[Pgpool-hackers] Master-Save Load-Balancing: Bind / Execute can be sent to different backends - causes failures.

[Bryan Varner]

I've put together a simple java application test-case (attached) to 
re-create the issue as reliably as I can.

I'm also attaching a .diff against CVS HEAD that seems to fix my issue.

In the Parse, Bind, Execute chain: Bind, Describe, and Execute could be 
directed at servers that did not receive the preceding required messages.

My fix for this (so far) is to simply not re-apply the logic of where to 
send the query in subsequent steps, but to trust that the 
pool_where_to_send during the PARSE was correct.

I will be testing this change extensively over the next few weeks. I'd 
love to hear any other issues or thoughts from anyone else on this, and 
I'd really like to get this pushed into the repository once we're sure 
we have a proper fix.

Regards,
  - Bryan Varner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgpool-trust-parse-location.diff
Type: text/x-patch
Size: 1211 bytes
Desc: pgpool-trust-parse-location.diff
URL: <http://pgfoundry.org/pipermail/pgpool-hackers/attachments/20111025/59a0a143/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgpool-buster.tar.gz
Type: application/x-gzip
Size: 2061 bytes
Desc: pgpool-buster.tar.gz
URL: <http://pgfoundry.org/pipermail/pgpool-hackers/attachments/20111025/59a0a143/attachment-0001.bin>