Cluster crashes if a query fails when using PostgreSQL and V2 protocol
----------------------------------------------------------------------
Key: SEQUOIA-881
URL: https://forge.continuent.org/jira/browse/SEQUOIA-881
Project: Sequoia
Type: Bug
Components: Core
Versions: Sequoia 2.10.3
Environment: Sequoia 2.10 CVS branch - PostgreSQL 8.1 - JDBC driver uses V2
protocol instead of V3
Reporter: Guillaume Smet
Hi all,
The application we are using (Red Hat WAF/CCM) forces us to use the version 2
of the protocol used by the PostgreSQL JDBC driver instead of the V3 one (due
to types problem which are not fixable).
When a query fails inside a transaction, the entire cluster goes down because
the backends are declared inactive by DatabaseBackend.isValidConnection().
In this method, there is a hack for PostgreSQL:
if ("25P02".equals(e.getSQLState())
|| (e.getMessage() != null && e
.getMessage()
.indexOf(
"current transaction is aborted, queries ignored until end of
transaction block") > 0))
{
// see bug item #300873 on the forge for details
// postgres throws an exception if a query is issued after a request has
// failed within a transaction, we now have to check for this exception
// as it is means the connection is valid
//
// postgres versions after 7.4 will return the SQLState, whereas
// postgres versions prior to 7.4 will have to be checked for the
// message text
This comment is misleading. It's not a problem with a particular version of
PostgreSQL but with the version of the protocol (V2 does not support SQLState
whereas V3 does). If you are using PostgreSQL 8.1 with V2 protocol
(protocolVersion=2 parameter in your JDBC URL), you won't have the valid
SQLState returned. Note that using the V2 protocol is the recommended
workaround when you cannot upgrade the application. Apart from that, V3
protocol can have performances problem on particular queries that V2 does not
have so it's really important to support it too.
BTW it shouldn't be a problem if the error message didn't change between 7.4
and 8.1. In 8.1 "queries" is replaced by "commands".
I attached a patch which:
- fixes the comment,
- removed the start of the sentence and only considers "ignored until end of
transaction block" - it's not used anywhere else in PostgreSQL source code and
is really tied to this problem.
This patch fixes the problem on our platform.
This is a really annoying problem for us as the entire cluster goes down as
soon as a query fails.
Regards,
--
Guillaume Smet
Open Wide
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://forge.continuent.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
