[Bugs] [Bug 1196632] dist-geo-rep: Concurrent renames and node reboots results in slave having both source and destination of file with destination being 0 byte sticky file
bugzilla at redhat.com
bugzilla at redhat.com
Mon Mar 16 04:54:03 UTC 2015
https://bugzilla.redhat.com/show_bug.cgi?id=1196632
--- Comment #5 from Anand Avati <aavati at redhat.com> ---
COMMIT: http://review.gluster.org/9759 committed in master by Vijay Bellur
(vbellur at redhat.com)
------
commit f0224ce93ae9ad420e23612fe6e6707a821f9cab
Author: Kotresh HR <khiremat at redhat.com>
Date: Mon Feb 23 14:46:48 2015 +0530
feature/geo-rep: Active Passive Switching logic flock
CURRENT DESIGN AND ITS LIMITATIONS:
-----------------------------------
Geo-replication syncs changes across geography using changelogs captured
by changelog translator. Changelog translator sits on server side just
above posix translator. Hence, in distributed replicated setup, both
replica pairs collect changelogs w.r.t their bricks. Geo-replication
syncs the changes using only one brick among the replica pair at a time,
calling it as "ACTIVE" and other non syncing brick as "PASSIVE".
Let's consider below example of distributed replicated setup where
NODE-1 as b1 and its replicated brick b1r is in NODE-2
NODE-1 NODE-2
b1 b1r
At the beginning, geo-replication chooses to sync changes from NODE-1:b1
and NODE-2:b1r will be "PASSIVE". The logic depends on virtual getxattr
'trusted.glusterfs.node-uuid' which always returns first up subvolume
i.e., NODE-1. When NODE-1 goes down, the above xattr returns NODE-2 and
that is made 'ACTIVE'. But when NODE-1 comes back again, the above xattr
returns NODE-1 and it is made 'ACTIVE' again. So for a brief interval of
time, if NODE-2 had not finished processing the changelog, both NODE-2
and NODE-1 will be ACTIVE causing rename race as mentioned in the bug.
SOLUTION:
---------
1. Have a shared replicated storage, a glusterfs management volume specific
to geo-replication.
2. Geo-rep creates a file per replica set on management volume.
3. fcntl lock on the above said file is used for synchronization
between geo-rep workers belonging to same replica set.
4. If management volume is not configured, geo-replication will back
to previous logic of using first up sub volume.
Each worker tries to lock the file on shared storage, who ever wins will
be ACTIVE. With this, we are able to solve the problem but there is an
issue when the shared replicated storage goes down (when all replicas
goes down). In that case, the lock state is lost. So AFR needs to rebuild
the
lock state after brick comes up.
NOTE:
-----
This patch brings in the, pre-requisite step of setting up management
volume
for geo-replication during creation.
1. Create mgmt-vol for geo-replicatoin and start it. Management volume
should
be part of master cluster and recommended to be three way replicated
volume having each brick in different nodes for availability.
2. Create geo-rep session.
3. Configure mgmt-vol created with geo-replication session as follows.
gluster vol geo-rep <mastervol> slavenode::<slavevol> config meta_volume
\
<meta-vol-name>
4. Start geo-rep session.
Backward Compatiability:
-----------------------
If management volume is not configured, it falls back to previous logic of
using node-uuid virtual xattr. But it is not recommended.
Change-Id: I7319d2289516f534b69edd00c9d0db5a3725661a
BUG: 1196632
Signed-off-by: Kotresh HR <khiremat at redhat.com>
Reviewed-on: http://review.gluster.org/9759
Tested-by: Gluster Build System <jenkins at build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur at redhat.com>
--
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=kjh1Xrjixw&a=cc_unsubscribe
More information about the Bugs
mailing list