glusterfs.git/xlators/cluster/afr/src/afr-inode-write.c, branch v3.7.18

cluster/afr: Prevent split-brain when bricks are brought off and on in cyclic order

2016-08-22T10:05:08+00:00

        Backport of: http://review.gluster.org/15080

When the bricks are brought offline and then online in cyclic
order while writes are in progress on a file, thanks to inode
refresh in write txns, AFR will mostly fail the write attempt
when the only good copy is offline. However, there is still a
remote possibility that the file will run into split-brain if
the brick that has the lone good copy goes offline *after* the
inode refresh but *before* the write txn completes (I call it
in-flight split-brain in the patch for ease of reference),
requiring intervention from admin to resolve the split-brain
before the IO can resume normally on the file. To get around this,
the patch does the following things:
i) retains the dirty xattrs on the file
ii) avoids marking the last of the good copies as bad (or accused)
    in case it is the one to go down during the course of a write.
iii) fails that particular write with the appropriate errno.

This way, we still have one good copy left despite the split-brain situation
which when it is back online, will be chosen as source to do the heal.

Change-Id: I7c13c6ddd5b8fe88b0f2684e8ce5f4a9c3a24a08
BUG: 1367270
Signed-off-by: Krutika Dhananjay 
Reviewed-on: http://review.gluster.org/15222
Smoke: Gluster Build System 
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System 
Reviewed-by: Oleksandr Natalenko 
Reviewed-by: Pranith Kumar Karampuri

cluster/afr: copy loc before passing to syncop

2016-08-17T11:44:11+00:00

Problem:
When io-threads is enabled on the client side, io-threads destroys the
call-stub in which the loc is stored as soon as the c-stack unwinds.
Because afr is creating a syncop with the address of loc passed in
setxattr by the time syncop tries to access it, io-threads would have
already freed the call-stub. This will lead to crash.

Fix:
Copy loc to frame->local and use it's address.

> Reviewed-on: http://review.gluster.org/15070
> Smoke: Gluster Build System 
> CentOS-regression: Gluster Build System 
> NetBSD-regression: NetBSD Build System 
> Reviewed-by: Ravishankar N 

BUG: 1367305
Change-Id: I16987e491e24b0b4e3d868a6968e802e47c77f7a
Signed-off-by: Pranith Kumar K 
Signed-off-by: Oleksandr Natalenko 
Reviewed-on: http://review.gluster.org/15168
Reviewed-by: Ravishankar N 
Smoke: Gluster Build System 
CentOS-regression: Gluster Build System 
NetBSD-regression: NetBSD Build System

cluster/afr: Unwind xdata_rsp even in case of failures

2016-06-03T13:24:00+00:00

DHT expects GF_PREOP_CHECK_FAILED to be present in xdata_rsp in case of mkdir
failures because of stale layout. But AFR was unwinding null xdata_rsp in case
of failures. This was leading to mkdir failures just after remove-brick. Unwind
the xdata_rsp in case of failures to make sure the response from brick reaches
dht.

 >BUG: 1340623
 >Change-Id: Idd3f7b95730e8ea987b608e892011ff190e181d1
 >Signed-off-by: Pranith Kumar K 
 >Reviewed-on: http://review.gluster.org/14553
 >NetBSD-regression: NetBSD Build System 
 >Reviewed-by: Ravishankar N 
 >Smoke: Gluster Build System 
 >CentOS-regression: Gluster Build System 
 >Reviewed-by: Anuradha Talur 
 >Reviewed-by: Krutika Dhananjay 

BUG: 1340992
Change-Id: I2641d35a851be692aa223dfea5d082245ac6c2bc
Signed-off-by: Pranith Kumar K 
Reviewed-on: http://review.gluster.org/14633
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System 
Smoke: Gluster Build System

cluster/afr: Don't let NFS cache stat after writes.

2016-06-03T13:23:47+00:00

Problem:
Afr does post-ops after write but the stat buffer it unwinds is at the
time of write, so if nfs client caches this, it will see different
ctime when it does stat on it after post-op is done. From NFS client's
perspective it thinks the file is changed. Tar which depends on this
to be correct keeps giving 'file changed as we read it' warning.
If Afr instead has to choose to unwind after post-op, eager-lock,
delayed-post-op will have to be disabled which will lead to bad
performance for all write usecases.

Fix:
Don't let client cache stat after write.

 >Change-Id: Ic6062acc6e5cdd97a9c83c56bd529ec83cee8a23
 >BUG: 1302948
 >Signed-off-by: Pranith Kumar K 
 >Signed-off-by: Anuradha Talur 
 >Reviewed-on: http://review.gluster.org/13785
 >Smoke: Gluster Build System 
 >NetBSD-regression: NetBSD Build System 
 >CentOS-regression: Gluster Build System 
 >Reviewed-by: Niels de Vos 

BUG: 1312721
Change-Id: I42a5d524bcf2a2034fe48ee8454812ca26a98c37
Signed-off-by: Pranith Kumar K 
Reviewed-on: http://review.gluster.org/14454
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System 
Smoke: Gluster Build System

arbiter: write performance improvement

2016-04-11T12:04:04+00:00

Backport of: http://review.gluster.org/#/c/13906

Problem: The throughput for a 'dd' workload was much less for arbiter
configuration when compared to normal replica-3 volume. There were 2
issues:

i)arbiter_writev was using the request dict as response dict while
unwinding, leading to incorect GLUSTERFS_WRITE_IS_APPEND and
GLUSTERFS_OPEN_FD_COUNT values (=4), leading to immediate post-ops
because is_afr_delayed_changelog_post_op_needed() failed due to
afr_are_multiple_fds_opened() check.

ii) The arbiter code in afr was setting local->transaction.{start and len} =0
to take full file locks. What this meant was even for simultaenous but
non-overlapping writevs, afr_transaction_eager_lock_init() was not
happening because afr_locals_overlap() always stays true. Consequently
is_afr_delayed_changelog_post_op_needed() failed due to
local->delayed_post_op not being set.

Fix:
i) Send appropriate response dict values in arbiter_writev.
ii) Modify flock params instead of local->transaction.{start and len} to
take full file locks in the transaction.

Also changed _fill_writev_xdata() in posix to fill rsp_xdata for
whatever key is requested for.

Change-Id: I1c5fc5e98aba49ade540bb441a022e65b753432a
BUG: 1324809
Signed-off-by: Ravishankar N 
Reported-by: Robert Rauch 
Reported-by: Russel Purinton 
Reviewed-on: http://review.gluster.org/13925
Smoke: Gluster Build System 
Reviewed-by: Pranith Kumar Karampuri 
CentOS-regression: Gluster Build System 
NetBSD-regression: NetBSD Build System

cluster/afr: Fix read-child selection in entry create fop

2016-03-29T08:07:39+00:00

When an entry is being created the inode is yet to be linked
so args must be filled with gfid and ia_type for it to give
consistent iatt.

Also handle Dht sending fops on inode not yet linked.

 >BUG: 1302948
 >Change-Id: I6969cacb437cad02f66716f3bf8ec004ffe7c691
 >Signed-off-by: Pranith Kumar K 
 >Reviewed-on: http://review.gluster.org/13827
 >Reviewed-by: Krutika Dhananjay 
 >Smoke: Gluster Build System 
 >Reviewed-by: Anuradha Talur 
 >NetBSD-regression: NetBSD Build System 
 >CentOS-regression: Gluster Build System 
 >(cherry picked from commit 2d0c2a9e15a581950a966841866f1f18a8888440)

Change-Id: I4be73580368b17af71486693682af66d97d6f70d
BUG: 1312721
Signed-off-by: Pranith Kumar K 
Reviewed-on: http://review.gluster.org/13836
Smoke: Gluster Build System 
CentOS-regression: Gluster Build System 
NetBSD-regression: NetBSD Build System 
Reviewed-by: Krutika Dhananjay

afr : Enable auto heal when replica count increases

2016-03-23T11:49:36+00:00

            Backport of: http://review.gluster.org/12454

    This patch is part two change to prevent data loss
    in a replicate volume on doing a add-brick operation.

    Problem: After doing add-brick, there is a chance that
    self heal might happen from the newly added brick rather
    than the source brick, leading to data loss.

    Solution: Mark pending changelogs on afr children for
    the new afr-child so that heal is performed in the
    correct direction.

            >Change-Id: I11871e55eef3593aec874f92214a2d97da229b17
            >BUG: 1276203
            >Signed-off-by: Anuradha Talur 
            >Reviewed-on: http://review.gluster.org/12454
            >Smoke: Gluster Build System 
            >Reviewed-by: Pranith Kumar Karampuri 
            >Tested-by: Pranith Kumar Karampuri 
            >CentOS-regression: Gluster Build System 
            >NetBSD-regression: NetBSD Build System 

Change-Id: Iae6af44f97e612cb3ee8c642254ec3d15ac063f5
BUG: 1320020
Signed-off-by: Anuradha Talur 
Reviewed-on: http://review.gluster.org/13807
Smoke: Gluster Build System 
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System 
Reviewed-by: Raghavendra G

cluster/afr: Move remaining gf_logs to gf_msgs

2016-03-07T07:19:26+00:00

 >Change-Id: I48d9e5313bd3ccf9fe26c90a7051a8a174d75c49
 >BUG: 1296818
 >Signed-off-by: Pranith Kumar K 
 >Reviewed-on: http://review.gluster.org/13195
 >Reviewed-by: Krutika Dhananjay 
 >Smoke: Gluster Build System 
 >NetBSD-regression: NetBSD Build System 
 >CentOS-regression: Gluster Build System 
 >Reviewed-by: Vijay Bellur 

BUG: 1315142
Change-Id: Iab55c3fd41777e9fe295f674c3ee410fd780b418
Signed-off-by: Pranith Kumar K 
Reviewed-on: http://review.gluster.org/13617
Smoke: Gluster Build System 
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System 
Reviewed-by: Krutika Dhananjay

cluster/ec: Mark self-heal fops as internal

2016-03-05T01:26:21+00:00

 >Change-Id: I8ae7af266d3e00460f0cfdc9389a926e5f2fee36
 >BUG: 1282761
 >Signed-off-by: Pranith Kumar K 
 >Reviewed-on: http://review.gluster.org/12598
 >Tested-by: Gluster Build System 
 >Tested-by: NetBSD Build System 
 >Reviewed-by: Xavier Hernandez 

BUG: 1283757
Change-Id: Ic20d4ee031265305db1a6ed2cf591ce94b7d0749
Signed-off-by: Pranith Kumar K 
Reviewed-on: http://review.gluster.org/12668
Smoke: Gluster Build System 
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System

afr: perform replace-brick in a synctask

2015-09-16T02:27:44+00:00

Backport of http://review.gluster.org/#/c/12169/

Problem:
replace-brick setxattr is not performed inside a synctask. This can lead
to hangs if the setxattr is executed by epoll thread, as the epoll
thread will be waiting for replies to come where as epoll thread is the
thread that needs to epoll_ctl for reading from socket and listen.

Fix:
Move replace-brick to synctask to prevent epoll thread hang.

This patch is in line with the fix performed in
http://review.gluster.org/#/c/12163/

Change-Id: I7284930ead9b0adaa0257f21ec2d893fa5a7146f
BUG: 1262547
Signed-off-by: Ravishankar N 
Reviewed-on: http://review.gluster.org/12172
Tested-by: NetBSD Build System 
Tested-by: Gluster Build System 
Reviewed-by: Pranith Kumar Karampuri