// updateACLReplicationStatus safely updates the ACL replication status.
func (s *Server) updateACLReplicationStatus(status structs.ACLReplicationStatus) {
// Fixup the times to shed some useless precision to ease formattting,
// and always report UTC.
status.LastError = status.LastError.Round(time.Second).UTC()
status.LastSuccess = status.LastSuccess.Round(time.Second).UTC()
// Set the shared state.
s.aclReplicationStatusLock.Lock()
s.aclReplicationStatus = status
s.aclReplicationStatusLock.Unlock()
}
// runACLReplication is a long-running goroutine that will attempt to replicate
// ACLs while the server is the leader, until the shutdown channel closes.
func (s *Server) runACLReplication() {
var status structs.ACLReplicationStatus
status.Enabled = true
status.SourceDatacenter = s.config.ACLDatacenter
s.updateACLReplicationStatus(status)
// Show that it's not running on the way out.
defer func() {
status.Running = false
s.updateACLReplicationStatus(status)
}()
// Give each server's replicator a random initial phase for good
// measure.
select {
case <-s.shutdownCh:
return
case <-time.After(lib.RandomStagger(s.config.ACLReplicationInterval)):
}
// We are fairly conservative with the lastRemoteIndex so that after a
// leadership change or an error we re-sync everything (we also don't
// want to block the first time after one of these events so we can
// show a successful sync in the status endpoint).
var lastRemoteIndex uint64
replicate := func() {
if !status.Running {
lastRemoteIndex = 0 // Re-sync everything.
status.Running = true
s.updateACLReplicationStatus(status)
s.logger.Printf("[INFO] consul: ACL replication started")
}
index, err := s.replicateACLs(lastRemoteIndex)
if err != nil {
lastRemoteIndex = 0 // Re-sync everything.
status.LastError = time.Now()
s.updateACLReplicationStatus(status)
s.logger.Printf("[WARN] consul: ACL replication error (will retry if still leader): %v", err)
} else {
lastRemoteIndex = index
status.ReplicatedIndex = index
status.LastSuccess = time.Now()
s.updateACLReplicationStatus(status)
s.logger.Printf("[DEBUG] consul: ACL replication completed through remote index %d", index)
}
}
pause := func() {
if status.Running {
lastRemoteIndex = 0 // Re-sync everything.
status.Running = false
s.updateACLReplicationStatus(status)
s.logger.Printf("[INFO] consul: ACL replication stopped (no longer leader)")
}
}
// This will slowly poll to see if replication should be active. Once it
// is and we've caught up, the replicate() call will begin to block and
// only wake up when the query timer expires or there are new ACLs to
// replicate. We've chosen this design so that the ACLReplicationInterval
// is the lower bound for how quickly we will replicate, no matter how
// much ACL churn is happening on the remote side.
//
// The blocking query inside replicate() respects the shutdown channel,
// so we won't get stuck in here as things are torn down.
for {
select {
case <-s.shutdownCh:
return
case <-time.After(s.config.ACLReplicationInterval):
if s.IsLeader() {
replicate()
} else {
pause()
}
}
}
}
