kevinrr888 commented on PR #5813:
URL: https://github.com/apache/accumulo/pull/5813#issuecomment-3214328197
@keith-turner @dlmarion - Some more details that I didn't mention in my
initial comment:
* This test fails in several different possible ways:
* One is the 3 tasks are still executing when we expect 0 that Dave
mentioned. I confirmed that only 2 FATE ops were created and reserved in this
case, so my only idea is that a TransactionRunner is somehow stuck/still
running (which again, does not seem possible given the shutdown logic). The
other 2 threads may be stuck because the test failed before the latch
countdown, I can't remember when the failure occurs for this case. Either way,
it seems at least one TransactionRunner is stuck in the while loop.
* Another failure occurs at this `waitFor`:
`fate.getTotalTxRunnersActive() == newNumWorkersSet1 + 1 + numWorkersSet3 +
numWorkersSet4`. For the 30 second period, it will see that the pool which is
shutdown due to config changes has 2 threads still running instead of the 1
expected (the only 1 stuck should be the worker working on the transaction).
This causes the `waitFor` to fail, then the shutdown will fail and hide the
`waitFor` exception since the shutdown is in a `finally`.
* Another failure occurs after ALL checks and waitFors in the try occur
(confirmed with prints). So the very last assertion in the try passes. Then
when it tries to shutdown fate, one thread is stuck (even though we've
confirmed our FATE work has completed successfully).
All these failure cases appear to show that a thread is working on some
phantom FATE transaction (seems impossible given how FATE and this test are
setup), is continuously executing the `while` (seems impossible given the FATE
shutdown logic), or got blocked on some operation in the `while` (from
jstacking the process, the thread was stuck on/repeatedly retrying the
`poll()`, which again seems impossible)
--
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]
