Decouple xds capacity controller and raft-autopilot (#20511)

Decouple xds capacity controller and autopilot

This prevents a potential bug where autopilot deadlocks while attempting
to execute `AutopilotDelegate.NotifyState()` on an xdscapacity controller
that stopped consuming messages.
This commit is contained in:
Derek Menteer 2024-02-08 15:31:44 -06:00 committed by GitHub
parent 26661a1c3b
commit a1c8d4dd19
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 8 additions and 4 deletions

3
.changelog/20511.txt Normal file
View File

@ -0,0 +1,3 @@
```release-note:bug
connect: Remove code coupling where the xDS capacity controller could negatively affect raft autopilot performance.
```

View File

@ -15,6 +15,7 @@ import (
"golang.org/x/time/rate"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib/channels"
"github.com/hashicorp/consul/lib/retry"
)
@ -72,7 +73,7 @@ type SessionLimiter interface {
func NewController(cfg Config) *Controller {
return &Controller{
cfg: cfg,
serverCh: make(chan uint32),
serverCh: make(chan uint32, 1),
doneCh: make(chan struct{}),
}
}
@ -109,9 +110,9 @@ func (c *Controller) Run(ctx context.Context) {
// SetServerCount updates the number of healthy servers that is used when
// determining capacity. It is called by the autopilot delegate.
func (c *Controller) SetServerCount(count uint32) {
select {
case c.serverCh <- count:
case <-c.doneCh:
if err := channels.DeliverLatest(count, c.serverCh); err != nil {
// This should not be possible since only one producer should ever exist, but log anyway.
c.cfg.Logger.Error("failed to deliver latest server count to xDS capacity controller")
}
}