Re: Creating a function for exposing memory usage of backend process

Kasahara Tatsuhito <kasahara.tatsuhito@gmail.com>

From: Kasahara Tatsuhito <kasahara.tatsuhito@gmail.com>
To: Fujii Masao <masao.fujii@oss.nttdata.com>
Cc: torikoshia <torikoshia@oss.nttdata.com>, PostgreSQL-development <pgsql-hackers@postgresql.org>, robertmhaas@gmail.com
Date: 2020-06-26T05:53:00Z
Lists: pgsql-hackers
Hi !

On Thu, Jun 18, 2020 at 12:56 PM Fujii Masao
<masao.fujii@oss.nttdata.com> wrote:
> Agreed. The feature to view how local memory contexts are used in
> each process is very useful!
+1

> >    =# SELECT * FROM pg_stat_get_backend_memory_context(PID);
>
> I'm afraid that this interface is not convenient when we want to monitor
> the usages of local memory contexts for all the processes. For example,
> I'd like to monitor how much memory is totally used to store prepared
> statements information. For that purpose, I wonder if it's more convenient
> to provide the view displaying the memory context usages for
> all the processes.
How about separating a function that examines memory consumption
trends for all processes and a function that examines memory
consumption for a particular phase of a particular process?

For the former, as Fujii said, the function shows the output limited
information for each context type. All processes calculation and
output the information at idle status.

I think the latter is useful for debugging and other purposes.
For example, imagine preparing a function for registration like the following.
=# SELECT pg_stat_get_backend_memory_context_regist (pid, context,
max_children, calc_point)

pid: A target process
context: The top level of the context of interest
max_children: Maximum number of output for the target context's children
 (similar to MemoryContextStatsInternal()'s max_children)
calc_point: Single or multiple position(s) to calculate and output
context information
 (Existing hooks such as planner_hook, executor_start, etc.. could be used. )

This function informs the target PID to output the information of the
specified context at the specified calc_point.
When the target PID process reaches the calc_point, it calculates and
output the context information one time to a file or DSM.

(Currently PostgreSQL has no formal ways of externally modifying the
parameters of a particular process, so it may need to be
implemented...)

Sometimes I want to know the memory usage in the planning phase or
others with a query_string and/or plan_tree that before target process
move to the idle status.
So it would be nice to retrieve memory usage at some arbitrary point in time !

Regards,

-- 
Tatsuhito Kasahara
kasahara.tatsuhito _at_ gmail.com



Commits

  1. Add regression test for pg_backend_memory_contexts.

  2. Move codes for pg_backend_memory_contexts from mmgr/mcxt.c to adt/mcxtfuncs.c.

  3. Prevent non-superusers from reading pg_backend_memory_contexts, by default.

  4. Add pg_backend_memory_contexts system view.

  5. Fix yet another issue with step generation in partition pruning.