RE: [HACKERS] Sure enough, SI buffer overrun is broken

> -----Original Message-----
> From: owner-pgsql-hackers(at)postgreSQL(dot)org
> [mailto:owner-pgsql-hackers(at)postgreSQL(dot)org]On Behalf Of Tom Lane
>
> I built the current sources with MAXNUMMESSAGES set to 32 in
> src/include/storage/sinvaladt.h. The regular regress tests
> run OK, with just a few NOTICEs about 'cache state reset'
> and 'SI buffer overflow' inserted in the normal outputs
> (as you'd expect, if SI overrun occurs).
>
> However, the parallel tests crash spectacularly, with weird errors
> and Assert() coredumps.

Unfortunately,I don't understand the cause yet.
The cause may be not unique.

Is the call RelationCacheInvalidate(false not true) in ResetSys-
temCaches() right ? Relation descriptors would be destoryed if
ResetSystemCaches() occurs in CommandConterIncrement().
Recent change setheapoverride -> CommandCounterIncrement
may need reopen of relations after CommandCounterIncrement.

static void
ResetSystemCaches()
{
ResetSystemCache();
RelationCacheInvalidate(false);
^^^^^^^^
}

I changed false -> true and tried.
Crash decreased but still occurs.

> Some of the unexpected messages in the
> postmaster log are:
>
> NOTICE: LockRelease: locktable lookup failed, no lock

I have seen this NOTICE only once or twice.
This seems because of setting MyProc->xid to InvalidTransa
ctionId in CommitTransaction() and AbortTransaction().
There's a little time until AtCommit(Abort)_Locks.
I have no idea to solve this now.

Regards.

Hiroshi Inoue
Inoue(at)tpf(dot)co(dot)jp