This continues the work done in #14908 where a crude solution to prevent a
goroutine leak was implemented. The former code would launch a perpetual
goroutine family every iteration (+1 +1) and the fixed code simply caused a
new goroutine family to first cancel the prior one to prevent the
leak (-1 +1 == 0).
This PR refactors this code completely to:
- make it more understandable
- remove the recursion-via-goroutine strangeness
- prevent unnecessary RPC fetches when the prior one has errored.
The core issue arose from a conflation of the entry.Fetching field to mean:
- there is an RPC (blocking query) in flight right now
- there is a goroutine running to manage the RPC fetch retry loop
The problem is that the goroutine-leak-avoidance check would treat
Fetching like (2), but within the body of a goroutine it would flip that
boolean back to false before the retry sleep. This would cause a new
chain of goroutines to launch which #14908 would correct crudely.
The refactored code uses a plain for-loop and changes the semantics
to track state for "is there a goroutine associated with this cache entry"
instead of the former.
We use a uint64 unique identity per goroutine instead of a boolean so
that any orphaned goroutines can tell when they've been replaced when
the expiry loop deletes a cache entry while the goroutine is still running
and is later replaced.
To support Destinations on the service-defaults (for tproxy with terminating gateway), we need to now also make servers watch service-defaults config entries.
This commit updates the establish endpoint to bubble up a 403 status
code to callers when the establishment secret from the token is invalid.
This is a signal that a new peering token must be generated.
* peering: skip register duplicate node and check from the peer
* Prebuilt the nodes map and checks map to avoid repeated for loop
* use key type to struct: node id, service id, and check id
Fix an issue where rpc_hold_timeout was being used as the timeout for non-blocking queries. Users should be able to tune read timeouts without fiddling with rpc_hold_timeout. A new configuration `rpc_read_timeout` is created.
Refactor some implementation from the original PR 11500 to remove the misleading linkage between RPCInfo's timeout (used to retry in case of certain modes of failures) and the client RPC timeouts.
There is a bug in the error handling code for the Agent cache subsystem discovered:
1. NotifyCallback calls notifyBlockingQuery which calls getWithIndex in
a loop (which backs off on-error up to 1 minute)
2. getWithIndex calls fetch if there’s no valid entry in the cache
3. fetch starts a goroutine which calls Fetch on the cache-type, waits
for a while (again with backoff up to 1 minute for errors) and then
calls fetch to trigger a refresh
The end result being that every 1 minute notifyBlockingQuery spawns an
ancestry of goroutines that essentially lives forever.
This PR ensures that the goroutine started by `fetch` cancels any prior
goroutine spawned by the same line for the same key.
In isolated testing where a cache type was tweaked to indefinitely
error, this patch prevented goroutine counts from skyrocketing.
In practice this was masked by #14956 and was only uncovered fixing the
other bug.
go test ./agent -run TestAgentConnectCALeafCert_goodNotLocal
would fail when only #14956 was fixed.
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.
When peering through mesh gateways we expect outbound dials to peer
servers to flow through the local mesh gateway addresses.
Now when establishing a peering we get a list of dial addresses as a
ring buffer that includes local mesh gateway addresses if the local DC
is configured to peer through mesh gateways. The ring buffer includes
the mesh gateway addresses first, but also includes the remote server
addresses as a fallback.
This fallback is present because it's possible that direct egress from
the servers may be allowed. If not allowed then the leader will cycle
back to a mesh gateway address through the ring.
When attempting to dial the remote servers we retry up to a fixed
timeout. If using mesh gateways we also have an initial wait in
order to allow for the mesh gateways to configure themselves.
Note that if we encounter a permission denied error we do not retry
since that error indicates that the secret in the peering token is
invalid.
memdb's `WatchCh` method creates a goroutine that will publish to the
returned channel when the watchset is triggered or the given context
is canceled. Although this is called out in its godoc comment, it's
not obvious that this method creates a goroutine who's lifecycle you
need to manage.
In the xDS capacity controller, we were calling `WatchCh` on each
iteration of the control loop, meaning the number of goroutines would
grow on each autopilot event until there was catalog churn.
In the catalog config source, we were calling `WatchCh` with the
background context, meaning that the goroutine would keep running after
the sync loop had terminated.
R.B. Boyer
Chris S. Kim
cskh
Iryna Shustava
Venu Yanamandra
freddygv
Nitya Dhanushkodi
Kyle Havlovitz
Dan Upton
Derek Menteer
Freddy
malizz