Thread: buildfarm failures after pgstat patch

The buildfarm appears to be failing after the recent pgstat patch.

The failure seems to be caused by this failed assertion, which appears to
occur fairly consistently in the ECPG tests, in the postmaster log:

TRAP: FailedAssertion("!(entry->trans == 0L)", File: "pgstat.c", Line: 696)


-- 
Disco is to music what Etch-A-Sketch is to art.

Jeremy Drake <pgsql@jdrake.com> writes:
> The buildfarm appears to be failing after the recent pgstat patch.

Yeah, just saw that myself.  Fixed the backend-side problem, but it
would be interesting to find out what ECPG is doing that wasn't exposed
by the core regression tests ... maybe we need another regression test.

Also, for me the ecpg "variable" test case seemed to go into an infinite
loop when the backend dies at this stage; that seems like a lack of
robustness on ECPG's part.
        regards, tom lane

On Sun, May 27, 2007 at 01:45:45AM -0400, Tom Lane wrote:
> Yeah, just saw that myself.  Fixed the backend-side problem, but it
> would be interesting to find out what ECPG is doing that wasn't exposed
> by the core regression tests ... maybe we need another regression test.

Don't see anything that special before the backend crashes. Seems to
crash on the very first insert that is done after settign datestyle to
iso.

> Also, for me the ecpg "variable" test case seemed to go into an infinite
> loop when the backend dies at this stage; that seems like a lack of
> robustness on ECPG's part.

Don't worry, it's not an ECPG problem. The regression test just had a
bad way to check for errors. I might have to check all tests again,
because they were doen to check the stuff in ecpg and might not take a
backend crash into regard at all. 

Simple problem this time was that the test only printed out errors and
not stopped.

Fixed in HEAD. 

Michael
-- 
Michael Meskes
Email: Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
ICQ: 179140304, AIM/Yahoo: michaelmeskes, Jabber: meskes@jabber.org
Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!

Michael Meskes <meskes@postgresql.org> writes:
> On Sun, May 27, 2007 at 01:45:45AM -0400, Tom Lane wrote:
>> Yeah, just saw that myself.  Fixed the backend-side problem, but it
>> would be interesting to find out what ECPG is doing that wasn't exposed
>> by the core regression tests ... maybe we need another regression test.

> Don't see anything that special before the backend crashes.

Actually, what was provoking it was that several of the ECPG tests
disconnect in the middle of a transaction, which was exposing the fact
that pgstat's on_proc_exit hook ran before we'd performed the abort
in ShutdownPostgres.  The Asserts I'd sprinkled in there to test the
transaction-awareness logic were unhappy.  The visible failure would
be at some random later point, when the crash-recovery logic killed the
backend running the next test.

Is it worth adding something to the regular regression tests to exercise
that code path?
        regards, tom lane