Add a crash recovery testcase for innodb_log_block_size

[Laurynas Biveinis]

Without testing innodb_log_block_size in crash recovery log parse, we cannot recommend this option for any production setup.

Thus add a testcase that stops server checkpoint activity, writes a variety of log records, crashes the server, and verifies the data after the crash recovery.

Reuse the testcase for default log block size too, as MySQL doesn't have good public testing coverage in that area.

[Laurynas Biveinis (laurynas-biveinis)]

See bug 1242563.

[Alexey Kopytov (akopytov)]

This has been covered by the XtraBackup test suite for quite some time.

[Laurynas Biveinis (laurynas-biveinis)]

That's of course good, and partially a reason the triage was not
higher than Wishlist, but shouldn't the server testing be
self-contained? We have archived log and bitmap testcases in the
server that partially overlap with the corresponding XtraBackup
testcases, don't we? XtraBackup testsuite and source code rebases
happen less often than PS point version merges and releases. A
regression there might go undetected for months. In fact, right now
the XB 2.1 trunk has no testing for log block size for Percona Server
5.6 because of this reason.

Please do not close this bug again until the above is discussed.

[Alexey Kopytov (akopytov)]

The server testing should not be, cannot be and is never self-contained, otherwise there would be no need for other tools like RQG.
I don't see any overlap between bitmap/archiving test cases, both server and XB ones test different code paths and generally provide testing from different perspectives.

I agree this request is borderline Invalid/Wishlist, but given the extremely low relevance and MTR's limitations in testing scenarios like this, I lean more towards Invalid.

And by the way, the XB 2.1 trunk does have testing for innodb_log_block_size for Percona Server 5.6.

[Laurynas Biveinis (laurynas-biveinis)]

Alexey -

I agree that not everything can be nor should be covered by MTR. I
just happened to believe that MTR is the right place in this case.
Such MTR testcase right now would need new server-side support to
pause checkpoint activity, other than that, I don't see why it would
be not straightforward. Let's say RQG / XB / etc. testing shows a
bug. Do we not add an MTR regression testcase then?

You say that XB 2.1 trunk tests innodb_log_block_size for XtraDB
5.6. Since that disagreed with my findings (yes - I checked), I had
to check some more to find out that currently the testing happens only
on CentOS, but it will be fixed on the next test PS binary rebase.

As I have already asked, what about the current situation that XB
testsuite and source rebases happen once in a few months, providing no
testing coverage for the server versions released in the meantime?

I have also kindly asked not to close the bug with an unfinished
discussion.

Thanks,
Laurynas

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3042