On Mon, Feb 24, 2020 at 2:28 PM Uday Bhaskar jalagam
<jalagam.ceph(a)gmail.com> wrote:
Thanks Patrick,
is this the bug you are referring to
https://tracker.ceph.com/issues/42515 ?
Yes
We also see performance issues mainly on metadata
operations like finding file stats operations , however mds perf dump shows no sign of any
latencies . could this bug cause any performance issues ?
Unlikely.
do you see any clue in this that could cause slow down
in such operations ? our metadara pool has around 1.7 GB of data I gave mds cache 3 GB,
3GB cache is probably too small for your cluster. How many users? Your
perf dump indicates it probably is 8GB and not 3GB.
I am not sure where to check how much used in the 3 GB
or what is hit and miss count/ration in cache .
"mds_co_bytes": 8160499164,
in your perf dump. You can also look at inodes added/removed (to
identify churn):
"mds_mem": {
"ino": 2740340,
"ino+": 19461742,
"ino-": 16721402,
--
Patrick Donnelly, Ph.D.
He / Him / His
Senior Software Engineer
Red Hat Sunnyvale, CA
GPG: 19F28A586F808C2402351B93C3301A3E258DD79D