Receiving an "acl not found" error from an RPC in the agent cache and the
streaming/event components will cause any request loops to cease under the
assumption that they will never work again if the token was destroyed. This
prevents log spam (#14144, #9738).
Unfortunately due to things like:
- authz requests going to stale servers that may not have witnessed the token
creation yet
- authz requests in a secondary datacenter happening before the tokens get
replicated to that datacenter
- authz requests from a primary TO a secondary datacenter happening before the
tokens get replicated to that datacenter
The caller will get an "acl not found" *before* the token exists, rather than
just after. The machinery added above in the linked PRs will kick in and
prevent the request loop from looping around again once the tokens actually
exist.
For `consul-dataplane` usages, where xDS is served by the Consul servers
rather than the clients ultimately this is not a problem because in that
scenario the `agent/proxycfg` machinery is on-demand and launched by a new xDS
stream needing data for a specific service in the catalog. If the watching
goroutines are terminated it ripples down and terminates the xDS stream, which
CDP will eventually re-establish and restart everything.
For Consul client usages, the `agent/proxycfg` machinery is ahead-of-time
launched at service registration time (called "local" in some of the proxycfg
machinery) so when the xDS stream comes in the data is already ready to go. If
the watching goroutines terminate it should terminate the xDS stream, but
there's no mechanism to re-spawn the watching goroutines. If the xDS stream
reconnects it will see no `ConfigSnapshot` and will not get one again until
the client agent is restarted, or the service is re-registered with something
changed in it.
This PR fixes a few things in the machinery:
- there was an inadvertent deadlock in fetching snapshot from the proxycfg
machinery by xDS, such that when the watching goroutine terminated the
snapshots would never be fetched. This caused some of the xDS machinery to
get indefinitely paused and not finish the teardown properly.
- Every 30s we now attempt to re-insert all locally registered services into
the proxycfg machinery.
- When services are re-inserted into the proxycfg machinery we special case
"dead" ones such that we unilaterally replace them rather that doing that
conditionally.
Prior to this PR, servers / agents would panic and crash if an ingress
or api gateway were configured to use a discovery chain that both:
1. Referenced a peered service
2. Had a mesh gateway mode of local
This could occur, because code for handling upstream watches was shared
between both connect-proxy and the gateways. As a short-term fix, this
PR ensures that the maps are always initialized for these gateway services.
This PR also wraps the proxycfg execution and service
registration calls with recover statements to ensure that future issues
like this do not put the server into an unrecoverable state.
* Stub proxycfg handler for API gateway
* Add Service Kind constants/handling for API Gateway
* Begin stubbing for SDS
* Add new Secret type to xDS order of operations
* Continue stubbing of SDS
* Iterate on proxycfg handler for API gateway
* Handle BoundAPIGateway config entry subscription in proxycfg-glue
* Add API gateway to config snapshot validation
* Add API gateway to config snapshot clone, leaf, etc.
* Subscribe to bound route + cert config entries on bound-api-gateway
* Track routes + certs on API gateway config snapshot
* Generate DeepCopy() for types used in watch.Map
* Watch all active references on api-gateway, unwatch inactive
* Track loading of initial bound-api-gateway config entry
* Use proper proto package for SDS mapping
* Use ResourceReference instead of ServiceName, collect resources
* Fix typo, add + remove TODOs
* Watch discovery chains for TCPRoute
* Add TODO for updating gateway services for api-gateway
* make proto
* Regenerate deep-copy for proxycfg
* Set datacenter on upstream ID from query source
* Watch discovery chains for http-route service backends
* Add ServiceName getter to HTTP+TCP Service structs
* Clean up unwatched discovery chains on API Gateway
* Implement watch for ingress leaf certificate
* Collect upstreams on http-route + tcp-route updates
* Remove unused GatewayServices update handler
* Remove unnecessary gateway services logic for API Gateway
* Remove outdate TODO
* Use .ToIngress where appropriate, including TODO for cleaning up
* Cancel before returning error
* Remove GatewayServices subscription
* Add godoc for handlerAPIGateway functions
* Update terminology from Connect => Consul Service Mesh
Consistent with terminology changes in https://github.com/hashicorp/consul/pull/12690
* Add missing TODO
* Remove duplicate switch case
* Rerun deep-copy generator
* Use correct property on config snapshot
* Remove unnecessary leaf cert watch
* Clean up based on code review feedback
* Note handler properties that are initialized but set elsewhere
* Add TODO for moving helper func into structs pkg
* Update generated DeepCopy code
* gofmt
* Generate DeepCopy() for API gateway listener types
* Improve variable name
* Regenerate DeepCopy() code
* Fix linting issue
* Temporarily remove the secret type from resource generation
Adds a user-configurable rate limiter to proxycfg snapshot delivery,
with a default limit of 250 updates per second.
This addresses a problem observed in our load testing of Consul
Dataplane where updating a "global" resource such as a wildcard
intention or the proxy-defaults config entry could starve the Raft or
Memberlist goroutines of CPU time, causing general cluster instability.
Replaces the reflection-based implementation of proxycfg's
ConfigSnapshot.Clone with code generated by deep-copy.
While load testing server-based xDS (for consul-dataplane) we discovered
this method is extremely expensive. The ConfigSnapshot struct, directly
or indirectly, contains a copy of many of the structs in the agent/structs
package, which creates a large graph for copystructure.Copy to traverse
at runtime, on every proxy reconfiguration.
This is the OSS portion of enterprise PR 2339.
It improves our handling of "irrecoverable" errors in proxycfg data sources.
The canonical example of this is what happens when the ACL token presented by
Envoy is deleted/revoked. Previously, the stream would get "stuck" until the
xDS server re-checked the token (after 5 minutes) and terminated the stream.
Materializers would also sit burning resources retrying something that could
never succeed.
Now, it is possible for data sources to mark errors as "terminal" which causes
the xDS stream to be closed immediately. Similarly, the submatview.Store will
evict materializers when it observes they have encountered such an error.
Mesh gateways will now enable tcp connections with SNI names including peering information so that those connections may be proxied.
Note: this does not change the callers to use these mesh gateways.
Envoy's SPIFFE certificate validation extension allows for us to
validate against different root certificates depending on the trust
domain of the dialing proxy.
If there are any trust bundles from peers in the config snapshot then we
use the SPIFFE validator as the validation context, rather than the
usual TrustedCA.
The injected validation config includes the local root certificates as
well.
For mTLS to work between two proxies in peered clusters with different root CAs,
proxies need to configure their outbound listener to use different root certificates
for validation.
Up until peering was introduced proxies would only ever use one set of root certificates
to validate all mesh traffic, both inbound and outbound. Now an upstream proxy
may have a leaf certificate signed by a CA that's different from the dialing proxy's.
This PR makes changes to proxycfg and xds so that the upstream TLS validation
uses different root certificates depending on which cluster is being dialed.
This is the OSS portion of enterprise PRs 1904, 1905, 1906, 1907, 1949,
and 1971.
It replaces the proxycfg manager's direct dependency on the agent cache
with interfaces that will be implemented differently when serving xDS
sessions from a Consul server.
OSS port of enterprise PR 1822
Includes the necessary changes to the `proxycfg` and `xds` packages to enable
Consul servers to configure arbitrary proxies using catalog data.
Broadly, `proxycfg.Manager` now has public methods for registering,
deregistering, and listing registered proxies — the existing local agent
state-sync behavior has been moved into a separate component that makes use of
these methods.
When an xDS session is started for a proxy service in the catalog, a goroutine
will be spawned to watch the service in the server's state store and
re-register it with the `proxycfg.Manager` whenever it is updated (and clean
it up when the client goes away).
OSS portion of enterprise PR 1857.
This removes (most) references to the `cache.UpdateEvent` type in the
`proxycfg` package.
As we're going to be direct usage of the agent cache with interfaces that
can be satisfied by alternative server-local datasources, it doesn't make
sense to depend on this type everywhere anymore (particularly on the
`state.ch` channel).
We also plan to extract `proxycfg` out of Consul into a shared library in
the future, which would require removing this dependency.
Aside from a fairly rote find-and-replace, the main change is that the
`cache.Cache` and `health.Client` types now accept a callback function
parameter, rather than a `chan<- cache.UpdateEvents`. This allows us to
do the type conversion without running another goroutine.
Prior to this PR for the envoy xDS golden tests in the agent/xds package we
were hand-creating a proxycfg.ConfigSnapshot structure in the proper format for
input to the xDS generator. Over time this intermediate structure has gotten
trickier to build correctly for the various tests.
This PR proposes to switch to using the existing mechanism for turning a
structs.NodeService and a sequence of cache.UpdateEvent copies into a
proxycfg.ConfigSnapshot, as that is less error prone to construct and aligns
more with how the data arrives.
NOTE: almost all of this is in test-related code. I tried super hard to craft
correct event inputs to get the golden files to be the same, or similar enough
after construction to feel ok that i recreated the spirit of the original test
cases.
Due to timing, a transparent proxy could have two upstreams to dial
directly with the same address.
For example:
- The orders service can dial upstreams shipping and payment directly.
- An instance of shipping at address 10.0.0.1 is deregistered.
- Payments is scaled up and scheduled to have address 10.0.0.1.
- The orders service receives the event for the new payments instance
before seeing the deregistration for the shipping instance. At this
point two upstreams have the same passthrough address and Envoy will
reject the listener configuration.
To disambiguate this commit considers the Raft index when storing
passthrough addresses. In the example above, 10.0.0.1 would only be
associated with the newer payments service instance.
The gist here is that now we use a value-type struct proxycfg.UpstreamID
as the map key in ConfigSnapshot maps where we used to use "upstream
id-ish" strings. These are internal only and used just for bidirectional
trips through the agent cache keyspace (like the discovery chain target
struct).
For the few places where the upstream id needs to be projected into xDS,
that's what (proxycfg.UpstreamID).EnvoyID() is for. This lets us ALWAYS
inject the partition and namespace into these things without making
stuff like the golden testdata diverge.
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.
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.
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.
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.
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.
