[postgres-xl-bugs] GTM fails from time to time
tomas.vondra at 2ndquadrant.com
Mon Nov 20 05:50:43 PST 2017
On 11/15/2017 10:43 AM, Pilar de Teodoro wrote:
> We are using postgres-xl 10 alpha2 and postgres-xl 9.5 in 2 different
> clusters but in both happens the same:
> We have found out that GTM fails without leaving any message although we
> have seen in Ganglia the node where GTM is installed (VM) reached a peak
> in memory. We have increased the node memory to 64GB from 32GB but it is
> still happening. Nothing was running at that time in postgres-XL.
> We have setup a GTM slave but the promote it is not automatic. How can
> we understand what is happening to the GTM and why?
I suppose this is caused by the memory leak we've fixed about a month
ago (and so it's not included in the latest releases). The GTM should
not really need more than 1GB of memory, but we've been leaking memory
on disconnect from GTM (e.g. if you have many short-lived database
connections, that may be the cause).
This should be fixed by b114973b60 , although Krzysztof reported some
new issues with that commit (but we've been unable to reproduce that).
There are two simple "workarounds" you may try:
1) deploying GTM proxy
This multiplexes connections to GTM, and so keeps them open longer. That
significantly reduces the amount of leaked memory.
2) using connection pool
Again, this keeps the GTM connections open (because the backends are
kept around for reuse), and so reduces the amount of memory.
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
More information about the postgres-xl-bugs