Hi,
I want to instrument database connections with app
module/action/username/transaction id/etc.
The main use case is for troubleshooting purposes:
1) I want some kind of report from the database (e.g. select * from
pg_stat_activitiy) to show not just SQL texts, but application
modules/components as well
2) In case of connection leak, it would be nice to know where the
connection was last used
application_name is limited with 64 characters (or bytes?), so it is
not sufficient to store the ids.
What is the recommended way of solving that?
I'm afraid, I cannot ask clients recompiling postgresql for changing namelen.
Can the length of application_name be configurable via regular
configuration parameter?
Is there a way to read guc from another backend?
Can we have more fields like "application_name"?
For instance, there is a paper on distributed tracing (see [1]) that
suggests having a couple of 64-bit ids to identify the call (span id
and parent span id).
I know there are security/concurrency implications, so I wonder if
there are hard limits.
In the mean time, my workaround would be unlogged table like
app_connection_tracker(pid, app_user, module, action, ...) with index
on pid.
Any pitfalls with that kind of "update mostly table"?
[1]: http://research.google.com/pubs/pub36356.html
--
Regards,
Vladimir Sitnikov
