Hi Abhisek,
On 8/12/24 07:37, Sahu, Abhisek Kumar wrote:
> Hi Daniel,
>
> Kindly let us know if you have any updates.
>
It's hard to say what happened, based on the information provided so
far. It might be that there really is a bug / memory leak, or maybe the
selected query plan is not great, or maybe there's a configuration
issue. We'll need more information to find out.
1) What query was the failed backend running? Was it a big long-running
query (e.g. in ETL/BI application), or many small OLTP queries?
2) Do you know what execution plan was used?
3) If you rerun the query, does the memory usage grow the same way? This
would allow us to collect important data. For example, it would be very
useful if you could attach a debugger to a backend consuming a lot of
memory, and get memory context stats.
4) Can you share the configuration parameters? I think the memory config
is the most interesting one (shared_buffers, work_mem, ...). How much
RAM does the system have?
5) Is this a parallel query? In the snippet from /var/log/messages there
are 3-5 postgres processes that look like they might be parallel workers
(nearly sequential PIDs, same memory pattern).
6) Is this a long-running problem, or did it just start to happen
recently? Perhaps after some change/upgrade?
regards
--
Tomas Vondra
