fix(leader-election): introduce delay before attempting to reacquire …

…leadership (#794) Fixes #785. Introduce a retry delay before attempting to reacquire leadership, fixing the log noise generated from the API server being unavailable.
buehler · Jul 25, 2024 · 0965d23 · 0965d23
1 parent d026e3b
commit 0965d23
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 3 deletions.
diff --git a/src/KubeOps.Operator/LeaderElection/LeaderElectionBackgroundService.cs b/src/KubeOps.Operator/LeaderElection/LeaderElectionBackgroundService.cs
@@ -82,6 +82,8 @@ static async ValueTask CastAndDispose(IDisposable resource)
 
     private async Task RunAndTryToHoldLeadershipForeverAsync()
     {
+        uint leadershipRetries = 0;
+
         while (!_cts.IsCancellationRequested)
         {
             try
@@ -94,7 +96,14 @@ private async Task RunAndTryToHoldLeadershipForeverAsync()
             }
             catch (Exception exception)
             {
-                logger.LogError(exception, "Failed to hold leadership.");
+                leadershipRetries++;
+
+                var delay = TimeSpan
+                    .FromSeconds(Math.Pow(2, Math.Clamp(leadershipRetries, 0, 5)))
+                    .Add(TimeSpan.FromMilliseconds(new Random().Next(0, 1000)));
+
+                logger.LogError(exception, "Failed to hold leadership. Wait {Seconds}s before attempting to reacquire leadership.", delay.TotalSeconds);
+                await Task.Delay(delay);
             }
         }
     }

diff --git a/test/KubeOps.Operator.Test/LeaderElector/LeaderElectionBackgroundService.Test.cs b/test/KubeOps.Operator.Test/LeaderElector/LeaderElectionBackgroundService.Test.cs
@@ -50,8 +50,8 @@ public async Task Elector_Throws_Should_Retry()
         await leaderElectionBackgroundService.StartAsync(CancellationToken.None);
 
         // Starting the background service should result in the lock attempt throwing, and then a subsequent attempt being made.
-        // Wait for the subsequent event to be signalled, if we time out the test fails.
-        electionLockSubsequentCallEvent.WaitOne(TimeSpan.FromMilliseconds(500)).Should().BeTrue();
+        // Wait for the subsequent event to be signalled, if we time out the test fails. The retry delay requires us to wait at least 3 seconds.
+        electionLockSubsequentCallEvent.WaitOne(TimeSpan.FromMilliseconds(3100)).Should().BeTrue();
 
         await leaderElectionBackgroundService.StopAsync(CancellationToken.None);
     }