Ceph is collecting all kinds of metrics internally that are exposed
through the admin socket or via the 'ceph perf dump' command. There's no
convenient way to watch these values from the command line, though.
A while back I wrote script/perf-watch.py that tries to provide something
that spits out a line every second (ala vmstat, iostat, etc) of whatever
metrics you specify. It's a bit kludgey but look something like this:
$ script/perf-watch.py -s out/osd.0.asok filestore.bytes osd.wr filestore.commitcycle osd.op osd.op_rw
# filestore.bytes filestore.commitcycle osd.op osd.op_rw
0 0 0 0
0 0 0 0
0 0 0 0
29375672 0 1 0
37768744 0 6 0
33572224 0 4 0
0 0 0 0
46161808 1 3 0
0 0 0 0
0 0 0 0
# filestore.bytes filestore.commitcycle osd.op osd.op_rw
0 0 0 0
0 0 0 0
25179152 1 2 0
37768736 0 6 0
0 1 0 0
0 0 0 0
...
You can specify either individual metrics ('osd.op') or a full category
(just 'osd'). The problem is usually that there are so many metrics that
doing the full set requires a *really* wide monitor to be useful.
Anyway, two thoughts:
1) We could incorporate this into the normal 'ceph' cli tool. Maybe a
'ceph perf <daemon or asok> <metrics ...>', which works similar to
'ceph daemon ...'.
2) We could mark certain metrics as the 'interesting' ones and make it
default to showing those when no others are specified. And/or possibly do
the same for each group (osd, filestore, etc.). That would steer
admins towards the ones that are actually helpful in telling what the
cluster is doing.
Thoughts?
sage
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
