You are viewing a plain text version of this content. The canonical link for it is here.
Posted to derby-user@db.apache.org by Kathey Marsden <km...@sbcglobal.net> on 2006/09/12 19:09:34 UTC
User community role 10.2 testing of optimizer changes
In the licencing discussion, I mentioned that we really need more user
feedback before we release 10.2.
Discussion has occurred on the developer list on that point and is on
this thread.
http://www.nabble.com/10.2-plans-%28was-Re%3A-10.2-licensing-issue%29-tf2256208.html
The summary is that right now the development community is not in a
position to work on any known regression and we should release to
motivate users to try 10.2. Kathy Saunders probably summed it up the
concern best when she said:
"If we really believe we need more testing, then what is that testing
and who is going to do it? "
My opinion is that our overall confidence in the quality and
robustness of Derby needs to increase release to release and we should
feel confident about that and should communicate risk in areas that are
likely impacted in terms that users can understand so you can either
adjust your expectations or do your part to bring your expectations
back to their prior level. Whether you count on Derby to control your
I.V drip, use it in the critical path of your business, or just use it
to manage your CD collection, you have the right to know the information
and can then assess whether you want to take your necessary role of
flushing out optimizer issues, before or after the release. The
summary is this:
There were significant optimizer performance changes in 10.2 that are
taking queries that were running for hours to seconds. (Maybe someone
can point to the data). These changes involve the optimizer and it is
really not possible to have comprehensive regression tests in this
area. The changes in my opinion are worth while, high quality, risky
and regression prone. We have had optimizer feedback from a single
user who exposed several issues. Other issues have been exposed by
development. We have fixed what we can. We need users especially
those who have complex or performance sensitive queries to try their
existing applications and give us feedback. Army can you please give an
overview of the changes from a functional perspective and explain what
types of usage you think could pop issues? Also I would like your
opinion on the value of such testing from the user community?
Users, please register your results at:
http://wiki.apache.org/db-derby/TenTwoApplicationTesting
Kathey
Re: User community role 10.2 testing of optimizer changes
Posted by Kathey Marsden <km...@sbcglobal.net>.
Kathey Marsden wrote:
> We need users especially those who have complex or performance
> sensitive queries to try their existing applications and give us
> feedback.
Yikes, I forgot the beta link.
http://people.apache.org/~rhillegas/10.2.1.3-beta/
Other 10.2 information
http://wiki.apache.org/db-derby/TenTwoRelease
http://wiki.apache.org/db-derby/TenTwoSnapshot
Register your application testing results and the type of derby usage
you work think they work out at
http://wiki.apache.org/db-derby/TenTwoApplicationTesting
File issues and list them for the contest at
http://wiki.apache.org/db-derby/RegressionSearchAndDestroy
If we can get testing of twenty query intensive applications I would
feel better about a release. I see the issues that have come in
recently
as a indicator that we need a larger data set to understand the
likelihood that users will hit regressions when they upgrade. Please help.
Thanks
Kathey
Re: User community role 10.2 testing of optimizer changes
Posted by "Jean T. Anderson" <jt...@bristowhill.com>.
Kathey Marsden wrote:
> In the licencing discussion, I mentioned that we really need more user
> feedback before we release 10.2.
...
> Users, please register your results at:
> http://wiki.apache.org/db-derby/TenTwoApplicationTesting
or, alternatively, how about posting feedback to derby-user?
Anyone can create a Wiki profile at http://wiki.apache.org/db-derby .
But I can appreciate that those who don't already know MoinMoin wiki
syntax (or similar syntax from other wiki software) might decide to add
feedback "later" when they can find time to spin up on it. And we all
know how soon "later" sometimes happens. :-)
-jean
Re: User community role 10.2 testing of optimizer changes
Posted by Kathey Marsden <km...@sbcglobal.net>.
Army wrote:
>
> Users may also want to remember the policy of "scratch your own itch"
> or "fry your own fish": I as a developer tend to have more time and
> inclination to address issues with contributed code closer to the time
> I actually made the contribution.
This is a good point and I bet your immediate understanding and ability
to turn around fixes is better when you are actively working on it.
It is really critical that users test the beta now and not later but
already it is very late. An even earlier partnership between the user
community and development would have greatly reduced the regression risk
for this change. Folks who wanted this improvement could have signed
up to test along the way and we could have gotten much more input. We
treat our user/developer relationship too much like a fine dining
establishment. Developers deliver a dish cooked, users take a bite
and see if they like it and then send it back if they don't. We need
our users in the kitchen tasting as we go along. Army is right. If the
dish is free and the cook was cooking for fun, it is hard to get a
rebake. Please test now.
http://people.apache.org/~rhillegas/10.2.1.3-beta/
For other popular performance and feature improvements like DERBY-47,
I hope we can use a different model. A development/user partnership
where users get the high quality improvements they want by providing
development the testing support needed along the way and development
communicates where user help is needed and most valuable and where they
are at risk. We break down the wall between the kitchen and the dining
room and work together to product "consistently high quality software."
For now we need users urgent help to try 10.2 beta. It's 5:55 and
dinner is scheduled at 6:00. We have a feeling something is wrong but
don't know what it is or what to do about it. Please come take a taste
before we put it on the table.
Thanks
Kathey
Re: User community role 10.2 testing of optimizer changes
Posted by Army <qo...@gmail.com>.
Kathey Marsden wrote:
> We need users especially those who have complex or performance
> sensitive queries to try their existing applications and give us
> feedback. Army can you please give an overview of the changes
> from a functional perspective and explain what types of usage
> you think could pop issues?
For "an overview of the changes from a functional perspective", users should see
the problem descriptions attached to the relevant Jira issues:
https://issues.apache.org/jira/browse/DERBY-805
--> DERBY-805_v5.html, sections I and II
https://issues.apache.org/jira/browse/DERBY-781
--> DERBY-781_v1.html, sections I and II
As for the "types of usage", there are generally three areas that are most
directly affected by the optimizer changes:
1. Queries with UNIONs in them (the more unions, the more likely the query is to
be affected).
2. Queries with subqueries in them, either explicitly or indirectly through
views. The more deeply nested subqueries there are, the more likely the query
is to be affected.
3. Queries which perform joins in which at least one of the result sets to be
joined is a UNION, a subquery, or a combination of the two.
Thus far all optimizer regressions have been discovered by queries posted by a
single user whose app uses large, deeply-nested queries that involve all three
of the above areas. See in particular DERBY-1205, DERBY-1633, DERBY-1777.
That said, it's possible that other queries will also be affected since
additional optimizer-related bug fixes have been made--esp. DERBY-1007 and
DERBY-1357. So as Kathey said, anyone with query-intensive applications might
find it beneficial to do some testing with 10.2.
> Also I would like your opinion on the value of such testing
> from the user community?
Very valuable, no doubt there. If users don't test it beforehand, they run the
risk of finding problems the hard way. I know it can take a lot of time and
effort to test a beta candidate, so it's not too surprising that we've had
little response to Derby's multiple requests for more user testing. I guess
it's up to the user to decide when and how the time and effort is going to be
spent: early on in beta, or later when regressions are (presumably) more critical.
Users may also want to remember the policy of "scratch your own itch" or "fry
your own fish": I as a developer tend to have more time and inclination to
address issues with contributed code closer to the time I actually made the
contribution. If the code is untested for several months and then a user hits a
regression at release time, what am I going to be doing at that time? And am I
going to have the time and means to resolve the problem right then? Will that
regression be my "itch" or my "fish"? I would of course hope the answer is an
immediate "Yes"--but there's no guarantee in the world of opensource, and a lot
can happen in a couple of months.
So yes, more testing is better. Thanks to Kathey for continuing to push for
more user feedback. It can only help.
Army