|
From: | Vijay Bellur |
Subject: | Re: [Gluster-devel] glusterfs-3.5.0beta4 released |
Date: | Sun, 09 Mar 2014 21:15:31 +0530 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.3.0 |
On 03/09/2014 04:17 PM, Emmanuel Dreyfus wrote:
Vijay Bellur <address@hidden> wrote:Do not observe any fixes related to memory leaks. Would it be possible to get a valgrind report for the glusterfsd processes if you happen to observe the leak again?Unfortunately, valgrind was not ported on NetBSD.
In that case, we can look at utilizing the built in memory accounting provided by glusterfs. Steps for that would include:
1. gluster volume set <volname> config.memory-accounting on 2. (Re)start the volume3. After the memory consumption increases, trigger a statedump by sending SIGUSR1 to the brick process.
4. statedump file would be created at $(INSTALL_PREFIX)/var/run/gluster/ 5. grep -A4 memusage <statedump_file>6. Noticing those allocations which have significant size & num_allocs in the output of grep can help in narrowing down the leak.
(usage-type can be looked up from mem-types.h in libglusterfs and <xlator>-mem-types.h located in the respective translator directory).
- Vijay
[Prev in Thread] | Current Thread | [Next in Thread] |