On 11-11-23 09:27 AM, Simon Riggs wrote:
> On Wed, Nov 23, 2011 at 1:51 PM, Steve Singer<ssinger at ca.afilias.info>  wrote:
>> On 11-11-23 04:28 AM, Simon Riggs wrote:
>>>
>>> On Tue, Nov 22, 2011 at 10:26 PM, Steve Singer<ssinger at ca.afilias.info>
>>>   wrote:
>>>>
>>>> On 11-11-20 05:47 AM, Cédric Villemain wrote:
>>>>>
>>
>>>
>>> ISTM that setting the remote worker to REPEATABLE READ would work well
>>> for this case. Patch attached.
>>>
>>
>> Simon, Did you forget to attach the patch? I don't see it.
>
> Looks that way.
>

Thanks for the patch.

So why would a SERIALIZABLE READ ONLY DEFERRED transaction produce fewer 
conflicts than a READ COMMITTED transaction?  Currently the 
remote_listener gets the default isolation level (READ COMMITTED).

The attached patch combines your two patches plus performs the same 
change to other places in remote_worker (there are places in 
remote_worker.c where that initial transaction is rolledback and 
restarted, this version also makes sure that those transactions are 
started as READ COMMITTED).   It also makes the connections that the 
remote helpers do to the remote database for querying sl_log_x READ ONLY 
DEFERRED.

When I run this patch through the test suite on 9.1 I don't see any 
serialization pivot failures (at least not yet).

-------------- next part --------------
A non-text attachment was scrubbed...
Name: read_only_patch.v3.diff
Type: text/x-patch
Size: 6342 bytes
Desc: not available
Url : http://lists.slony.info/pipermail/slony1-general/attachments/20111123/d6eb12cc/attachment.bin