Hi All,

> I forgot to mention that Ashutosh has tested this patch for a day
> using Jeff's tool and he didn't found any problem.  Also, he has found
> a way to easily reproduce the problem.  Ashutosh, can you share your
> changes to the script using which you have reproduce the problem?

I made slight changes in Jeff Janes patch (crash_REL10.patch) to
reproduce the issue reported by him earlier [1]. I changed the
crash_REL10.patch such that a torn page write happens only for a
bitmap page in hash index. As described by Amit in [2] & [3] we were
not logging full page image for bitmap page and that's the reason we
were not be able to restore bitmap page in case it is a torn page
which would eventually result in below error.

38422  01000 2016-09-19 16:25:50.061 PDT:WARNING:  page verification
failed, calculated checksum 65067 but expected 21260
38422  01000 2016-09-19 16:25:50.061 PDT:CONTEXT:  xlog redo at
3F/22053B50 for Hash/ADD_OVFL_PAGE: bmsize 4096, bmpage_found T
38422  XX001 2016-09-19 16:25:50.071 PDT:FATAL:  invalid page in block
9 of relation base/16384/17334

Jeff Jane's original patch had following conditions for a torn page write:

if (JJ_torn_page > 0 && counter++ > JJ_torn_page && !RecoveryInProgress())
    nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ/3);

Ashutosh has added some more condition's in above if( ) to ensure that
the issue associated with bitmap page gets reproduced easily:

if (JJ_torn_page > 0 && counter++ > JJ_torn_page &&
!RecoveryInProgress() && bucket_opaque->hasho_page_id == HASHO_PAGE_ID
&& bucket_opaque->hasho_flag & LH_BITMAP_PAGE)
    nbytes = FileWrite(v->mdfd_vfd, buffer, 24);

Also, please note that i am allowing only 24 bytes of data i.e. just a
page header to be written into a disk file so that even if a single
overflow page is added into hash index table the issue will be

Note: You can find Jeff Jane's patch for torn page write at - [4].

[1] - 
[2] - 
[3] - 
[4] - 

With Regards,
Ashutosh Sharma
EnterpriseDB: http://www.enterprisedb.com

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to