Cluster leader election multiple issues #860
Comments
|
Following on from this: if I restart a node in the cluster while all are running, the restarted node is always elected as leader. This should not happen if the weights were respected. If the leader is restarted the next-highest weighted node should become leader until the restart has completed and the higher-weighted node takes over. |
|
There hasn't been any activity on this issue recently. In an effort to provide a better overview of current issues we automatically clean some of the old ones. Many of them may already be resolved in newer versions of room-assistant. |
|
Is there a way of reopening this issue? I am seeing the same behaviour on v3 beta 4 (HA addon is beta 3): |
|
I re-opened the issue for transparency. I believe this won't be handled as fixes to the current mechanism though. Instead, #780 seems like the best way forward to me. |
Describe the bug
noneas cluster leader for newly-added nodes.To reproduce
I have 3 nodes:
office,lounge,workshop. I have the peer IDs explicitly listed in each config file, auto discovery is disabled, each node is weighted as 10, 2, and 1, respectively. My MQTT broker is running on a different machine on the local network. Home assistant integration and room-assistant logs is used to view leader election.lounge. Wait for discovery of devicesoffice.officereportsloungeis added to cluster butofficeis not elected as leader.officedoes not log any leader election change. Home Assistant reportsofficecluster leader asnonewith 2 instances in cluster.workshop.loungeandofficeboth reportworkshopas added to cluster.loungeandofficeboth logworkshopas being elected as leader.workshopdoes not log itself as being elected as leader.loungeandofficethen immediately logofficeas elected as leader,workshopdoes not log any leadership change. Home Assistant shows thelounge cluster leaderto beloungewhileofficeandworkshopshow the cluster leader asoffice.Relevant logs
lounge
office
workshop
Relevant configuration
All of the configs look like this. Only the
instanceNameandweightdiffers.Expected behavior
loungeshould first log itself as leader. It's the only node.officeshould be elected as leader when added to the cluster. It has the highest weight (10). However it was not.officeshould publish the current value of the leader (even if incorrect) to MQTT, notnone(which may be the absence of a publication) because no re-election has taken place.loungeshould log the addition ofofficeto the cluster. It does not.workshopshould be added to the cluster but never elected as leader.officeas leader for the cluster as reported by each node's "cluster leader" sensor. It does not.Environment
officeis an Raspberry Pi 4 on ethernet,loungeandworkshopare Raspberry Pi Zero W on Wi-Fi. MQTT broker is hardwired home server running mosquitto in Docker on Ubuntu 21.04 on an Intel Core i5 10400, 32 GB RAM.The text was updated successfully, but these errors were encountered: