github
/
consul
mirror of https://github.com/hashicorp/consul
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

add config toggle to disable host metrics and docs

pull/17038/head
Nick Ethier 2023-05-12 16:09:54 -04:00
parent fafb68b28d
commit baffef45bd
12 changed files with 53 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
agent/config/builder.go
View File

@ -941,6 +941,7 @@ func (b *builder) build() (rt RuntimeConfig, err error) {
CirconusCheckTags: stringVal(c.Telemetry.CirconusCheckTags),
CirconusSubmissionInterval: stringVal(c.Telemetry.CirconusSubmissionInterval),
CirconusSubmissionURL: stringVal(c.Telemetry.CirconusSubmissionURL),
DisableHostMetrics: boolVal(c.Telemetry.DisableHostMetrics),
DisableHostname: boolVal(c.Telemetry.DisableHostname),
DogstatsdAddr: stringVal(c.Telemetry.DogstatsdAddr),
DogstatsdTags: c.Telemetry.DogstatsdTags,

1
agent/config/config.go
View File

@ -690,6 +690,7 @@ type Telemetry struct {
CirconusCheckTags *string `mapstructure:"circonus_check_tags" json:"circonus_check_tags,omitempty"`
CirconusSubmissionInterval *string `mapstructure:"circonus_submission_interval" json:"circonus_submission_interval,omitempty"`
CirconusSubmissionURL *string `mapstructure:"circonus_submission_url" json:"circonus_submission_url,omitempty"`
DisableHostMetrics *bool `mapstructure:"disable_host_metrics" json:"disable_host_metrics,omitempty"`
DisableHostname *bool `mapstructure:"disable_hostname" json:"disable_hostname,omitempty"`
DogstatsdAddr *string `mapstructure:"dogstatsd_addr" json:"dogstatsd_addr,omitempty"`
DogstatsdTags []string `mapstructure:"dogstatsd_tags" json:"dogstatsd_tags,omitempty"`

1
agent/config/runtime_test.go
View File

@ -6754,6 +6754,7 @@ func TestLoad_FullConfig(t *testing.T) {
Expiration: 15 * time.Second,
Name: "ftO6DySn", // notice this is the same as the metrics prefix
},
DisableHostMetrics: true,
},
TLS: tlsutil.Config{
InternalRPC: tlsutil.ProtocolConfig{

3
agent/config/testdata/TestRuntimeConfig_Sanitize.golden vendored
View File

@ -462,6 +462,7 @@
"CirconusSubmissionInterval": "",
"CirconusSubmissionURL": "",
"Disable": false,
"DisableHostMetrics": false,
"DisableHostname": false,
"DogstatsdAddr": "",
"DogstatsdTags": [],
@ -510,4 +511,4 @@
"VersionPrerelease": "",
"Watches": [],
"XDSUpdateRateLimit": 0
}
}

1
agent/config/testdata/full-config.hcl vendored
View File

@ -690,6 +690,7 @@ telemetry {
circonus_check_tags = "prvO4uBl"
circonus_submission_interval = "DolzaflP"
circonus_submission_url = "gTcbS93G"
disable_host_metrics = true
disable_hostname = true
dogstatsd_addr = "0wSndumK"
dogstatsd_tags = [ "3N81zSUB","Xtj8AnXZ" ]

1
agent/config/testdata/full-config.json vendored
View File

@ -808,6 +808,7 @@
"circonus_check_tags": "prvO4uBl",
"circonus_submission_interval": "DolzaflP",
"circonus_submission_url": "gTcbS93G",
"disable_host_metrics": true,
"disable_hostname": true,
"dogstatsd_addr": "0wSndumK",
"dogstatsd_tags": [

4
agent/setup.go
View File

@ -119,7 +119,9 @@ func NewBaseDeps(configLoader ConfigLoader, logOut io.Writer, providedLogger hcl
if err != nil {
return d, fmt.Errorf("failed to initialize telemetry: %w", err)
}
hoststats.NewCollector(context.Background(), d.Logger, cfg.DataDir)
if !cfg.Telemetry.DisableHostMetrics {
hoststats.NewCollector(context.Background(), d.Logger, cfg.DataDir)
}
d.TLSConfigurator, err = tlsutil.NewConfigurator(cfg.TLS, d.Logger)
if err != nil {

2
lib/hoststats/cpu.go
View File

@ -21,9 +21,9 @@ func (h *cpuStatsCalculator) calculate(times cpu.TimesStat) (idle float64, user
currentIdle := times.Idle
currentUser := times.User
currentSystem := times.System
currentTotal := times.Total()
currentBusy := times.User + times.System + times.Nice + times.Iowait + times.Irq +
times.Softirq + times.Steal + times.Guest + times.GuestNice
currentTotal := currentBusy + currentIdle
deltaTotal := currentTotal - h.prevTotal
idle = ((currentIdle - h.prevIdle) / deltaTotal) * 100

23
lib/hoststats/cpu_test.go
View File

@ -27,12 +27,9 @@ func TestCpuStats_percent(t *testing.T) {
}
func TestHostStats_CPU(t *testing.T) {
assert := assert.New(t)
logger := testutil.Logger(t)
cwd, err := os.Getwd()
assert.Nil(err)
assert.Nil(t, err)
hs := initCollector(logger, cwd)
// Collect twice so we can calculate percents we need to generate some work
@ -45,18 +42,18 @@ func TestHostStats_CPU(t *testing.T) {
}
hs.collect()
stats := hs.Stats()
assert.NotZero(len(stats.CPU))
assert.NotZero(t, len(stats.CPU))
for _, cpu := range stats.CPU {
assert.False(math.IsNaN(cpu.Idle))
assert.False(math.IsNaN(cpu.Total))
assert.False(math.IsNaN(cpu.System))
assert.False(math.IsNaN(cpu.User))
assert.False(t, math.IsNaN(cpu.Idle))
assert.False(t, math.IsNaN(cpu.Total))
assert.False(t, math.IsNaN(cpu.System))
assert.False(t, math.IsNaN(cpu.User))
assert.False(math.IsInf(cpu.Idle, 0))
assert.False(math.IsInf(cpu.Total, 0))
assert.False(math.IsInf(cpu.System, 0))
assert.False(math.IsInf(cpu.User, 0))
assert.False(t, math.IsInf(cpu.Idle, 0))
assert.False(t, math.IsInf(cpu.Total, 0))
assert.False(t, math.IsInf(cpu.System, 0))
assert.False(t, math.IsInf(cpu.User, 0))
}
}

5
lib/telemetry.go
View File

@ -210,6 +210,11 @@ type TelemetryConfig struct {
//
// hcl: telemetry { prometheus_retention_time = "duration" }
PrometheusOpts prometheus.PrometheusOpts
// DisableHostMetrics will disable metrics collected about the host system such as cpu memory and disk usage.
//
// hcl: telemetry { disable_host_metrics = (true|false) }
DisableHostMetrics bool
}
// MetricsHandler provides an http.Handler for displaying metrics.

3
website/content/docs/agent/config/config-files.mdx
View File

@ -1817,6 +1817,9 @@ subsystem that provides Consul's service mesh capabilities.
be used based on **where** this particular instance is running (e.g. a specific
geo location or datacenter, dc:sfo). By default, this is left blank and not used.
- `disable_host_metrics` ((#telemetry-disable_host_metrics))
This disables reporting of host metrics about system resources, defaults to false.
- `disable_hostname` ((#telemetry-disable_hostname))
This controls whether or not to prepend runtime telemetry with the machine's
hostname, defaults to false.

24
website/content/docs/agent/telemetry.mdx
View File

@ -755,3 +755,27 @@ Consul attaches the following labels to metric values.
| `peer_id` | The ID of a peer connected to the reporting cluster or leader. | Any UUID |
| `partition` | <EnterpriseAlert inline /> Name of the partition that the peering is created in. | Any defined partition name in the cluster |
## Server Host Metrics
Consul servers report the following metrics about the host's system resources
**Requirements:**
- Consul 1.15.3+
| Metric | Description | Unit | Type |
| ----------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------- | ------- |
| `consul.host.memory.total` | The total physical memory in bytes | mixed | mixed |
| `consul.host.memory.available` | The available physical memory in bytes | mixed | mixed |
| `consul.host.memory.free` | The free physical memory in bytes | mixed | mixed |
| `consul.host.memory.used` | The used physical memory in bytes | mixed | mixed |
| `consul.host.memory.used_percent` | The used physical memory as a percentage of total physical memory | mixed | mixed |
| `consul.host.cpu.total` | The host's total cpu utilization
| `consul.host.cpu.user` | The cpu utilization in user space
| `consul.host.cpu.idle` | The cpu utilization in idle state
| `consul.host.cpu.system` | The cpu utilization in system space
| `consul.host.disk.size` | The size in bytes of the data_dir disk
| `consul.host.disk.used` | The number of bytes used on the data_dir disk
| `consul.host.disk.available` | The number of bytes available on the data_dir disk
| `consul.host.disk.used_percent` | The percentage of disk space used on the data_dir disk
| `consul.host.disk.inodes_percent` | The percentage of inode usage on the data_dir disk
| `consul.host.uptime` | The uptime of the host in seconds