On Sat, May 22, 2021 at 11:00 AM osumi.takami...@fujitsu.com
<osumi.takami...@fujitsu.com> wrote:
> On Friday, May 21, 2021 9:45 PM I worte:
> > On Friday, May 21, 2021 4:43 PM Amit Langote <amitlangot...@gmail.com>
> > wrote:
> > > On Fri, May 21, 2021 at 3:55 PM osumi.takami...@fujitsu.com
> > > <osumi.takami...@fujitsu.com> wrote:
> > > > But, I've detected segmentation faults caused by the patch, which
> > > > can happen during 100_bugs.pl in src/test/subscription.
> > >
> > > Hmm, maybe get_rel_syn_entry() should explicitly set map to NULL when
> > > first initializing an entry.  It's possible that without doing so, the
> > > map remains set to a garbage value, which causes the invalidation
> > > callback that runs into such partially initialized entry to segfault
> > > upon trying to deference that garbage pointer.
> > Just in case, I prepared a new PG and
> > did a check to make get_rel_sync_entry() print its first pointer value with 
> > elog.
> > Here, when I executed 100_bugs.pl, I got some garbage below.
> >
> > * The change I did:
> > @@ -1011,6 +1011,7 @@ get_rel_sync_entry(PGOutputData *data, Oid relid)
> >                 entry->pubactions.pubinsert =
> > entry->pubactions.pubupdate =
> >                         entry->pubactions.pubdelete =
> > entry->pubactions.pubtruncate = false;
> >                 entry->publish_as_relid = InvalidOid;
> > +               elog(LOG, "**> the pointer's default value : %p",
> > + entry->map);
> >         }
> >
> (snip)
> >
> > So, your solution is right, I think.
> This was a bit indirect.
> I've checked the core file of v3's failure core and printed the entry
> to get more confidence. Sorry for inappropriate measure to verify the 
> solution.
>
> $1 = {relid = 16388, schema_sent = false, streamed_txns = 0x0, 
> replicate_valid = false, pubactions = {pubinsert = false, pubupdate = false, 
> pubdelete = false, pubtruncate = false}, publish_as_relid = 16388,
>   map = 0x7f7f7f7f7f7f7f7f}
>
> Yes, the process tried to free garbage.
> Now, we are convinced that we have addressed the problem. That's it !

Thanks for confirming that.

-- 
Amit Langote
EDB: http://www.enterprisedb.com


Reply via email to