mds: Kill C_SaferCond in evict_sessions() #9971
Conversation
|
NB: make check failure is bluestore unit tests. |
| @@ -587,6 +587,33 @@ void MDSDaemon::tick() | |||
| } | |||
| } | |||
|
|
|||
| void MDSDaemon::send_command_reply(MCommand *m, MDSRank *mds_rank, | |||
| int r, bufferlist outbl, | |||
| std::string outs) | |||
There was a problem hiding this comment.
The string should probably be a const reference?
|
Looks good to me; just a few nits. Since we test commands in the nightlies this should have plenty of coverage, but you did test it locally, right? |
|
Comments addressed, should I run this through a suite or something? |
|
There's probably something available under vstart-runner that tests the commands. If that passes we can mark it as needs-qa and let @jcsp or somebody bundle it up in a testing branch. |
|
I am seeing a hang inside evict sessions when running TestVolumeClient: @fullerdj can you reproduce locally with vstart_runner? |
|
Yes, I've been working on that.
|
fullerdj
force-pushed
the
wip-djf-16288
branch
2 times, most recently
from
71eda15
to
1da3d6a
Compare
|
This resolves the hang I was seeing, but it looks like you may have encountered a different case. It looks like you are seeing a nearly identical case to http://tracker.ceph.com/issues/16288, but I have never been able to reproduce that one. |
MDSRankDispatcher::evict_sessions waits on a C_SaferCond for kill_session to complete on each of its victims. Change the command handling flow to pass command messages all the way down to MDSRankDispatcher. Extract the MDSDaemon's reply path into a static function callable from a new context in the MDSRankDispatcher. See: http://tracker.ceph.com/issues/16288 Signed-off-by: Douglas Fuller <dfuller@redhat.com>
MDSRankDispatcher::evict_sessions waits on a C_SaferCond for
kill_session to complete on each of its victims. Change the
command handling flow to pass command messages all the way down
to MDSRankDispatcher. Extract the MDSDaemon's reply path into a
static function callable from a new context in the MDSRankDispatcher.
See: http://tracker.ceph.com/issues/16288
Signed-off-by: Douglas Fuller dfuller@redhat.com