[Bugs] [Bug 1457936] possible memory leak in glusterfsd with multiplexing

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1457936

Raghavendra G <rgowdapp at redhat.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |needinfo?(kramdoss at redhat.c
                   |                            |om)



--- Comment #12 from Raghavendra G <rgowdapp at redhat.com> ---
>From https://bugzilla.redhat.com/show_bug.cgi?id=1461543#c35

> Do comment 32 and 33 make us to think about why we are seeing a spike in the
> memory consumption in the brick mix scale setup, Raghavendra G?

I'll answer based on two observations by Sowmya

> 1) With per-thread mem-pools now, as the process threads scales, the mem-pools > allocated also scales at the rate of power of 2. This similar issue may had  been observed in case of CNS too (as the volume count is increased, there is drastic increase in VSS usage of the brick process)

Possible. There are around 10-12k threads in a single process which has 1k
bricks multiplexed. Need to check with shberry whether the tests involved any
I/O. If no I/O is involved mem pools may not necessarily be allocated as they
seem to be allocated during the first mem_get on an object provided by the
pool.

> 2) The un-used memory should be cleaned up by mem-pool sweeper which may not be happening. This needs to be looked into.

This issue is not present in brick-mux setup as we use a glusterfsd binary and
glusterfsd does indeed create a pool sweeper thread while it it is starting up.
This issue is present only for gfapi setups.

@shekhar, @karthick, was it just mount or any I/O done during your tests?

-- 
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=7W0EUmlMjK&a=cc_unsubscribe