| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
| |
Signed-off-by: Pavan Vilas Sondur <pavan@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 1042 (Use correct flock structures in lk fops)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1042
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use the frame->root address as lk-owner when FUSE does
not supply lk_owner.
Raghu,
I unit tested this patch with dbench and self heal tests.
Did not observe lk-owner=0 in any server logs. Can you
verify this patch with the other tests you had run today?
Signed-off-by: Pavan Vilas Sondur <pavan@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 1032 (Set lock-owner with pid when fuse does not supply value)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1032
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
completes successfully.
Signed-off-by: Simone Gotti <simone.gotti@gmail.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 884 (I/O errors after fixed split brain and successfully completed self heal)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=884
Signed-off-by: Pavan Vilas Sondur <pavan@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 884 (I/O errors after fixed split brain and successfully completed self heal)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=884
|
|
|
|
|
|
|
|
| |
Signed-off-by: Vijay Bellur <vijay@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 805 (memory leak in afr)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=805
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch makes the replicate readdir call fail over to the next
subvolume if the first call fails. It takes care to ensure that
entries are not duplicated.
The failover behavior of readdir only comes into effect if
the option 'strict-readdir' is on.
Signed-off-by: Vikas Gorur <vikas@dev.gluster.com>
Signed-off-by: root <root@client02.(none)>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 453 (afr_readdir does not fail over)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=453
|
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@blackhole.gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The janitor thread deletes all files and directories in
the "/" GF_REPLICATE_TRASH_DIR directory. This directory
is used by replicate self-heal to dump files and
directories it deletes.
This is needed because letting replicate walk the directory
tree and delete a directory and all its children is too racy.
Instead, replicate self-heal only does an atomic rename(),
and the janitor thread takes care of actually deleting them.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 227 (replicate selfheal does not remove directory with contents in it)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=227
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
doing self-heal.
This patch sets the read-subvolume equal to the self-heal "source"
even if we're not doing self-heal (because some one else is already
doing it).
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 320 (Improve self-heal performance)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=320
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
creation fops.
This fixes fuse_create_cbk conflict warnings and random errors while
running dbench (typically open handle failure with ENOENT).
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 315 (generation number support)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=315
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch completes the previous patch for self-heal of
open fds in replicate.
If an fd was never opened on a subvolume, we remember that
and do the open after we've done self-heal on that fd.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
| |
Cleaned up the self-heal interface to callers.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch brings in partial support for self-heal of open
fds. The precondition is that the fd should have been opened
successfully during the initial open() (or create()), and we
assume that protocol/client has successfully reopened the fd
when the subvolume comes back up.
It works by doing an "up/down flush" (a dummy flush transaction
to do post-op wherever necessary) and then triggering
data self-heal on the file in the post-post-op hook of the
dummy flush transaction. This ensures that any writes
that come in during self-heal will wait until self-heal completes.
The up/down flush is also done when a subvolume goes down,
so that post-op is done on all subvolumes where pre-op was done.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
| |
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Refactored the operation of the data self-heal algorithm
as:
* open all fd's (if fd not supplied by caller)
* lock 0-0 (if lock not supplied by caller)
* fxattrop, fstat (instead of lookup)
... self heal ...
* unlock (if lock not supplied by caller)
* close (if fd not supplied by caller).
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
|
| |
Data self-heal now holds blocking locks, and instead of locking
on all subvolumes, it only locks on {data-lock-server-count} subvolumes.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 170 (Auto-heal fails on files that are open()-ed/mmap()-ed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=170
|
|
|
|
|
|
|
|
|
|
|
| |
Set opendir_done and split_brain flags correctly
in the inode context.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 249 (Self heal of a file that does not exist on the first subvolume)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=249
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch does two things related to revalidate:
1) If a revalidate fails on any subvolume, the entire lookup
call is failed.
2) Self-heal is not triggered on a revalidate if revalidate
has failed on any subvolume.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 389 (auto-heal fails randomly and causes "Stale NFS file handle" errors)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=389
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The problem: If some files on the first subvolume disappeared
without leaving a trace in the entry changelog (this can happen,
for example, when an fsck has deleted files or when a hard drive
is replaced), those files would never be self-healed even though
they would be present on the second subvolume. This is because
readdir is sent only to the first subvolume, and since the files
don't appear in the directory listing, no lookup would ever be
sent on them.
This patch fixes this problem by doing a readdir on all the subvolumes
during the first opendir on a directory inode. If a discrepancy in the
contents is detected, entry self-heal in a special "force merge" mode
is triggered on that directory.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 249 (Self heal of a file that does not exist on the first subvolume)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=249
|
|
|
|
|
|
|
|
|
|
|
| |
If the inodelk_count or entrylk_count is positive on a
file/directory, don't try to do self-heal on it.
Signed-off-by: Vikas Gorur <vikas@gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 326 ([2.0.8rc9] Spurious self-heal)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=326
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If entry self-heal determines that a file/directory should
be deleted from a subvolume, move that entry to a directory
called "/.trash" on that subvolume. This is for two reasons:
1) It limits the damage that can be done by a "wrong" entry
self-heal.
2) It solves the problem of a to-be-deleted directory not
being empty.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 227 (replicate selfheal does not remove directory with contents in it)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=227
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 146 (Add setattr FOP)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=146
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 145 (NFSv3 related additions to 2.1 task list)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=145
|
|
|
|
|
|
|
|
|
|
| |
During entry self-heal, make sure not only that a symlink
exists on all subvolumes, but also that their targets match.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 193 (symlink contents not self-healed by replicate)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=193
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch introduces a new option "background-self-heal-count", with a
default value of 16.
This means that upto {background-self-heal-count} number of files/directories
will be healed in the background at any given time. If such number of self-heals
are already in progress, further self-heals take place in the foreground.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 320 (Improve self-heal performance)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=320
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Start upto "data-self-heal-window-size" instances of the read-write loop
of the "full" data self-heal algorithm simultaneously.
Add a new option "data-self-heal-window-size" with range [1-1024],
and a default value of 16.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 320 (Improve self-heal performance)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=320
|
|
|
|
|
|
|
|
|
|
|
| |
While creating/deleting an entry as part of entry self-heal,
set the parent directory's mtime to match that on the source
subvolume.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 137 (Parent directory mtime not reset after a create in self-heal)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=137
|
|
|
|
|
|
|
|
|
|
| |
afr selfheal now remembers all the nodes on which locks were successfully
held and sends unlocks only to those nodes
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 112 (parallel deletion of files mounted by different clients on the same back-end hangs and/or does not completely delete)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=112
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 285 ("first up child" can change during a transaction)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=285
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 146 (Add setattr FOP)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=146
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
create, mkdir, symlink, mknod: Prefer to return itransform'd inode number
from the first_up_child. If not, fall back on any other child that returned
succcess.
link, rename: Return the same inode number that
was passed as part of loc_t.
Also adds a new member to afr_local_t, local->first_up_child
which is initialized at the start of the transaction. This
fixes the race where a subvolume might go down during the transaction
and thus have the first_up_child change.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 285 ("first up child" can change during a transaction)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=285
|
|
|
|
|
|
|
|
|
|
| |
option: data-self-heal-algorithm type: string default: "full"
This option allows the user to specify the algorithm to
be used for data self-heal. Currently supported values
are "full" and "diff".
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
|
|
|
|
|
|
|
|
| |
Abstract the read/write loop part of data self-heal. This
patch has support for the "full" (i.e., read and write entire
file) algorithm.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
|
|
|
|
|
|
|
|
|
|
| |
Remember the inode number that had been returned in lookup_cbk
and set the stat buf->ino to the same.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 166 (libglusterfsclient: Cached stat buf inode is different from ino in inode_t)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=166
|
|
|
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 138 (create family calls do not return stat buf from read child)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=138
|
|
|
|
|
|
|
|
|
|
|
| |
Also modifies the inode-write ops to wait for the call
to read-child to return (whether success or failure)
before unwinding.
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 125 (stat information not returned from the same subvolume always)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=125
|
|
|
|
|
|
|
|
|
|
| |
Also fixes a bug in the "KLUDGE" part. It was setting lookup_buf
when it should have been setting local->cont.lookup.buf
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
BUG: 116 (Replicate: Need inode number from first subvolume on fresh lookup)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=116
|
|
|
|
|
|
|
|
| |
Save the original pid while locking and restore it
after the FOP is done. This ensures posix-locks can
release locks (fcntl) properly.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
|
|
| |
subvolumes while keeping existing data.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
|
|
| |
If a writev fails, remember it by marking it in the fd context.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
|
|
|
|
|
| |
Transaction fops earlier called afr_transaction_child_died only
if an fop failed due to ENOTCONN or EBADFD. Now they consider a child
dead regardless of the reason for failure. This handles cases such
as ENOSPC.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
| |
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
|
|
|
|
| |
Notification of a split-brain situation, which was earlier signalled
by the mere presence of inode context is now signalled by
the 'split_brain' member in the structure.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|
|
|
|
| |
updated copyright header to include 2009.
Signed-off-by: Anand V. Avati <avati@amp.gluster.com>
|
|
|