[Postgres-xl-general] Getting "Failed to get pooled connections" when 1 datanode is down

Rick Mtek rickmtek at gmail.com
Wed Feb 11 10:35:46 PST 2015


Pavan,

The SELECT [SKIP_UNREACHABLE] is something that could benefit us.

In the meantime, How do we know which are the reachable nodes? Is there any
catalog table from where we could get which ones are the
available/reachable nodes so that we can construct a query to hit only
those nodes skipping the unreachable ones?

I appreciate your feedback,
Thanks.

On Tue, Feb 10, 2015 at 11:34 PM, Pavan Deolasee <pavan.deolasee at gmail.com>
wrote:

>
>
> On Tue, Feb 10, 2015 at 11:05 PM, Rick Mtek <rickmtek at gmail.com> wrote:
>
>>
>>
>>>
>> Pavan, if I run the query against the other datanodes it works as you
>> mentioned. No doubts about it. That works fine, but the application does
>> not know anything about datanodes. For the application this is a just one
>> big database (behind scenes threre are several machines with GTM,
>> Coordinators and Datanodes).
>>
>> Still these questions are unanswered:
>>
>> 1) if by any chance one datanode is down or not accessible the whole system/cluster data will be not accessible by the UI application? (ERROR:  Failed to get pooled connections)
>>
>>
> The data hosted by that datanode will remain inaccessible as long as SQL
> queries do not require access to that data.
>
>> 2) How to keep getting results from all the others datanodes even if one is not accessible from the UI? (querying manually each datanode?????)
>>
>> Not really. I guess the application is expected to query only the data it
> needs. If it requires everything and if some datanodes are not reachable,
> what else will be acceptable? If we skip some datanodes, then the query
> result could be incorrect and that can lead to all kinds of consistency
> issues. We do recommend users to set up streaming replica of datanodes and
> failover to them for HA. There was a suggestion to have a configuration
> parameter so as the cluster can run in degraded mode and return data from
> the accessible nodes, without queries failing. Or we could have SQL
> extension, something like SELECT [SKIP_UNREACHABLE], which would return
> data from available nodes instead of reporting error. But to do something
> like this implicitly is something asking for bigger trouble.
>
>
>> 3) What should be the expected results when query a distributed table having one datanode down? (An Error?)
>>
>> Yes, that's what indeed happens today.
>
> Thanks,
> Pavan
>
> --
> Pavan Deolasee
> http://www.linkedin.com/in/pavandeolasee
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.postgres-xl.org/pipermail/postgres-xl-general-postgres-xl.org/attachments/20150211/3aaa56e1/attachment.htm>


More information about the postgres-xl-general mailing list