Re: Write Ahead Logging for Hash Indexes

On Wed, Sep 14, 2016 at 4:36 PM, Ashutosh Sharma <ashu(dot)coek88(at)gmail(dot)com> wrote:
> Hi All,
>
> Below is the backtrace for the issue reported in my earlier mail [1].
> From the callstack it looks like we are trying to release lock on a
> meta page twice in _hash_expandtable().
>

Thanks for the call stack. I think below code in patch is culprit.
Here we have already released the meta page lock and then again on
failure, we are trying to release it.

_hash_expandtable()
{
..

/* Release the metapage lock, before completing the split. */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
..

if (!buf_nblkno)
{
_hash_relbuf(rel, buf_oblkno);
goto fail;
}
..
fail:
/* We didn't write the metapage, so just drop lock */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
}

This is a problem of concurrent hash index patch. I will send the fix
in next version of the patch.

--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com