[Bugs] [Bug 1657743] Very high memory usage (25GB) on Gluster FUSE mountpoint

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1657743

ryan at magenta.tv changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|needinfo?(ryan at magenta.tv)  |



--- Comment #16 from ryan at magenta.tv ---
Thanks for the link, I believe I may have read this previously and decided to
set a lower ping based on it. I've increased it to 20s now.

I've set the logging level to warning and will monitor the system in use to see
if the issue re-occurs and prints anything of interest to the log.

Only thing I've found currently is the following error:
[2019-07-09 11:26:43.314018] E [MSGID: 101046]
[dht-common.c:1895:dht_revalidate_cbk] 3-mcv01-dht: dict is null

However I think i raised a bug for this before and that has been addressed as a
separate issue.

Unfortunately we're still unsure on how this issue is reproduced. The best I
can do at the moment is explain how the FUSE mountpoint is used:
Most of the users use the volume via the vfs_glusterfs module in Samba, however
we had many issues getting OS X machines to work with SMB, mainly due to bugs
between the Fruit, Streams_xattr and Glusterfs VFS modules. As a result, we run
Netatalk (Program to present the mount point over the AFP protocol) to improve
stability for OS X users. The users are mainly video editors accessing large
files. The workload access pattern tends to be 80% sequential access and around
80% read, with the remainder being random read/write IO for metadata and
sidecar files etc. The memory buildup occurs slowly overtime, which makes
testing for the bug significantly harder.

I can provide statedumps or anything else that may be able to assist.

I'll update by the end of today with an update on the logs.

Best,
Ryan

-- 
You are receiving this mail because:
You are on the CC list for the bug.