[Postgres-xl-general] "ERROR: Failed to get pooled connections" with cross-node join

Pavan Deolasee pavan.deolasee at gmail.com
Wed Sep 30 21:06:49 PDT 2015


On Wed, Sep 30, 2015 at 9:22 PM, Tobias Oberstein <
tobias.oberstein at gmail.com> wrote:

> Krzysztof,
>
> Am 30.09.2015 um 15:21 schrieb Krzysztof Nienartowicz:
> > Hello,
> > What is your max_pool_size?
>
> No, I didn't had set a value.
>
> Now I have set
>
> max_datanodes = 64
> max_connections = 2048
> max_pool_size = 2048
>
> for the coordinator, and
>
> max_datanodes = 64
> max_connections = 2048
>
> on the datanodes.
>
>
Try increasing max_pool_size on the datanodes too.


> I rebuilt the whole cluster and recreated the test dataset from scratch.
>
> The cross-join query fails with the exact same reason.
>
> This time, a second start of the query did run "forever". When canceling
> the query, one of the PG processes _crashed_ with
>
> PANIC: ERRORDATA_STACK_SIZE exceeded
>
>
I've this in my issue tracker. I've seen it happening in exactly similar
situations that you described above. Sorry, haven't had got chance to dig
deeper and fix this issue.


> As the other PG backends noticed the process crash, the whole cluster
> went into recovery.
>
> I restarted the cluster, and reran the query: same problem.
>
> Note that the log leaves:
>
> WARNING: can not connect to GTM: connection has been reset by peer
> ERROR: GTM generated global XID not available
>
> Could it be that the GTM is overwhelmed by connections?
>
> Is there any GTM tunable related?
>

There are some compile time #defines for worker threads etc, but other than
that there is no tunable for GTM.


>
> ===
>
> I wonder if anyone has run PG-XL with large node counts and driven by
> more than toy workloads .. anyone?
>
>
I haven't heard of cases when someone tried or using it with more than 16
nodes. But recently we are getting aware of cases where people are using it
in production, but those cases don't get reported here, especially if
things work out fine.

We (developers) are also limited by the hardware that is accessible to us.
So doing any tests which involves significant hardware is practically
impossible for us. Any help in that area is highly appreciated. I know
Krzysztof is doing many tests with real world use cases and reporting
issues. Thats really helpful and we are committed to keep improving
stability of the product, more than adding new features at this stage.

Thanks,
Pavan

-- 
 Pavan Deolasee                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.postgres-xl.org/pipermail/postgres-xl-general-postgres-xl.org/attachments/20151001/38418563/attachment.htm>


More information about the postgres-xl-general mailing list