Log In - Home Page

Patch: Checksums (initdb-time)

Edit Patch - Move To Another CommitFest - Delete Patch

CommitFest 2013-01
Topic Server Features
Patch Status Committed
Author Simon Riggs, Jeff Davis
Reviewers Greg Smith
Committer Simon Riggs
Close Date 2013-03-23
Patch by jdavis on 2012-11-15 01:42:46 AM: Initial version.
Patch by jdavis on 2012-11-18 08:06:08 PM: Cleaned up patch, and a few changes/fixes.
Patch by jdavis on 2012-11-26 02:35:29 AM: Updated patch. Fixed a bug, improved speed of checksum algorithm.
Comment by simon on 2012-12-03 01:20:29 PM: Extracted 3 generic parts of the patch and committed them. Needs rebase, much smalller/clearer.
Patch by jdavis on 2012-12-04 09:04:18 AM: Rebase and minor cleanup.
Comment by jdavis on 2012-12-19 11:33:25 PM: Performance test results.
Patch by jdavis on 2013-01-17 01:51:25 AM: Test program from Greg Smith.
Patch by jdavis on 2013-01-17 01:51:55 AM: Fixes two bugs.
Patch by jdavis on 2013-01-24 09:04:46 PM: Rebased, guard FSM against torn pages.
Patch by ringerc on 2013-03-03 02:19:35 PM: Rebased, fixed conflicts (Greg)
Comment by ringerc on 2013-03-03 02:20:45 PM: Daniel is considering applying this, shipping it and turning it on. Needs to check whether can be turned off to run without updating checksums in case of problems.
Comment by ringerc on 2013-03-03 02:22:58 PM: Queried status, whether BC test worked out
Review by gsmith on 2013-03-03 11:07:28 PM: Corruption testing review. Some concerns about header corruption leading to page data becoming unavailable.
Comment by gsmith on 2013-03-03 11:08:11 PM: Safety fix to testing corruption test harness in previous message.
Review by gsmith on 2013-03-03 11:09:08 PM: Worst-case performance test results duplicated via test harness. Some concerns about better documenting when these can happen.
Review by gsmith on 2013-03-07 04:19:44 AM: Suggestions to trim down page header use. Is the checksum method used the best approach?
Comment by gsmith on 2013-03-07 04:20:55 AM: Comments on Fletcher-16 vs. CRC16, in-place upgrade
Comment by gsmith on 2013-03-07 04:21:56 AM: Notes on similar use of CRC32C by Linux, potential hardware acceleration
Review by gsmith on 2013-03-07 04:23:10 AM: pgbench performance numbers show 2% performance hit is typical, 20% worse case slowdown expected is seen with some larger databases and low client counts.
Patch by jdavis on 2013-03-07 09:46:48 PM: Rebase, GUC, addressed Heikki's comments (except for checksum algorithm questions).
Patch by jdavis on 2013-03-13 06:34:25 AM: Rebase and fix temp table problem.
Comment by gsmith on 2013-03-20 03:47:04 AM: Initial commits started, using the WAL CRC32 truncated to 16 bits as the checksum. Other checksum computations might be selected based on performance optimization tests. With the initdb time checksum enabling used now, any change to that method just requires a standard catversion bump.

Add Comment

Please log in to comment on this patch.