You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
prometheus/discovery/consul/consul.go

530 lines
16 KiB

// Copyright 2015 The Prometheus Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package consul
import (
"context"
"fmt"
"net"
"net/http"
"strconv"
"strings"
"time"
"github.com/go-kit/kit/log"
"github.com/go-kit/kit/log/level"
consul "github.com/hashicorp/consul/api"
"github.com/mwitkow/go-conntrack"
"github.com/prometheus/client_golang/prometheus"
config_util "github.com/prometheus/common/config"
"github.com/prometheus/common/model"
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
"github.com/prometheus/prometheus/discovery/targetgroup"
"github.com/prometheus/prometheus/util/strutil"
)
const (
watchTimeout = 30 * time.Second
retryInterval = 15 * time.Second
// addressLabel is the name for the label containing a target's address.
addressLabel = model.MetaLabelPrefix + "consul_address"
// nodeLabel is the name for the label containing a target's node name.
nodeLabel = model.MetaLabelPrefix + "consul_node"
// metaDataLabel is the prefix for the labels mapping to a target's metadata.
metaDataLabel = model.MetaLabelPrefix + "consul_metadata_"
// serviceMetaDataLabel is the prefix for the labels mapping to a target's service metadata.
serviceMetaDataLabel = model.MetaLabelPrefix + "consul_service_metadata_"
// tagsLabel is the name of the label containing the tags assigned to the target.
tagsLabel = model.MetaLabelPrefix + "consul_tags"
// serviceLabel is the name of the label containing the service name.
serviceLabel = model.MetaLabelPrefix + "consul_service"
// serviceAddressLabel is the name of the label containing the (optional) service address.
serviceAddressLabel = model.MetaLabelPrefix + "consul_service_address"
//servicePortLabel is the name of the label containing the service port.
servicePortLabel = model.MetaLabelPrefix + "consul_service_port"
// datacenterLabel is the name of the label containing the datacenter ID.
datacenterLabel = model.MetaLabelPrefix + "consul_dc"
// serviceIDLabel is the name of the label containing the service ID.
serviceIDLabel = model.MetaLabelPrefix + "consul_service_id"
// Constants for instrumentation.
namespace = "prometheus"
)
var (
rpcFailuresCount = prometheus.NewCounter(
prometheus.CounterOpts{
Namespace: namespace,
Name: "sd_consul_rpc_failures_total",
Help: "The number of Consul RPC call failures.",
})
rpcDuration = prometheus.NewSummaryVec(
prometheus.SummaryOpts{
Namespace: namespace,
Name: "sd_consul_rpc_duration_seconds",
Help: "The duration of a Consul RPC call in seconds.",
},
[]string{"endpoint", "call"},
)
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
// DefaultSDConfig is the default Consul SD configuration.
DefaultSDConfig = SDConfig{
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
TagSeparator: ",",
Scheme: "http",
Server: "localhost:8500",
AllowStale: true,
RefreshInterval: model.Duration(watchTimeout),
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
}
)
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
// SDConfig is the configuration for Consul service discovery.
type SDConfig struct {
Server string `yaml:"server,omitempty"`
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
Token config_util.Secret `yaml:"token,omitempty"`
Datacenter string `yaml:"datacenter,omitempty"`
TagSeparator string `yaml:"tag_separator,omitempty"`
Scheme string `yaml:"scheme,omitempty"`
Username string `yaml:"username,omitempty"`
Password config_util.Secret `yaml:"password,omitempty"`
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// See https://www.consul.io/docs/internals/consensus.html#consistency-modes,
// stale reads are a lot cheaper and are a necessity if you have >5k targets.
AllowStale bool `yaml:"allow_stale"`
// By default use blocking queries (https://www.consul.io/api/index.html#blocking-queries)
// but allow users to throttle updates if necessary. This can be useful because of "bugs" like
// https://github.com/hashicorp/consul/issues/3712 which cause an un-necessary
// amount of requests on consul.
RefreshInterval model.Duration `yaml:"refresh_interval,omitempty"`
// See https://www.consul.io/api/catalog.html#list-services
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
// The list of services for which targets are discovered.
// Defaults to all services if empty.
Services []string `yaml:"services,omitempty"`
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// An optional tag used to filter instances inside a service. A single tag is supported
// here to match the Consul API.
ServiceTag string `yaml:"tag,omitempty"`
// Desired node metadata.
NodeMeta map[string]string `yaml:"node_meta,omitempty"`
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
TLSConfig config_util.TLSConfig `yaml:"tls_config,omitempty"`
}
// UnmarshalYAML implements the yaml.Unmarshaler interface.
func (c *SDConfig) UnmarshalYAML(unmarshal func(interface{}) error) error {
*c = DefaultSDConfig
type plain SDConfig
err := unmarshal((*plain)(c))
if err != nil {
return err
}
if strings.TrimSpace(c.Server) == "" {
return fmt.Errorf("Consul SD configuration requires a server address")
}
return nil
}
func init() {
prometheus.MustRegister(rpcFailuresCount)
prometheus.MustRegister(rpcDuration)
// Initialize metric vectors.
rpcDuration.WithLabelValues("catalog", "service")
rpcDuration.WithLabelValues("catalog", "services")
}
// Discovery retrieves target information from a Consul server
// and updates them via watches.
type Discovery struct {
client *consul.Client
clientDatacenter string
tagSeparator string
watchedServices []string // Set of services which will be discovered.
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
watchedTag string // A tag used to filter instances of a service.
watchedNodeMeta map[string]string
allowStale bool
refreshInterval time.Duration
finalizer func()
logger log.Logger
}
// NewDiscovery returns a new Discovery for the given config.
Refactor SD configuration to remove `config` dependency (#3629) * refactor: move targetGroup struct and CheckOverflow() to their own package * refactor: move auth and security related structs to a utility package, fix import error in utility package * refactor: Azure SD, remove SD struct from config * refactor: DNS SD, remove SD struct from config into dns package * refactor: ec2 SD, move SD struct from config into the ec2 package * refactor: file SD, move SD struct from config to file discovery package * refactor: gce, move SD struct from config to gce discovery package * refactor: move HTTPClientConfig and URL into util/config, fix import error in httputil * refactor: consul, move SD struct from config into consul discovery package * refactor: marathon, move SD struct from config into marathon discovery package * refactor: triton, move SD struct from config to triton discovery package, fix test * refactor: zookeeper, move SD structs from config to zookeeper discovery package * refactor: openstack, remove SD struct from config, move into openstack discovery package * refactor: kubernetes, move SD struct from config into kubernetes discovery package * refactor: notifier, use targetgroup package instead of config * refactor: tests for file, marathon, triton SD - use targetgroup package instead of config.TargetGroup * refactor: retrieval, use targetgroup package instead of config.TargetGroup * refactor: storage, use config util package * refactor: discovery manager, use targetgroup package instead of config.TargetGroup * refactor: use HTTPClient and TLS config from configUtil instead of config * refactor: tests, use targetgroup package instead of config.TargetGroup * refactor: fix tagetgroup.Group pointers that were removed by mistake * refactor: openstack, kubernetes: drop prefixes * refactor: remove import aliases forced due to vscode bug * refactor: move main SD struct out of config into discovery/config * refactor: rename configUtil to config_util * refactor: rename yamlUtil to yaml_config * refactor: kubernetes, remove prefixes * refactor: move the TargetGroup package to discovery/ * refactor: fix order of imports
7 years ago
func NewDiscovery(conf *SDConfig, logger log.Logger) (*Discovery, error) {
if logger == nil {
logger = log.NewNopLogger()
}
tls, err := config_util.NewTLSConfig(&conf.TLSConfig)
if err != nil {
return nil, err
}
transport := &http.Transport{
IdleConnTimeout: 5 * time.Duration(conf.RefreshInterval),
TLSClientConfig: tls,
DialContext: conntrack.NewDialContextFunc(
conntrack.DialWithTracing(),
conntrack.DialWithName("consul_sd"),
),
}
wrapper := &http.Client{
Transport: transport,
Timeout: 35 * time.Second,
}
clientConf := &consul.Config{
Address: conf.Server,
Scheme: conf.Scheme,
Datacenter: conf.Datacenter,
Token: string(conf.Token),
HttpAuth: &consul.HttpBasicAuth{
Username: conf.Username,
Password: string(conf.Password),
},
HttpClient: wrapper,
}
client, err := consul.NewClient(clientConf)
if err != nil {
return nil, err
}
cd := &Discovery{
client: client,
tagSeparator: conf.TagSeparator,
watchedServices: conf.Services,
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
watchedTag: conf.ServiceTag,
watchedNodeMeta: conf.NodeMeta,
allowStale: conf.AllowStale,
refreshInterval: time.Duration(conf.RefreshInterval),
clientDatacenter: conf.Datacenter,
finalizer: transport.CloseIdleConnections,
logger: logger,
}
return cd, nil
}
// shouldWatch returns whether the service of the given name should be watched.
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
func (d *Discovery) shouldWatch(name string, tags []string) bool {
return d.shouldWatchFromName(name) && d.shouldWatchFromTags(tags)
}
// shouldWatch returns whether the service of the given name should be watched based on its name.
func (d *Discovery) shouldWatchFromName(name string) bool {
// If there's no fixed set of watched services, we watch everything.
if len(d.watchedServices) == 0 {
return true
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
for _, sn := range d.watchedServices {
if sn == name {
return true
}
}
return false
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// shouldWatch returns whether the service of the given name should be watched based on its tags.
// This gets called when the user doesn't specify a list of services in order to avoid watching
// *all* services. Details in https://github.com/prometheus/prometheus/pull/3814
func (d *Discovery) shouldWatchFromTags(tags []string) bool {
// If there's no fixed set of watched tags, we watch everything.
if d.watchedTag == "" {
return true
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
for _, tag := range tags {
if d.watchedTag == tag {
return true
}
}
return false
}
// Get the local datacenter if not specified.
func (d *Discovery) getDatacenter() error {
// If the datacenter was not set from clientConf, let's get it from the local Consul agent
// (Consul default is to use local node's datacenter if one isn't given for a query).
if d.clientDatacenter != "" {
return nil
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
info, err := d.client.Agent().Self()
if err != nil {
level.Error(d.logger).Log("msg", "Error retrieving datacenter name", "err", err)
rpcFailuresCount.Inc()
return err
}
dc, ok := info["Config"]["Datacenter"].(string)
if !ok {
err := fmt.Errorf("Invalid value '%v' for Config.Datacenter", info["Config"]["Datacenter"])
level.Error(d.logger).Log("msg", "Error retrieving datacenter name", "err", err)
return err
}
d.clientDatacenter = dc
return nil
}
// Initialize the Discoverer run.
func (d *Discovery) initialize(ctx context.Context) {
// Loop until we manage to get the local datacenter.
for {
// We have to check the context at least once. The checks during channel sends
// do not guarantee that.
select {
case <-ctx.Done():
return
default:
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Get the local datacenter first, if necessary.
err := d.getDatacenter()
if err != nil {
time.Sleep(retryInterval)
continue
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// We are good to go.
return
}
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Run implements the Discoverer interface.
func (d *Discovery) Run(ctx context.Context, ch chan<- []*targetgroup.Group) {
if d.finalizer != nil {
defer d.finalizer()
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
d.initialize(ctx)
if len(d.watchedServices) == 0 || d.watchedTag != "" {
// We need to watch the catalog.
ticker := time.NewTicker(d.refreshInterval)
// Watched services and their cancellation functions.
services := make(map[string]func())
var lastIndex uint64
for {
select {
case <-ctx.Done():
ticker.Stop()
return
default:
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
d.watchServices(ctx, ch, &lastIndex, services)
<-ticker.C
}
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
} else {
// We only have fully defined services.
for _, name := range d.watchedServices {
d.watchService(ctx, ch, name)
}
<-ctx.Done()
}
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Watch the catalog for new services we would like to watch. This is called only
// when we don't know yet the names of the services and need to ask Consul the
// entire list of services.
func (d *Discovery) watchServices(ctx context.Context, ch chan<- []*targetgroup.Group, lastIndex *uint64, services map[string]func()) error {
catalog := d.client.Catalog()
level.Debug(d.logger).Log("msg", "Watching services", "tag", d.watchedTag)
t0 := time.Now()
srvs, meta, err := catalog.Services(&consul.QueryOptions{
WaitIndex: *lastIndex,
WaitTime: watchTimeout,
AllowStale: d.allowStale,
NodeMeta: d.watchedNodeMeta,
})
elapsed := time.Since(t0)
rpcDuration.WithLabelValues("catalog", "services").Observe(elapsed.Seconds())
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
if err != nil {
level.Error(d.logger).Log("msg", "Error refreshing service list", "err", err)
rpcFailuresCount.Inc()
time.Sleep(retryInterval)
return err
}
// If the index equals the previous one, the watch timed out with no update.
if meta.LastIndex == *lastIndex {
return nil
}
*lastIndex = meta.LastIndex
// Check for new services.
for name := range srvs {
// catalog.Service() returns a map of service name to tags, we can use that to watch
// only the services that have the tag we are looking for (if specified).
// In the future consul will also support server side for service metadata.
// https://github.com/hashicorp/consul/issues/1107
if !d.shouldWatch(name, srvs[name]) {
continue
}
if _, ok := services[name]; ok {
continue // We are already watching the service.
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
wctx, cancel := context.WithCancel(ctx)
d.watchService(wctx, ch, name)
services[name] = cancel
}
// Check for removed services.
for name, cancel := range services {
if _, ok := srvs[name]; !ok {
// Call the watch cancellation function.
cancel()
delete(services, name)
// Send clearing target group.
select {
case <-ctx.Done():
return ctx.Err()
case ch <- []*targetgroup.Group{{Source: name}}:
}
}
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
return nil
}
// consulService contains data belonging to the same service.
type consulService struct {
name string
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
tag string
labels model.LabelSet
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
discovery *Discovery
client *consul.Client
tagSeparator string
logger log.Logger
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Start watching a service.
func (d *Discovery) watchService(ctx context.Context, ch chan<- []*targetgroup.Group, name string) {
srv := &consulService{
discovery: d,
client: d.client,
name: name,
tag: d.watchedTag,
labels: model.LabelSet{
serviceLabel: model.LabelValue(name),
datacenterLabel: model.LabelValue(d.clientDatacenter),
},
tagSeparator: d.tagSeparator,
logger: d.logger,
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
go func() {
ticker := time.NewTicker(d.refreshInterval)
var lastIndex uint64
catalog := srv.client.Catalog()
for {
select {
case <-ctx.Done():
ticker.Stop()
return
default:
srv.watch(ctx, ch, catalog, &lastIndex)
<-ticker.C
}
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
}()
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Get updates for a service.
func (srv *consulService) watch(ctx context.Context, ch chan<- []*targetgroup.Group, catalog *consul.Catalog, lastIndex *uint64) error {
level.Debug(srv.logger).Log("msg", "Watching service", "service", srv.name, "tag", srv.tag)
t0 := time.Now()
nodes, meta, err := catalog.Service(srv.name, srv.tag, &consul.QueryOptions{
WaitIndex: *lastIndex,
WaitTime: watchTimeout,
AllowStale: srv.discovery.allowStale,
NodeMeta: srv.discovery.watchedNodeMeta,
})
elapsed := time.Since(t0)
rpcDuration.WithLabelValues("catalog", "service").Observe(elapsed.Seconds())
// Check the context before in order to exit early.
select {
case <-ctx.Done():
return ctx.Err()
default:
// Continue.
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
if err != nil {
level.Error(srv.logger).Log("msg", "Error refreshing service", "service", srv.name, "tag", srv.tag, "err", err)
rpcFailuresCount.Inc()
time.Sleep(retryInterval)
return err
}
// If the index equals the previous one, the watch timed out with no update.
if meta.LastIndex == *lastIndex {
return nil
}
*lastIndex = meta.LastIndex
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
tgroup := targetgroup.Group{
Source: srv.name,
Labels: srv.labels,
Targets: make([]model.LabelSet, 0, len(nodes)),
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
for _, node := range nodes {
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// We surround the separated list with the separator as well. This way regular expressions
// in relabeling rules don't have to consider tag positions.
var tags = srv.tagSeparator + strings.Join(node.ServiceTags, srv.tagSeparator) + srv.tagSeparator
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// If the service address is not empty it should be used instead of the node address
// since the service may be registered remotely through a different node
var addr string
if node.ServiceAddress != "" {
addr = net.JoinHostPort(node.ServiceAddress, fmt.Sprintf("%d", node.ServicePort))
} else {
addr = net.JoinHostPort(node.Address, fmt.Sprintf("%d", node.ServicePort))
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
labels := model.LabelSet{
model.AddressLabel: model.LabelValue(addr),
addressLabel: model.LabelValue(node.Address),
nodeLabel: model.LabelValue(node.Node),
tagsLabel: model.LabelValue(tags),
serviceAddressLabel: model.LabelValue(node.ServiceAddress),
servicePortLabel: model.LabelValue(strconv.Itoa(node.ServicePort)),
serviceIDLabel: model.LabelValue(node.ServiceID),
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Add all key/value pairs from the node's metadata as their own labels
for k, v := range node.NodeMeta {
name := strutil.SanitizeLabelName(k)
labels[metaDataLabel+model.LabelName(name)] = model.LabelValue(v)
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
// Add all key/value pairs from the service's metadata as their own labels
for k, v := range node.ServiceMeta {
name := strutil.SanitizeLabelName(k)
labels[serviceMetaDataLabel+model.LabelName(name)] = model.LabelValue(v)
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
tgroup.Targets = append(tgroup.Targets, labels)
}
consul: improve consul service discovery (#3814) * consul: improve consul service discovery Related to #3711 - Add the ability to filter by tag and node-meta in an efficient way (`/catalog/services` allow filtering by node-meta, and returns a `map[string]string` or `service`->`tags`). Tags and nore-meta are also used in `/catalog/service` requests. - Do not require a call to the catalog if services are specified by name. This is important because on large cluster `/catalog/services` changes all the time. - Add `allow_stale` configuration option to do stale reads. Non-stale reads can be costly, even more when you are doing them to a remote datacenter with 10k+ targets over WAN (which is common for federation). - Add `refresh_interval` to minimize the strain on the catalog and on the service endpoint. This is needed because of that kind of behavior from consul: https://github.com/hashicorp/consul/issues/3712 and because a catalog on a large cluster would basically change *all* the time. No need to discover targets in 1sec if we scrape them every minute. - Added plenty of unit tests. Benchmarks ---------- ```yaml scrape_configs: - job_name: prometheus scrape_interval: 60s static_configs: - targets: ["127.0.0.1:9090"] - job_name: "observability-by-tag" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 tag: marathon-user-observability # Used in After refresh_interval: 30s # Used in After+delay relabel_configs: - source_labels: [__meta_consul_tags] regex: ^(.*,)?marathon-user-observability(,.*)?$ action: keep - job_name: "observability-by-name" scrape_interval: "60s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - observability-cerebro - observability-portal-web - job_name: "fake-fake-fake" scrape_interval: "15s" metrics_path: "/metrics" consul_sd_configs: - server: consul.service.par.consul.prod.crto.in:8500 services: - fake-fake-fake ``` Note: tested with ~1200 services, ~5000 nodes. | Resource | Empty | Before | After | After + delay | | -------- |:-----:|:------:|:-----:|:-------------:| |/service-discovery size|5K|85MiB|27k|27k|27k| |`go_memstats_heap_objects`|100k|1M|120k|110k| |`go_memstats_heap_alloc_bytes`|24MB|150MB|28MB|27MB| |`rate(go_memstats_alloc_bytes_total[5m])`|0.2MB/s|28MB/s|2MB/s|0.3MB/s| |`rate(process_cpu_seconds_total[5m])`|0.1%|15%|2%|0.01%| |`process_open_fds`|16|*1236*|22|22| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="services"}[5m])`|~0|1|1|*0.03*| |`rate(prometheus_sd_consul_rpc_duration_seconds_count{call="service"}[5m])`|0.1|*80*|0.5|0.5| |`prometheus_target_sync_length_seconds{quantile="0.9",scrape_job="observability-by-tag"}`|N/A|200ms|0.2ms|0.2ms| |Network bandwidth|~10kbps|~2.8Mbps|~1.6Mbps|~10kbps| Filtering by tag using relabel_configs uses **100kiB and 23kiB/s per service per job** and quite a lot of CPU. Also sends and additional *1Mbps* of traffic to consul. Being a little bit smarter about this reduces the overhead quite a lot. Limiting the number of `/catalog/services` queries per second almost removes the overhead of service discovery. * consul: tweak `refresh_interval` behavior `refresh_interval` now does what is advertised in the documentation, there won't be more that one update per `refresh_interval`. It now defaults to 30s (which was also the current waitTime in the consul query). This also make sure we don't wait another 30s if we already waited 29s in the blocking call by substracting the number of elapsed seconds. Hopefully this will do what people expect it does and will be safer for existing consul infrastructures.
7 years ago
select {
case <-ctx.Done():
return ctx.Err()
case ch <- []*targetgroup.Group{&tgroup}:
}
return nil
}