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

Tobias Oberstein tobias.oberstein at gmail.com
Wed Sep 30 12:16:06 PDT 2015


Am 30.09.2015 um 19:06 schrieb Krzysztof Nienartowicz:
> I would give it a spin, it all depends on your workload. Be also sure

Ok. I will try running with a GTM proxy tomorrow.

> all the (data) nodes recovered before you start any new operation..
> We test on 4 nodes x 20 HTcores heavy bulk ingestion and transformations
> with tens of TBs and billions of records in partitioned tables, with
> 64-300 cores accessing the DB cluster to stream in, update some small
> tables etc..

Interesting. Thanks for describing your setup and workload!

Just to be sure I understand: you run 4 PG-XL nodes on a single 4 CPU 
socket machine or on 4 machines, each with 1 PG-XL node?

Right now, we have a single machine with 4 sockets á 12 non-HT cores. We 
might get a 2nd machine with 4 sockets and 64 cores plus 100GbE 
interconnect soon. And possibly more and more of these boxes in the 
future. So we are looking for an "upgrade path" -- PG-XL is hot.

> The main problem XL has is the stability _after_ the configuration or
> even application errors. Sometimes it is possible to recover by easy
> cleanup, sometimes restart helps, sometimes scratch of the cluster is
> needed. The robustness should be the main development focus, to get
> nearer to what Postgres got us accustomed to over last years. Until it
> happens, you might face such errors til your setup gets into an equilibrium.

I see. This is making me slightly nervous .. "stability". We are 
evaluating this for DWH production use for a customer in the finance 
sector. We can't afford "sorry, we need to scratch the DWH" situations. 
There is a certain amount of breakage we can accept, but I am afraid to 
run into "unresolvable situations" ..

/Tobias

>
>
>
>
> On Wed, Sep 30, 2015 at 6:30 PM, Tobias Oberstein
> <tobias.oberstein at gmail.com <mailto:tobias.oberstein at gmail.com>> wrote:
>
>     WARNING: can not connect to GTM: connection has been reset by peer
>     ERROR: GTM generated global XID not available
>
>     What .. why?
>
>     Could it be, that running a GTM proxy helps?
>
>





More information about the postgres-xl-general mailing list