Merge pull request #76 from openshift-cherrypick-robot/cherry-pick-75…

…-to-release-0.5 [release-0.5] Fixing wrong scenario of control-plane reboot
medik8s · Dec 25, 2022 · 3e60fd9 · 3e60fd9
2 parents c901f32 + dc8c666
commit 3e60fd9
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 7 deletions.
diff --git a/pkg/apicheck/check.go b/pkg/apicheck/check.go
@@ -193,7 +193,7 @@ func (c *ApiConnectivityCheck) getWorkerPeersResponse() peers.Response {
 		return peers.Response{IsHealthy: false, Reason: peers.UnHealthyBecauseNodeIsIsolated}
 	} else {
 		c.config.Log.Info("Ignoring no peers response error, time is below threshold for no peers response", "time without peers response (seconds)", now.Sub(c.timeOfLastPeerResponse).Seconds(), "threshold (seconds)", c.config.MaxTimeForNoPeersResponse.Seconds())
-		return peers.Response{IsHealthy: true, Reason: peers.HealthyBecauseNoPeersResponseNotReachedMaxAttempts}
+		return peers.Response{IsHealthy: true, Reason: peers.HealthyBecauseNoPeersResponseNotReachedTimeout}
 	}
 
 }

diff --git a/pkg/controlplane/manager.go b/pkg/controlplane/manager.go
@@ -65,7 +65,7 @@ func (manager *Manager) IsControlPlaneHealthy(workerPeerResponse peers.Response,
 	case peers.UnHealthyBecauseNodeIsIsolated:
 		return canOtherControlPlanesBeReached
 	//reported healthy by worker peers
-	case peers.HealthyBecauseErrorsThresholdNotReached, peers.HealthyBecauseCRNotFound:
+	case peers.HealthyBecauseErrorsThresholdNotReached, peers.HealthyBecauseCRNotFound, peers.HealthyBecauseNoPeersResponseNotReachedTimeout:
 		return true
 	//controlPlane node has connection to most workers, we assume it's not isolated (or at least that the controlPlane node that does not have worker peers quorum will reboot)
 	case peers.HealthyBecauseMostPeersCantAccessAPIServer:

diff --git a/pkg/peers/response.go b/pkg/peers/response.go
@@ -8,11 +8,11 @@ type Response struct {
 type reason string
 
 const (
-	HealthyBecauseCRNotFound                           reason = "CR Not found, node is considered healthy"
-	HealthyBecauseErrorsThresholdNotReached            reason = "Errors number hasn't reached threshold not querying peers yet, node is considered healthy"
-	HealthyBecauseNoPeersResponseNotReachedMaxAttempts reason = "No response from peer hasn't passed the non responsive time threshold so still considered healthy"
-	HealthyBecauseNoPeersWereFound                     reason = "No Peers where found, node is considered healthy"
-	HealthyBecauseMostPeersCantAccessAPIServer         reason = "Most peers couldn't access API server, node is considered healthy"
+	HealthyBecauseCRNotFound                       reason = "CR Not found, node is considered healthy"
+	HealthyBecauseErrorsThresholdNotReached        reason = "Errors number hasn't reached threshold not querying peers yet, node is considered healthy"
+	HealthyBecauseNoPeersResponseNotReachedTimeout reason = "No response from peer. The duration of peer not responding hasn't passed the threshold so still considered healthy"
+	HealthyBecauseNoPeersWereFound                 reason = "No Peers where found, node is considered healthy"
+	HealthyBecauseMostPeersCantAccessAPIServer     reason = "Most peers couldn't access API server, node is considered healthy"
 
 	UnHealthyBecausePeersResponse  reason = "Node is reported unhealthy by it's peers"
 	UnHealthyBecauseNodeIsIsolated reason = "Node is isolated, node is considered unhealthy"