The duo of `makeUpstreamFilterChainForDiscoveryChain` and `makeListenerForDiscoveryChain` were really hard to reason about, and led to concealing a bug in their branching logic. There were several issues here:
- They tried to accomplish too much: determining filter name, cluster name, and whether RDS should be used.
- They embedded logic to handle significantly different kinds of upstream listeners (passthrough, prepared query, typical services, and catch-all)
- They needed to coalesce different data sources (Upstream and CompiledDiscoveryChain)
Rather than handling all of those tasks inside of these functions, this PR pulls out the RDS/clusterName/filterName logic.
This refactor also fixed a bug with the handling of [UpstreamDefaults](https://www.consul.io/docs/connect/config-entries/service-defaults#defaults). These defaults get stored as UpstreamConfig in the proxy snapshot with a DestinationName of "*", since they apply to all upstreams. However, this wildcard destination name must not be used when creating the name of the associated upstream cluster. The coalescing logic in the original functions here was in some situations creating clusters with a `*.` prefix, which is not a valid destination.
Fixes an issue described in #10132, where if two DCs are WAN federated
over mesh gateways, and the gateway in the non-primary DC is terminated
and receives a new IP address (as is commonly the case when running them
on ephemeral compute instances) the primary DC is unable to re-establish
its connection until the agent running on its own gateway is restarted.
This was happening because we always preferred gateways discovered by
the `Internal.ServiceDump` RPC (which would fail because there's no way
to dial the remote DC) over those discovered in the federation state,
which is replicated as long as the primary DC's gateway is reachable.
This will behave the way we handle SNI and SPIFFE IDs, where the default
partition is excluded.
Excluding the default ensures that don't attempt to compare default.dc2
to dc2 in OSS.
This commit updates mesh gateway watches for cross-partitions
communication.
* Mesh gateways are keyed by partition and datacenter.
* Mesh gateways will now watch gateways in partitions that export
services to their partition.
* Mesh gateways in non-default partitions will not have cross-datacenter
watches. They are not involved in traditional WAN federation.
Previously the datacenter of the gateway was the key identifier, now it
is the datacenter and partition.
When dialing services in other partitions or datacenters we now watch
the appropriate partition.
These methods only called a single function. Wrappers like this end up making code harder to read
because it adds extra ways of doing things.
We already have many helper functions for constructing these types, we don't need additional methods.
Previously SAN validation for prepared queries was broken because we
validated against the name, namespace, and datacenter for prepared
queries.
However, prepared queries can target:
- Services with a name that isn't their own
- Services in multiple datacenters
This means that the SpiffeID to validate needs to be based on the
prepared query endpoints, and not the prepared query's upstream
definition.
This commit updates prepared query clusters to account for that.
- The TestNodeService helper created services with the fixed name "web",
and now that name is overridable.
- The discovery chain snapshot didn't have prepared query endpoints so
the endpoints tests were missing data for prepared queries
Knowing that blocking queries are firing does not provide much
information on its own. If we know the correlation IDs we can
piece together which parts of the snapshot have been populated.
Some of these responses might be empty from the blocking
query timing out. But if they're returning quickly I think we
can reasonably assume they contain data.
There is no interaction between these handlers, so splitting them into separate files
makes it easier to discover the full implementation of each kindHandler.
This commit extracts all the kind-specific logic into handler types, and
keeps the generic parts on the state struct. This change should make it
easier to add new kinds, and see the implementation of each kind more
clearly.
These two new struct types will allow us to make polymorphic handler for each kind, instad of
having all the logic for each proxy kind on the state struct.
context.Context should never be stored on a struct (as it says in the godoc) because it is easy to
to end up with the wrong context when it is stored.
Also see https://blog.golang.org/context-and-structs
This change is also in preparation for splitting state into kind-specific handlers so that the
implementation of each kind is grouped together.
Co-authored-by: R.B. Boyer <4903+rboyer@users.noreply.github.com>
Previously we would associate the address of a discovery chain target
with the discovery chain's filter chain. This was broken for a few reasons:
- If the upstream is a virtual service, the client proxy has no way of
dialing it because virtual services are not targets of their discovery
chains. The targets are distinct services. This is addressed by watching
the endpoints of all upstream services, not just their discovery chain
targets.
- If multiple discovery chains resolve to the same target, that would
lead to multiple filter chains attempting to match on the target's
virtual IP. This is addressed by only matching on the upstream's virtual
IP.
NOTE: this implementation requires an intention to the redirecting
virtual service and not just to the final destination. This is how
we can know that the virtual service is an upstream to watch.
A later PR will look into traversing discovery chains when computing
upstreams so that intentions are only required to the discovery chain
targets.
No config entry needs a Kind field. It is only used to determine the Go type to
target. As we introduce new config entries (like this one) we can remove the kind field
and have the GetKind method return the single supported value.
In this case (similar to proxy-defaults) the Name field is also unnecessary. We always
use the same value. So we can omit the name field entirely.
This adds support for the Incremental xDS protocol when using xDS v3. This is best reviewed commit-by-commit and will not be squashed when merged.
Union of all commit messages follows to give an overarching summary:
xds: exclusively support incremental xDS when using xDS v3
Attempts to use SoTW via v3 will fail, much like attempts to use incremental via v2 will fail.
Work around a strange older envoy behavior involving empty CDS responses over incremental xDS.
xds: various cleanups and refactors that don't strictly concern the addition of incremental xDS support
Dissolve the connectionInfo struct in favor of per-connection ResourceGenerators instead.
Do a better job of ensuring the xds code uses a well configured logger that accurately describes the connected client.
xds: pull out checkStreamACLs method in advance of a later commit
xds: rewrite SoTW xDS protocol tests to use protobufs rather than hand-rolled json strings
In the test we very lightly reuse some of the more boring protobuf construction helper code that is also technically under test. The important thing of the protocol tests is testing the protocol. The actual inputs and outputs are largely already handled by the xds golden output tests now so these protocol tests don't have to do double-duty.
This also updates the SoTW protocol test to exclusively use xDS v2 which is the only variant of SoTW that will be supported in Consul 1.10.
xds: default xds.Server.AuthCheckFrequency at use-time instead of construction-time
This config entry is being renamed primarily because in k8s the name
cluster could be confusing given that the config entry applies across
federated datacenters.
Additionally, this config entry will only apply to Consul as a service
mesh, so the more generic "cluster" name is not needed.
Setting this field to a value is equivalent to using the 'near' query paramter.
The intent is to sort the results by proximity to the node requesting
them. However with connect we send the results to envoy, which doesn't
care about the order, so setting this field is increasing the work
performed for no gain.
It is necessary to unset this field now because we would like connect
to use streaming, but streaming does not support sorting by proximity.
This PR replaces the original boolean used to configure transparent
proxy mode. It was replaced with a string mode that can be set to:
- "": Empty string is the default for when the setting should be
defaulted from other configuration like config entries.
- "direct": Direct mode is how applications originally opted into the
mesh. Proxy listeners need to be dialed directly.
- "transparent": Transparent mode enables configuring Envoy as a
transparent proxy. Traffic must be captured and redirected to the
inbound and outbound listeners.
This PR also adds a struct for transparent proxy specific configuration.
Initially this is not stored as a pointer. Will revisit that decision
before GA.
Deadlock scenario:
1. Due to scheduling, the state runner sends one snapshot into
snapCh and then attempts to send a second. The first send succeeds
because the channel is buffered, but the second blocks.
2. Separately, Manager.Watch is called by the xDS server after
getting a discovery request from Envoy. This function acquires the
manager lock and then blocks on receiving the CurrentSnapshot from
the state runner.
3. Separately, there is a Manager goroutine that reads the snapshots
from the channel in step 1. These reads are done to notify proxy
watchers, but they require holding the manager lock. This goroutine
goes to acquire that lock, but can't because it is held by step 2.
Now, the goroutine from step 3 is waiting on the one from step 2 to
release the lock. The goroutine from step 2 won't release the lock until
the goroutine in step 1 advances. But the goroutine in step 1 is waiting
for the one in step 3. Deadlock.
By making this send non-blocking step 1 above can proceed. The coalesce
timer will be reset and a new valid snapshot will be delivered after it
elapses or when one is requested by xDS.
Add a skip condition to all tests slower than 100ms.
This change was made using `gotestsum tool slowest` with data from the
last 3 CI runs of master.
See https://github.com/gotestyourself/gotestsum#finding-and-skipping-slow-tests
With this change:
```
$ time go test -count=1 -short ./agent
ok github.com/hashicorp/consul/agent 0.743s
real 0m4.791s
$ time go test -count=1 -short ./agent/consul
ok github.com/hashicorp/consul/agent/consul 4.229s
real 0m8.769s
```
This implements a solution for #7863
It does:
Add a new config cache.entry_fetch_rate to limit the number of calls/s for a given cache entry, default value = rate.Inf
Add cache.entry_fetch_max_burst size of rate limit (default value = 2)
The new configuration now supports the following syntax for instance to allow 1 query every 3s:
command line HCL: -hcl 'cache = { entry_fetch_rate = 0.333}'
in JSON
{
"cache": {
"entry_fetch_rate": 0.333
}
}
The rationale behind removing them is that all of our own code (xDS, builtin connect proxy) use the cache notification mechanism. This ensures that the blocking fetch behind the scenes is always executing. Therefore the only way you might go to get a certificate and have to wait is when 1) the request has never been made for that cert before or 2) you are using the v1/agent/connect/ca/leaf API for retrieving the cert yourself.
In the first case, the refresh change doesn’t alter the behavior. In the second case, it can be mitigated by using blocking queries with that API which just like normal cache notification mechanism will cause the blocking fetch to be initiated and to get leaf certs as soon as needed.
If you are not using blocking queries, or Envoy/xDS, or the builtin connect proxy but are retrieving the certs yourself then the HTTP endpoint might take a little longer to respond.
This also renames the RefreshTimeout field on the register options to QueryTimeout to more accurately reflect that it is used for any type that supports blocking queries.
And fix the 'value not used' issues.
Many of these are not bugs, but a few are tests not checking errors, and
one appears to be a missed error in non-test code.
Co-authored-by: Matt Keeler <mkeeler@users.noreply.github.com>
Currently when passing hostname clusters to Envoy, we set each service instance registered with Consul as an LbEndpoint for the cluster.
However, Envoy can only handle one per cluster:
[2020-06-04 18:32:34.094][1][warning][config] [source/common/config/grpc_subscription_impl.cc:87] gRPC config for type.googleapis.com/envoy.api.v2.Cluster rejected: Error adding/updating cluster(s) dc2.internal.ddd90499-9b47-91c5-4616-c0cbf0fc358a.consul: LOGICAL_DNS clusters must have a single locality_lb_endpoint and a single lb_endpoint, server.dc2.consul: LOGICAL_DNS clusters must have a single locality_lb_endpoint and a single lb_endpoint
Envoy is currently handling this gracefully by only picking one of the endpoints. However, we should avoid passing multiple to avoid these warning logs.
This PR:
* Ensures we only pass one endpoint, which is tied to one service instance.
* We prefer sending an endpoint which is marked as Healthy by Consul.
* If no endpoints are healthy we emit a warning and skip the cluster.
* If multiple unique hostnames are spread across service instances we emit a warning and let the user know which will be resolved.
The DNS resolution will be handled by Envoy and defaults to LOGICAL_DNS. This discovery type can be overridden on a per-gateway basis with the envoy_dns_discovery_type Gateway Option.
If a service contains an instance with a hostname as an address we set the Envoy cluster to use DNS as the discovery type rather than EDS. Since both mesh gateways and terminating gateways route to clusters using SNI, whenever there is a mix of hostnames and IP addresses associated with a service we use the hostname + CDS rather than the IPs + EDS.
Note that we detect hostnames by attempting to parse the service instance's address as an IP. If it is not a valid IP we assume it is a hostname.
Three of the checks are temporarily disabled to limit the size of the
diff, and allow us to enable all the other checks in CI.
In a follow up we can fix the issues reported by the other checks one
at a time, and enable them.
* Standardize support for Tagged and BindAddresses in Ingress Gateways
This updates the TaggedAddresses and BindAddresses behavior for Ingress
to match Mesh/Terminating gateways. The `consul connect envoy` command
now also allows passing an address without a port for tagged/bind
addresses.
* Update command/connect/envoy/envoy.go
Co-authored-by: Freddy <freddygv@users.noreply.github.com>
* PR comments
* Check to see if address is an actual IP address
* Update agent/xds/listeners.go
Co-authored-by: Freddy <freddygv@users.noreply.github.com>
* fix whitespace
Co-authored-by: Chris Piraino <cpiraino@hashicorp.com>
Co-authored-by: Freddy <freddygv@users.noreply.github.com>
- Use correct enterprise metadata for finding config entry
- nil out cancel functions on config snapshot copy
- Look at HostsSet when checking validity
- Validate that this cannot be set on a 'tcp' listener nor on a wildcard
service.
- Add Hosts field to api and test in consul config write CLI
- xds: Configure envoy with user-provided hosts from ingress gateways
This commit adds the necessary changes to allow an ingress gateway to
route traffic from a single defined port to multiple different upstream
services in the Consul mesh.
To do this, we now require all HTTP requests coming into the ingress
gateway to specify a Host header that matches "<service-name>.*" in
order to correctly route traffic to the correct service.
- Differentiate multiple listener's route names by port
- Adds a case in xds for allowing default discovery chains to create a
route configuration when on an ingress gateway. This allows default
services to easily use host header routing
- ingress-gateways have a single route config for each listener
that utilizes domain matching to route to different services.
This commit copies many of the connect-proxy xds testcases and reuses
for ingress gateways. This allows us to more easily see changes to the
envoy configuration when make updates to ingress gateways.
Previously the SupportsBlocking option was specified by a method on the
type, and all the other options were specified from RegisterOptions.
This change moves RegisterOptions to a method on the type, and moves
SupportsBlocking into the options struct.
Currently there are only 2 cache-types. So all cache-types can implement
this method by embedding a struct with those predefined values. In the
future if a cache type needs to be registered more than once with different
options it can remove the embedded type and implement the method in a way
that allows for paramaterization.
* Implements a simple, tcp ingress gateway workflow
This adds a new type of gateway for allowing Ingress traffic into Connect from external services.
Co-authored-by: Chris Piraino <cpiraino@hashicorp.com>
If a proxied service is a gRPC or HTTP2 service, but a path is exposed
using the HTTP1 or TCP protocol, Envoy should not be configured with
`http2ProtocolOptions` for the cluster backing the path.
A situation where this comes up is a gRPC service whose healthcheck or
metrics route (e.g. for Prometheus) is an HTTP1 service running on
a different port. Previously, if these were exposed either using
`Expose: { Checks: true }` or `Expose: { Paths: ... }`, Envoy would
still be configured to communicate with the path over HTTP2, which would
not work properly.
This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch:
There are several distinct chunks of code that are affected:
* new flags and config options for the server
* retry join WAN is slightly different
* retry join code is shared to discover primary mesh gateways from secondary datacenters
* because retry join logic runs in the *agent* and the results of that
operation for primary mesh gateways are needed in the *server* there are
some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur
at multiple layers of abstraction just to pass the data down to the right
layer.
* new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers
* the function signature for RPC dialing picked up a new required field (the
node name of the destination)
* several new RPCs for manipulating a FederationState object:
`FederationState:{Apply,Get,List,ListMeshGateways}`
* 3 read-only internal APIs for debugging use to invoke those RPCs from curl
* raft and fsm changes to persist these FederationStates
* replication for FederationStates as they are canonically stored in the
Primary and replicated to the Secondaries.
* a special derivative of anti-entropy that runs in secondaries to snapshot
their local mesh gateway `CheckServiceNodes` and sync them into their upstream
FederationState in the primary (this works in conjunction with the
replication to distribute addresses for all mesh gateways in all DCs to all
other DCs)
* a "gateway locator" convenience object to make use of this data to choose
the addresses of gateways to use for any given RPC or gossip operation to a
remote DC. This gets data from the "retry join" logic in the agent and also
directly calls into the FSM.
* RPC (`:8300`) on the server sniffs the first byte of a new connection to
determine if it's actually doing native TLS. If so it checks the ALPN header
for protocol determination (just like how the existing system uses the
type-byte marker).
* 2 new kinds of protocols are exclusively decoded via this native TLS
mechanism: one for ferrying "packet" operations (udp-like) from the gossip
layer and one for "stream" operations (tcp-like). The packet operations
re-use sockets (using length-prefixing) to cut down on TLS re-negotiation
overhead.
* the server instances specially wrap the `memberlist.NetTransport` when running
with gateway federation enabled (in a `wanfed.Transport`). The general gist is
that if it tries to dial a node in the SAME datacenter (deduced by looking
at the suffix of the node name) there is no change. If dialing a DIFFERENT
datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh
gateways to eventually end up in a server's :8300 port.
* a new flag when launching a mesh gateway via `consul connect envoy` to
indicate that the servers are to be exposed. This sets a special service
meta when registering the gateway into the catalog.
* `proxycfg/xds` notice this metadata blob to activate additional watches for
the FederationState objects as well as the location of all of the consul
servers in that datacenter.
* `xds:` if the extra metadata is in place additional clusters are defined in a
DC to bulk sink all traffic to another DC's gateways. For the current
datacenter we listen on a wildcard name (`server.<dc>.consul`) that load
balances all servers as well as one mini-cluster per node
(`<node>.server.<dc>.consul`)
* the `consul tls cert create` command got a new flag (`-node`) to help create
an additional SAN in certs that can be used with this flavor of federation.
* xDS Mesh Gateway Resolver Subset Fixes
The first fix was that clusters were being generated for every service resolver subset regardless of there being any service instances of the associated service in that dc. The previous logic didn’t care at all but now it will omit generating those clusters unless we also have service instances that should be proxied.
The second fix was to respect the DefaultSubset of a service resolver so that mesh-gateways would configure the endpoints of the unnamed subset cluster to only those endpoints matched by the default subsets filters.
* Refactor the gateway endpoint generation to be a little easier to read
R.B. Boyer