librbd: flush and invalidate cache via admin socket #6453
Conversation
|
@dillaman thank you for review. Rebased after addressing your comments. |
| @@ -41,6 +41,9 @@ enum { | |||
| l_librbd_readahead, | |||
| l_librbd_readahead_bytes, | |||
|
|
|||
| l_librbd_flush_cache, | |||
| l_librbd_invalidate_cache, | |||
There was a problem hiding this comment.
these two counter seemed a little different with other counters. Maybe another name?
There was a problem hiding this comment.
The l_librbd_flush_cache can just use the existing l_librbd_flush perf counter. As for l_librbd_invalidate_cache, to me it is in-line with the other function-level perf counters such as l_librbd_snap_xyz, l_librbd_resize, etc.
|
@dillaman Thanks. Rebased after applying your suggestions. |
|
lgtm |
| local counter=$2 | ||
|
|
||
| ceph --admin-daemon $(rbd_watch_asok ${image}) perf dump | | ||
| jq ".\"librbd--rbd/${image}\".${counter}" |
There was a problem hiding this comment.
I had to change this to the following to work under Fedora 22: jq ".[\"librbd--rbd/${image}\"].${counter}". Also, the jq application is not necessarily installed. An alternative that doesn't require a new package dependency is something like:
ceph --admin-daemon $(rbd_watch_asok ${image}) perf dump librbd--rbd/${image} | grep ${counter} | awk '{print $NF}'
or use --format xml and the existing xmlstarlet dependency.
There was a problem hiding this comment.
Thanks, I will fix jq query.
The --format xml and xmlstarlet is the first thing I tried. Unfortunately it does not work due to "/" in "librbd--rbd/testimg23790" tag:
zhuzha:~/ceph/ceph/src% ceph --format xml --admin-daemon out/client.admin.13067.asok perf dump | xmlstarlet sel -t -m "."
-:1.37: error parsing attribute name
<perfcounter_collection><librbd--rbd/testimg23790><rd>0</rd><rd_bytes>0</rd_byte
^
(I think I have to create a PR for this).
There was a problem hiding this comment.
As for your comment that jq is not necessarily installed, we have in run-make-check.sh the instruction to install jq, before running the tests. That is why I decided it would be ok to use jq here, and not parse this e.g. with awk, which is fragile. I could also use a python oneliner with json module, which could be correct but heavier.
There was a problem hiding this comment.
@ldachary do you think it is ok to use jq for parsing json output in our tests or we'd rather avoid this?
There was a problem hiding this comment.
OK, if we are already using it, I have no issue.
|
Rebased after fixing the jq query in the test. |
|
lgtm |
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
Fixes: ceph#2468 Signed-off-by: Mykola Golub <mgolub@mirantis.com>
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
|
Rebased after updating the tests not to use |
librbd: flush and invalidate cache via admin socket Reviewed-by: Jason Dillaman <dillaman@redhat.com>
Fixes: #2468
Signed-off-by: Mykola Golub mgolub@mirantis.com