On Tue, Apr 2, 2019 at 10:59 PM ANASTACIO Tiago
<tiago_anastacio@yahoo.fr> wrote:
> A pity background worker are no more verbose (DEBUG5) than:
>
> 2019-04-02 09:42:38.993 GMT [17654] ERROR: invalid name syntax
I agree, it would be nice to improve that. It seems to be a general
problem with any GUC that requires a transaction to check: parallel
workers will force that, but you might not understand the error
because it comes with no context. That gives me an idea: perhaps we
should set the elog() context while restoring GUCs, so that it is
displayed in the error output somehow.
> About Tracing
> I used strace and grep background woker pid. But I didn't get that trace, which tool did you use ?
I modified the source so that all the elog() calls that produce that
message were at PANIC level (which means abort(), which delivers
SIGABRT, which normally dumps core) instead of ERROR level, then I ran
ulimit -c unlimited to make sure my OS would dump core files before
starting the server, then I triggered the error, then I loaded the
resulting core file into my debugger and ran bt to see the backtrace
and understand how we finished up there.
--
Thomas Munro
https://enterprisedb.com