kotman12 commented on PR #5301:
URL: https://github.com/apache/solr/pull/5301#issuecomment-6752208095
## Setup (same in both cases)
```
ZooKeeper sends, in this order:
B-Connected (seq 5) B-Disconnected (seq 6) B-Expired (seq 7)
Where each event runs:
B-Connected, B-Disconnected -> thread pool (may finish in any order)
B-Expired -> the single Expired thread
The Expired thread is already busy recovering from A's expiry:
it created B and is stuck in: while (!connected) wait();
```
In both cases the pool happens to finish B-Disconnected before B-Connected.
## Before Houston's change: no sequence check
```
time pool thread connected Expired thread
---- --------------------- --------- --------------------------------
t1 false waiting for connected...
t2 apply B-Disconnected false wakes, still false, waits again
t3 apply B-Connected TRUE <-- late, but applied anyway
t4 true wakes, sees true -> EXITS the wait
t5 finishes recovery, thread is free
t6 runs B-Expired
t7 -> starts a new session C, recovers
```
Applying B-Connected late is technically wrong, since B isn't connected at
t3. But it flips the flag, the waiter
exits, the thread frees up, and B-Expired then gets processed. The bad state
is brief and the system recovers.
## After Houston's change: stale events are ignored
```
time pool thread connected Expired thread
---- --------------------- --------- --------------------------------
t1 false waiting for connected...
t2 apply B-Disconnected false wakes, still false, waits again
t3 B-Connected arrives: false (nothing changes)
seq 5 < 6, "stale"
-> IGNORED
t4 false still waiting...
t5 still waiting...
...
B-Expired is queued here:
[ Expired thread: busy waiting ] <- [ B-Expired waiting for this
thread ]
^ |
+------ never released -------------------+
```
Ignoring B-Connected is correct, since B really is disconnected. But that
late `Connected` was the only thing that
ever released the waiter. The one event that could say "give up, B is dead"
is B-Expired, and it needs the very
thread that's stuck.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
