On 3/26/22 17:19, Tom Lane wrote: > Andrew Dunstan <and...@dunslane.net> writes: >> On 3/26/22 15:49, Andres Freund wrote: >>> One interesting bit in the config is: >>> [ lack of ] >>> 'update_process_title = off' >> I'd forgotten about that. Let me do that for REL_14_STABLE and see where >> we get to. > Hm. But if that does mitigate it, it still seems like a bug no? > Why would that be preferentially crashing partitioned-table creation?
Yes it seems like a bug, but hard to diagnose. It seemed like a bug back in May: see <https://postgr.es/m/4baee39d-0ebe-8327-7878-5bc11c95e...@dunslane.net> I vaguely theorize about a buffer overrun somewhere that scribbles on the stack. The answer to Andres's question about where the stackdumps go is that they go in the data directory, AFAIK. You can see the buildfarm logic for collecting them at <https://github.com/PGBuildFarm/client-code/blob/main/PGBuild/Utils.pm> starting at line 149. There are various appropriate invocations of get_stack_trace() in run_build.pl. cheers andrew -- Andrew Dunstan EDB: https://www.enterprisedb.com
