Thanks Imran. Created an issue https://github.com/apache/helix/issues/691 for tracking that. I think we may need more discussion on the expected behavior and how to handle it.
Feel free to update the issue with any suggestions or further questions. Best, Junkai On Sun, Jan 19, 2020 at 9:18 PM kishore g <[email protected]> wrote: > That's a great idea. Handling the actual session expiry or re-connect > event at a later time might be a bit tricky but doable. > > thanks > Kishore G > > On Sun, Jan 19, 2020 at 8:35 PM Imran S. Shah <[email protected]> > wrote: > >> Hi, >> >> I have a helix cluster setup with a resource in MASTER-SLAVE >> configuration with 3 replicas. We are testing what happens when a client >> temporarily looses network connectivity to the ZK cluster. The observations >> are: >> 1. The client continues to think it is MASTER (DISCONNECT event is >> triggered in the background and reconnecting messages are seen) >> 2. The controller makes another node MASTER >> 3. Once the client is able to reconnect a session expired is triggered, >> and reset is called on the MASTER-SLAVE, moving the partitions to OFFLINE >> state. >> >> I would like the node to stop serving requests as MASTER as soon as >> client detects that session expiration timeout has passed even though the >> session may not be expired. Curator framework injects a session timeout >> event based on percent of the negotiated timeout, and I was hoping that in >> Helix something similar could be configured. Otherwise there is a state >> where multiple nodes think they are masters. >> >> Thanks, >> Imran >> > -- Junkai Xue
