Problem: in load balancing environment time-to-time sessions are dead in Active sessions pool. (ad_session.isProcessed=N). Actually Sessions are "deleted" when server started. (ad_session.isProcessed=Y - according to discussion with CarlosRuiz.)
Solution: We agree with CarlosRuiz would be solution to improve shutdown process by - similar to start session - to "clean ad_session"
IMPORTANT: server must process only related session, not other servers session.
hi , not only case use kill, it's cover case power suddenly turn off, or something make server crash or wrong network.
I see "ServerName column" but maybe i just think over. in test environment sometime you have to run muilti instance of idempiere on same PC
Yep - ServerName has more information, like port and IP I think.
So, if I understood in your approach other servers can mark sessions as finished when one server fails - is there one leading that task? Or all of them do it?
Can it be done as a scheduled process specific to an IP address? That checks (ping) the server in some way to see if is alive?
yes. it done at step 3
3. server will update active time of record each 5 minute (this value can configuration)
i create a new view to fill out session of server have active time > 5 minute (this value can configuration)
other server not need ping to check status of one server.
Revisiting this - I wonder if this can be done as a process, maybe using some kind of hazelcast broadcasting (like KillAllSession).
If this is done as a process then is easy to schedule it to run periodically.
However, I'm not sure if that's a good option to have by default - maybe better to have it as a plugin. There is always the risk of marking a session that is still alive but that was somehow disconnected from the hazelcast.
Regards,
Carlos Ruiz
Hi Carlos,
yes this isgood to revisit and resolve.
we tried to implement count user sessions, but we totally fall while unpredictable dead sessions. so simple cant count.
i see 2 solutions now
1. from experience - users mostly login many times - various circumstances - after close browser, multiple computer login sessions with same user etc - usually session not finished correctly with logout. So they create new session record - and old - stalled - are not touched - we can apply a simple approach. When user login mark all previous - live/dead user sessions as done. This way user sessions are cleaned ongoing (similar when server is restarted). Here would be serious to broadcast a message you was kicked out while parallel login.
2. for periodic cleanup - eg nightly - process can be fine.
hope helps
norbert