RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2025-03-25 11:48:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

doc: fix rst for watch/notify internals

Browse Source 
Signed-off-by: Sage Weil <sage@inktank.com>
This commit is contained in:
Sage Weil 2013-02-26 17:45:23 -08:00
parent c313423cfd
commit 75ce88fb69
1 changed files with 10 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
doc/dev/osd_internals/watch_notify.rst
View File

@ -28,8 +28,10 @@ Watches in order to be able to notify them upon ms_handle_reset()
Each Watch object tracks the set of currently un-acked notifies.
start_notify() on a Watch object adds a reference to a new in-progress
Notify to the Watch and either:
* if the Watch is *connected*, sends a Notify message to the client
* if the Watch is *unconnected*, does nothing.
* if the Watch is *connected*, sends a Notify message to the client
* if the Watch is *unconnected*, does nothing.
When the Watch becomes connected (in ReplicatedPG::do_osd_op_effects),
Notifies are resent to all remaining tracked Notify objects.
@ -41,11 +43,12 @@ is sent to the client.
Watch Lifecycle
---------------
A watch may be in one of 5 states:
1. Non existent.
2. On disk, but not registered with an object context.
3. Connected
4. Disconnected, callback registered with timer
5. Disconnected, callback in queue for scrub or is_degraded
1. Non existent.
2. On disk, but not registered with an object context.
3. Connected
4. Disconnected, callback registered with timer
5. Disconnected, callback in queue for scrub or is_degraded
Case 2 occurs between when an OSD goes active and the ObjectContext
for an object with watchers is loaded into memory due to an access.