Ceph » CephFS

I can't find the referenced ticket anywhere. Anybody know what this is supposed to be and if it still applies? (I think it's actually about the kclient, but not sure.)

I think sage means clean dirty pages at the same time of cleaning dirty metadata. This haven't been implemented yet

If I have correctly comprehended it and from the current code and my test we had already implemented it.

There has 2 cases where the mds sessions are closing:

Case 1) unmount:

In this case, the VFS will help us do the dirty pages cleaning and then sync all the ods requests to the ODSs:

ceph_kill_sb() -->
    ceph_mdsc_pre_umount() -->
        ceph_flush_dirty_caps(): will do the dirty caps flushing    
    generic_shutdown_super() -->
        sync_filesystem() -->
           sync_inodes_sb()/writeback_inodes_sb(): will help submit the dirty pages as ods requests
           sb->s_op->sync_fs() -->
              ceph_sync_fs() -->
                 ceph_osdc_sync() : will sync the ods requests to the ODSs
        sop->put_super() -->
           ceph_put_super() -->
              ceph_mdsc_close_sessions() : will close the sessions

Case 2) in handle_session and (op is CEPH_SESSION_OPEN && mdsc->stopping):
This case is the same as Case 1, it is also doing the unmounting or mount fail path. And since the relating session was not even opened and there shouldn't any dirty pages exist, so no need to do the clean too.

I'm not sure what this tracker is really asking for, tbqh.

Hmm...now that I look, I do see this:

commit 6c99f2545dbb9e53afe0d1d037c51ab04ef1ff4e
Author: Sage Weil <sage@newdream.net>
Date: Mon May 10 16:12:25 2010 -0700

ceph: throw out dirty caps metadata, data on session teardown

The remove_session_caps() helper is called when an MDS closes out our
session (either normally, or as a result of a failed reconnect), and when
we tear down state for umount. If we remove the last cap, and there are
no cap migrations in progress, then there is little hope of us flushing
out that data to the mds (without heroic efforts to reconnect and flush).

So, to avoid leaving inodes pinned (due to dirty state) and crashing after
umount, throw out dirty caps state and unpin the inodes. Print a warning
to the console so we know something was lost.

NOTE: Although we drop wrbuffer refs, we don't actually mark pages clean;
maybe a truncate should be queued?

Signed-off-by: Sage Weil <sage@newdream.net>

Maybe implementing what's in that last NOTE is what was meant? I'm not sure we're really gaining anything by doing this though. Is anything actually broken here? What needs to be fixed?

Case1:
In the case when unmounting, the vfs will do this for us.

Case2:
In the case when the session is reconnecting after rejected due to the client is blacklisted, we have already fulfilled the dirty pages' invalidation.

Case3:
And in another case when we are decreasing the max_mds in the cluster side, and the kclient will release some extra sessions, as discussed with Yan Zheng, there is no need to do the dirty pages' invalidation.

Case4:
In other cases, such as when the session is _OPENing and in the _OPEN reply, but mdsc->stopping is set, the session will be set tear down too, in this case there is no need care about the dirty pages.

So currently there is no any further work to do.
THanks.