On 27.01.11 18:16, Rick Hillegas wrote:
Hi Kristian,
Thanks for taking care of this while you are on leave. +1 to checking
in your patches. More inline...
FYI, I have now committed the remaining patches.
I have enabled logging by default for now, and the feature is also
enabled by default.
Both the feature itself and the logging can be turned off explicitly
using properties (I think they have all been described in DERBY-4771).
The easiest way to see if work is ongoing is to obtain a stack trace and
see if the index statistics daemon thread is present. That thread will
be created when needed and go away when there is no more work to do, so
this technique is awkward for small tables...
Grepping for "istat" in derby.log will give you some information about
what has happened. A summary will be printed to the log if you shut down
the database properly.
You can also enable tracing for more detailed information. To see what
the output looks like, run the test store.AutomaticIndexStatisticsTest
with tracing/logging enabled.
Note that the feature is still somewhat experimental. Feel free to ask
questions, I'll try to answer them as soon as I can. Again, I think
DERBY-4771 is the issue that contains most information about the feature.
Regards,
--
Kristian
On 1/26/11 12:11 PM, Kristian Waagan wrote:
Hello,
I plan to enable automatic calculation of index cardinality
statistics by committing DERBY-4938 and DERBY-4939 (I have broken up
the code from the latest DERBY-4771 patch, the patches are just not
accessable for me right now).
Before I do that, I want to ask is people think the regression tests
are stable enough.
I think that Dag has just checked in a fix for an instability in the
regression tests related to interrupt handling. So that instability
should go away and there should be less noise obscuring instabilities
introduced by your statistics work, if any.
Enabling the auto-calculation may cause intermittent failures, since
the work is carried out in a background thread. Tests that may
experience failures are tests asserting on statistics and locks, and
which also perform SQL queries using indexes ([1]). If you suspect
the feature has caused an error, the first thing to do is to grep for
"istat" in derby.log.
It would be helpful wrt debugging if there aren't many intermittent
failures popping up already.
I can hold off for a while longer, but I want the feature to be
tested of many different types of operating systems and hardware
before 10.8.
Yes, please.
Thanks,
-Rick
Opinions?
Regards,
