On 15.05.12 20:17, Rick Hillegas wrote:
A little less than a week remains before we build the first 10.9 release
candidate. At the end of this message, I summarize the status of
important issues, which people had hoped we would address for 10.9.
At this time, I do not see any reason to push back the 10.9 schedule and
I plan to build a release candidate next Monday. Please let me know if
you disagree with this decision.
Hi Rick,
I'm planning to get some fixes in for update statistics functionality.
The current code has been posted on DERBY-5684, but it may be that I
break it up into a few parts. It will be a little tight, but I hope I
can complete this some time during Monday.
The fixes would address the following issues:
o DERBY-5680: indexStat daemon processing tables over an over even
when there are no changes in the tables ([1])
o DERBY-3790: Investigate if request for update statistics can be
skipped for certain kind of indexes, one instance may be unique indexes
based on one column.
In addition it would optimize how the statistics are generated for
single-column foreign keys backed by a unique index. In this case the
row estimate from the backing index will be used.
To reduce the risk of something going seriously wrong in the wild, I
will include a debug flag that allows people to fall back to the 10.8
behavior ([2]). Combined with the new SYSCS_DROP_STATISTICS system
procedure and the knob to disable the istat daemon, that will hopefully
be enough to work around any issues detected after the release.
The current code passes suites.All, and I've added an upgrade test.
Some adjustments are required for some of our existing tests, since
there will be fewer statistics entries for some tables.
Regards,
--
Kristian
[1] This provides an automatic "self-healing" mechanism, whereas the new
SYSCS_DROP_STATISTICS provides a manual way to fix the problem.
[2] That is, don't drop any statistics automatically, always scan the
indexes, and scan all indexes.
Thanks,
-Rick
------------------------ IMPORTANT ISSUES ------------------------
I believe that the issues with identity/sequence generators have been
addressed (see DERBY-5495). People also expressed interest in tackling
the following issues:
-------
The following issue has been resolved:
DERBY-5667 Instability in UpdateLocksTest
-------
The following issues are being worked on and should be wrapped up by
next Monday:
DERBY-5565 Network Server should reject client connections that are not
Derby Network Cli
DERBY-5691 Document that Write Caching must be disabled to avoid
possible database corruption
DERBY-5508 Improve backup/restore documentation visibility and content
to encourage proper backups and restore procedures
-------
Partial progress has been made on some of the following issues. However,
some of these issues have not attracted a volunteer yet. I do not plan
to hold up the release for any of this stabilization work:
DERBY-4852 Instability in JMXTest.
DERBY-5197 Instability in Replication tests.
DERBY-5600 Instability in Replication tests.
DERBY-5666 Instability in NativeAuthenticationServiceTest
DERBY-5686 Instability in DriverMgrAuthenticationTest.
-------
No one has volunteered to pick up the following new features. I do not
plan to hold up the release, waiting for volunteers.
DERBY-4805 Increase the length of the RDBNAM field in the DRDA
DERBY-5578 Provide a way to invalidate stored prepared statements
-------
A fix has been checked in which addresses the repros attached to the
following issue. However, a user reports another version of the bug. I
am prepared to hold up the release if the new version of the bug can be
scripted. Otherwise, waiting for a repro is an unbounded problem, and I
think it is better to produce a 10.9 release which contains the current,
partial fix:
DERBY-5234 Data corruption caused by inplace table compression
