25 Sep
2012
25 Sep
'12
11:24 a.m.
See below.
On Sep 25, 2012, at 10:48 AM, Denis Gervalle <dgl(a)softec.lu> wrote:
On Tue, Sep 25, 2012 at 1:29 AM, Andreas Jonsson
<aj(a)member.fsf.org> wrote:
I understand. You should also consider that Apache DBCP is well known for:
* being slow
* being full of bugs
* having a stalled develoment
At least that's what I keep seeing mentioned in all my google everywhere.
I think you shouldn't think Apache == Good. At least not for Commons. Commons libs can
be very small and maintained by very few people.
If you check http://commons.apache.org/dbcp/team-list.html you'll see several
committers.
But if you check
http://svnsearch.org/svnsearch/repos/ASF/search?view=plot&path=%2Fcommo…
you'll see that only 2 committers have been active since 2009.
You should also note that Tomcat 7 has implemented its own connection pooling thus
dropping DBCP even though they're both in apache land… you should ask why. I don't
know the internal story but here's what they say:
http://stackoverflow.com/questions/520585/connection-pooling-options-with-j…
Thanks
-Vincent
PS: I don't disagree with your analysis but we shouldn't be too fearful either of
making discoveries/changes. Provided we can control them and revert easily it's not a
bad thing to discover new stuff.
PPS: BoneCP could not be the best choice. Maybe C3P0 is better now or maybe the new Tomcat
Connection Pool is better.
PPPS: In any case I'm not going to make any change till I get to the bottom of this
issue with HSQLDB in virtual mode….
2012-09-24 17:26, Vincent Massol skrev:
You should, we really do scary stuff to support multiple DB, and also
dynamic mapping.
Hi devs,
I was trying to make XEM work with HSQLDB and... I have succeeded :)
The only problem I have is that DBCP doesn't seem to work with
Hibernate. For
some weird reasons when we do a SET SCHEMA it makes calls to
fail afterwards. In any case if I configure Hibernate to not use DBCP all
work just fine.
Funny coincidence! I was just trying to get it to work with
PostgreSQL. It mostly works, as already commented on the Jira issue.
The only thing that seems to work unreliably is the initial template
creation. It seems likely that it is the same issue.
But having looked into this I'm a bit worried about the current XEM
implementation. I'm not an expert on Hibernate, but to me it
seems that
entities are expected to be bound to specific tables at initialization
time. So I'm wondering, does Hibernate really support changing the
mappings by switching databases or schemas underneath?
The tricks is that we do not really support multiple hibernate mapping. All
accessed DBs use the same mapping. These are serious limitations in
multi-wiki mode for mapping related features.
Hibernate does not even knows that we switch the DB under its feet. Even
the pool provider is not aware. This is why we need to set the correct DB
each time we open a session/transaction (always tight together), which
means getting a connection from the pool. We do not support nested
session/transaction, in particular to different DBs for the same reason.
What prevents it
from generating prepared queries for accessing the entities, which will
be resolved against the initial database/schema, and continue to use
these even after the database or schema changes?
Our config prevent using prepared statements on the DB server.
+1 for your action plan.
Best Regards,
/Andreas
Unless BoneCP really provide proven improvements, and not only a potential
performance increase, that will probably not be tremendous, I am not really
in favor of changing a well-know Apache project for a project that as not
really a great team of supporter.
I've googled around and found that there are lots of people complaining
about
DBCP.
I've googled for what connection pooling library to use and I've found
that most people are recommending BoneCP (http://jolbox.com/).
See:
*
http://stackoverflow.com/questions/5640146/java-jdbc-connection-pool-librar…
http://stackoverflow.com/questions/8057110/java-database-connection-pool-bo…
The other thing is that we currently have some 300 line of code that we
shouldn't have at all and that we need in XWiki just for handling DBCP (see
com.xpn.xwiki.store.DBCPConnectionProvider).
The pro of BoneCP is that it seems to be the fastest, see
http://jolbox.com/benchmarks.html
So here's what I'd like to propose:
* We move DBCPConnectionProvider to a legacy module
* We bundle the bonecp jar by default
* We configure our hibernate.cfg.xml by default to use boneCP:
<property name="bonecp.idleMaxAge">240</property>
<property name="bonecp.idleConnectionTestPeriod">60</property>
<property name="bonecp.partitionCount">3</property>
<property name="bonecp.acquireIncrement">10</property>
<property
name="bonecp.maxConnectionsPerPartition">60</property>
<property
name="bonecp.minConnectionsPerPartition">20</property>
<property name="bonecp.statementsCacheSize">50</property>
<property name="bonecp.releaseHelperThreads">3</property>
<property
name="connection.provider_class">com.jolbox.bonecp.provider.BoneCPConnectionProvider</property>
What this means:
* Existing users of XWiki will not have to change anything, it'll still
work
* New users will use bonecp without knowing
* Existing users can migrate to bonecp just by changing one line in
their
hibernate.cfg.xml file
I'd love to do this for 4.3 but we're already quite advanced in the 4.x
cycle. That said it's just a one line change in hibernate.cfg.xml in case
of problem to go back to DBCP so we could do this for 4.3 which leaves us
4.3, 4.4 and 4.5 to test this out. And if later on we find an issue we'll
always be able to release a 4.5.1 that has this one line change to go back
to DBCP.
WDYT?
Here's my +1 for this action plan.
From the project info page, I see only
one name. From the forum, there seems to be serious rumors that the project
is no more supported by that guy since more than a year.
So, currently, I am -1.
>>
>> Thanks
>> -Vincent