Re: core dumps in auto_prewarm, tests succeed

On Tue, Jan 23, 2024 at 06:33:25PM +0100, Alvaro Herrera wrote:
> On 2024-Jan-22, Nathan Bossart wrote:
>> This might be a topic for another thread, but I do wonder whether we could
>> put a generic pg_controldata check in node->stop that would at least make
>> sure that the state is some sort of expected shut-down state.  My first
>> thought is that it could be a tad expensive, but... maybe it wouldn't be
>> too bad.
> 
> Does this actually detect a problem if you take out the fix?  I think
> what's going to happen is that postmaster is going to crash, then do the
> recovery cycle, then stop as instructed by the test; so pg_controldata
> would report that it was correctly shut down.

Yes, the control data shows "in production" without it.  The segfault
happens within the shut-down path, and the test logs indicate that the
server continues shutting down without doing a recovery cycle:

2024-01-23 12:14:49.254 CST [2376301] LOG:  received fast shutdown request
2024-01-23 12:14:49.254 CST [2376301] LOG:  aborting any active transactions
2024-01-23 12:14:49.255 CST [2376301] LOG:  background worker "logical replication launcher" (PID 2376308) exited with
exitcode 1
 
2024-01-23 12:14:49.256 CST [2376301] LOG:  background worker "autoprewarm leader" (PID 2376307) was terminated by
signal11: Segmentation fault
 
2024-01-23 12:14:49.256 CST [2376301] LOG:  terminating any other active server processes
2024-01-23 12:14:49.257 CST [2376301] LOG:  abnormal database system shutdown
2024-01-23 12:14:49.261 CST [2376301] LOG:  database system is shut down

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com