| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
For write fops afr's transaction eager-lock init adds transactions
that can share eager-lock to fdctx list. But if eager-lock finodelk
fop fails the stub remains in the list. This could later lead to
corruption of the list and lead to infinite loop on the list
leading to a mount hang.
Fix:
Remove the stub when finodelk fails.
Change-Id: I0ed4bc6b62f26c5e891c1181a6871ee6e4f4f5fd
BUG: 1063190
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/6944
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Remove client side self-healing completely (opendir, openfd, lookup)
- Re-work readdir-failover to work reliably in case of NFS
- Remove unused/dead lock recovery code
- Consistently use xdata in both calls and callbacks in all FOPs
- Per-inode event generation, used to force inode ctx refresh
- Implement dirty flag support (in place of pending counts)
- Eliminate inode ctx structure, use read subvol bits + event_generation
- Implement inode ctx refreshing based on event generation
- Provide backward compatibility in transactions
- remove unused variables and functions
- make code more consistent in style and pattern
- regularize and clean up inode-write transaction code
- regularize and clean up dir-write transaction code
- regularize and clean up common FOPs
- reorganize transaction framework code
- skip setting xattrs in pending dict if nothing is pending
- re-write self-healing code using syncops
- re-write simpler self-heal-daemon
Change-Id: I1e4080c9796c8a2815c2dab4be3073f389d614a8
BUG: 1021686
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/6010
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
Under the entry self heal, readlink is done at the
source and sink. When readlink is done at the sink,
because link is not present at the sink, afr expects
ENOENT. AFR translator takes decisions for new link
creation based on ENOENT but server translator is modified
to return ESTALE because of which afr xlator is not able
to heal.
Fix:
The check for inode absence at server includes ESTALE as
well.
Change-Id: I319e4cb4156a243afee79365b7b7a5a7823e9a24
BUG: 1046624
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/6599
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Also renamed allow-sh-for-running-transaction -> attempt-self-heal
Change-Id: I134cc79e663b532e625ffc342c59e49e71644ab3
BUG: 1039544
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/6463
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: venkatesh somyajulu <vsomyaju@redhat.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
glfs_zerofill() can be potentially called to zero-out entire file and
hence allow for bigger value of length parameter.
Change-Id: I75f1d11af298915049a3f3a7cb3890a2d72fca63
BUG: 1028673
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-on: http://review.gluster.org/6266
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: M. Mohan Kumar <mohan@in.ibm.com>
Tested-by: M. Mohan Kumar <mohan@in.ibm.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add support for a new ZEROFILL fop. Zerofill writes zeroes to a file in
the specified range. This fop will be useful when a whole file needs to
be initialized with zero (could be useful for zero filled VM disk image
provisioning or during scrubbing of VM disk images).
Client/application can issue this FOP for zeroing out. Gluster server
will zero out required range of bytes ie server offloaded zeroing. In
the absence of this fop, client/application has to repetitively issue
write (zero) fop to the server, which is very inefficient method because
of the overheads involved in RPC calls and acknowledgements.
WRITESAME is a SCSI T10 command that takes a block of data as input and
writes the same data to other blocks and this write is handled
completely within the storage and hence is known as offload . Linux ,now
has support for SCSI WRITESAME command which is exposed to the user in
the form of BLKZEROOUT ioctl. BD Xlator can exploit BLKZEROOUT ioctl to
implement this fop. Thus zeroing out operations can be completely
offloaded to the storage device , making it highly efficient.
The fop takes two arguments offset and size. It zeroes out 'size' number
of bytes in an opened file starting from 'offset' position.
This patch adds zerofill support to the following areas:
- libglusterfs
- io-stats
- performance/md-cache,open-behind
- quota
- cluster/afr,dht,stripe
- rpc/xdr
- protocol/client,server
- io-threads
- marker
- storage/posix
- libgfapi
Client applications can exloit this fop by using glfs_zerofill introduced in
libgfapi.FUSE support to this fop has not been added as there is no system call
for this fop.
Changes from previous version 3:
* Removed redundant memory failure log messages
Changes from previous version 2:
* Rebased and fixed build error
Changes from previous version 1:
* Rebased for latest master
TODO :
* Add zerofill support to trace xlator
* Expose zerofill capability as part of gluster volume info
Here is a performance comparison of server offloaded zeofill vs zeroing
out using repeated writes.
[root@llmvm02 remote]# time ./offloaded aakash-test log 20
real 3m34.155s
user 0m0.018s
sys 0m0.040s
[root@llmvm02 remote]# time ./manually aakash-test log 20
real 4m23.043s
user 0m2.197s
sys 0m14.457s
[root@llmvm02 remote]# time ./offloaded aakash-test log 25;
real 4m28.363s
user 0m0.021s
sys 0m0.025s
[root@llmvm02 remote]# time ./manually aakash-test log 25
real 5m34.278s
user 0m2.957s
sys 0m18.808s
The argument log is a file which we want to set for logging purpose and
the third argument is size in GB .
As we can see there is a performance improvement of around 20% with this
fop.
Change-Id: I081159f5f7edde0ddb78169fb4c21c776ec91a18
BUG: 1028673
Signed-off-by: Aakash Lal Das <aakash@linux.vnet.ibm.com>
Signed-off-by: M. Mohan Kumar <mohan@in.ibm.com>
Reviewed-on: http://review.gluster.org/5327
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Bail out of fops if split brain has been detected during lookup
Change-Id: Id387dbb1a25eec4a121dedceadc6069bdea24b5d
BUG: 1010834
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-on: http://review.gluster.org/5988
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently to know the number of files to be healed, either user
has to go to backend and check the number of entries present in
indices/xattrop directory. But if a volume consists of large
number of bricks, going to each backend and counting the number
of entries is a time-taking task. Otherwise user can give
gluster volume heal vol-name info command but with this
approach if no. of entries are very hugh in the indices/
xattrop directory, it will comsume time.
So as a feature, new command is implemented.
Command 1: gluster volume heal vn statistics heal-count
This command will get the number of entries present in
every brick of a volume. The output displays only entries
count.
Command 2: gluster volume heal vn statistics heal-count
replica 192.168.122.1:/home/user/brickname
Here if we are concerned with just one replica.
So providing any one of the brick of a replica will get
the number of entries to be healed for that replica only.
Example:
Replicate volume with replica count 2.
Backend status:
--------------
[root@dhcp-0-17 xattrop]# ls -lia | wc -l
1918
NOTE: Out of 1918, 2 entries are <xattrop-gfid> dummy
entries so actual no. of entries to be healed are
1916.
[root@dhcp-0-17 xattrop]# pwd
/home/user/2ty/.glusterfs/indices/xattrop
Command output:
--------------
Gathering count of entries to be healed on volume volume3 has been successful
Brick 192.168.122.1:/home/user/22iu
Status: Brick is Not connected
Entries count is not available
Brick 192.168.122.1:/home/user/2ty
Number of entries: 1916
Change-Id: I72452f3de50502dc898076ec74d434d9e77fd290
BUG: 1015990
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/6044
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"gluster volume heal volumename statistics" command gives the summary
of the afr crawl done based on the entries present in the xattrop
directory. Whenever afr crawls are attempted, the beginning time of
crawl, end time of crawl, no of files healed, heal-failed count and
number of files in split brain are shown along with the type of the
crawl. If crawl is already in progress then it will give the number
of files healed, heal failed count and number of files in split-brain
from the beginning of the crawl and instead of telling the end time of
the crawl, "CRAWL IN PROGRESS" message will be shown.
Output format:
command: "gluster volume heal volume-name statistics"
Output:
Gathering afr crawl statistics crawl statistics on volume volume-name
has been successful
------------------------------------------------
Crawl statistics for brick no 0
Hostname of brick 192.168.122.248
Starting time of crawl: Wed Jul 10 15:52:38 2013
Ending time of crawl: Wed Jul 10 15:52:38 2013
Type of crawl: INDEX
No. of entries healed: 0
No. of entries in split-brain: 0
No. of heal failed entries: 0
Starting time of crawl: Wed Jul 10 15:52:38 2013
Ending time of crawl: Wed Jul 10 15:52:38 2013
Type of crawl: INDEX
No. of entries healed: 0
No. of entries in split-brain: 0
No. of heal failed entries: 0
------------------------------------------------
Crawl statistics for brick no 1
Hostname of brick 192.168.122.1
Starting time of crawl: Wed Jul 10 15:52:42 2013
Ending time of crawl: Wed Jul 10 15:52:42 2013
Type of crawl: INDEX
No. of entries healed: 0
No. of entries in split-brain: 0
No. of heal failed entries: 0
Starting time of crawl: Wed Jul 10 15:52:42 2013
Ending time of crawl: Wed Jul 10 15:52:42 2013
Type of crawl: INDEX
No. of entries healed: 0
No. of entries in split-brain: 0
No. of heal failed entries: 0
--------------------------------------------------
Change-Id: I10bf9d10b005741db9973fb1352e0dd59ed99aa9
BUG: 949400
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/4790
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
'-' can be present in a volume. This may lead to domain
collisions in future.
Tests:
Checked in gdb that domain comes with ':' separator:
Breakpoint 1, pl_common_inodelk (frame=0x7fdabcce51a4,
this=0x8bde20, volume=0x8b50d0 "r2-replicate-0:self-heal",
inode=0x7fdab822f0e8, cmd=6, flock=0x7fdabc76eee4,
loc=0x7fdabc76ede4, fd=0x0, xdata=0x7fdabc6e0ab0) at inodelk.c:597
Change-Id: I4456ae35ac8bf21e6361c34e9ad437f744a2e84b
BUG: 993981
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/6025
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
| |
Change-Id: Ia7b324b86d6a7051d187106d7a060155e77defc5
BUG: 910217
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5238
Reviewed-by: Ravishankar N <ravishankar@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Additional information for source and sinks are added.
Change-Id: I1704956ff86ac3ae36744efe7499c1d1c43faeaf
BUG: 968301
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/5638
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Durability of appending writes is implicit in the file size. Therefore
performing an explicit fsync() is unnecessary in such cases as self-heal
can check for the size of file when pending changelog is not unambiguous.
Change-Id: I05446180a91d20e0dbee5de5a7085b87d57f178a
BUG: 927146
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/5501
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lets say mount1 has eager-lock(full-lock) and after the eager-lock
is taken mount2 opened the same file, it won't be able to
perform any data operations until mount1 releases eager-lock.
To avoid such scenario do not enable eager-lock for transaction
if open-fd-count is > 1. Delaying of changelog piggybacking is
avoided in this situation.
Change-Id: I51b45d6a7c216a78860aff0265a0b8dabc6423a5
BUG: 910217
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5432
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: venkatesh somyajulu <vsomyaju@redhat.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
| |
Change-Id: I95e47e589419dc6a032cbd8ba01964b6c176c2d5
BUG: 927146
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5408
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
At the moment data-self-heal acquires locks in following
pattern. It takes full file lock then gets xattrs on files on both
replicas. Decides sources/sinks based on the xattrs. Now it acquires
lock from 0-128k then unlocks the full file lock. Syncs 0-128k range
from source to sink now acquires lock 128k+1 till 256k then unlocks
0-128k, syncs 128k+1 till 256k block... so on finally it takes full file
lock again then unlocks the final small range block.
It decrements pending counts and then unlocks the full file lock.
This pattern of locks is chosen to avoid more than 1 self-heal
to be in progress. BUT if another self-heal tries to take a full
file lock while a self-heal is already in progress it will be put in
blocked queue, further inodelks from writes by the application will
also be put in blocked queue because of the way locks xlator grants
inodelks. So until the self-heal is complete writes are blocked.
Here is the code:
xlators/features/locks/src/inodelk.c - line 225
if (__blocked_lock_conflict (dom, lock) && !(__owner_has_lock (dom, lock))) {
ret = -EAGAIN;
if (can_block == 0)
goto out;
gettimeofday (&lock->blkd_time, NULL);
list_add_tail (&lock->blocked_locks, &dom->blocked_inodelks);
}
This leads to hangs in applications.
Fix:
Since we want to prevent two parallel self-heals. We let them compete
in a separate "domain". Lets call the domain on which the locks have
been taken on in previous approach as "data-domain".
In the new approach When a self-heal is triggered,
it acquires a full lock in the new domain "self-heal-domain".
After this it performs data-self-heal using the locks in
"data-domain" as before.
unlock the full file lock in "self-heal-domain"
With this approach, application's writevs don't have to wait
in pending queue when more than 1 self-heal is triggered.
Change-Id: Id79aef3dfa888945977fb9758374ac41c320d0d5
BUG: 967717
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5100
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- afr_local_copy should not be memduping locked nodes, that would
mean that lock is taken in self-heal on those nodes even before
it actually takes the lock. So removed memdup code. Even entry
lock related copying (lockee info) is also not necessary for
self-heal functionality, so removing that as well. Since it is
not local_copy anymore changed its name.
- My editor changed tabs to spaces.
Change-Id: I8dfb92cb8338e9a967c06907a8e29a8404782d61
BUG: 967717
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5099
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
| |
Change-Id: I40eec20ca6b3f857245a2438883822e251077ee9
BUG: 979365
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5269
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
At the moment afr-flush makes sure that a delayed post-op
is woken up but it does not wait for it to complete the
post-op before flush unwinds.
These are the steps that are happening:
1) flush fop comes on an fd which wakes up a delayed post-op
and continues with the flush fop.
2) post-op sends fsync on the wire.
3) flush completes and unwinds to fuse.
4) graph switch happens on the fuse mount disconnecting the
old graph's client connections to bricks.
5) xattrop after fsync fails with ENOTCONN because the
connections from old graph are taken down now.
Fix:
Wait for post-op to complete before starting to flush.
We could make flush act similar to fsync (i.e.) wind
flush as is but wait for post-op to complete before unwinding
flush, but it is better to send flush as the final fop. So
wind of flush will start after post-op is complete. Had to
change fsync to accommodate this change.
Change-Id: I93aa642647751969511718b0e137afbd067b388a
BUG: 980548
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/5274
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
Currently whenever there is metadata split-brain, a variable
sh->op_failed is set to 1 to denote that self heal got failed.
But if we proceed for data self heal, even code-path of data
self heal also relies on the sh->op_failed variable. So if will
check for sh->op_failed variable and will eventually fails to
do data self heal. So needed a mechanism to allow data self heal
even if metadata is in split brain.
Fix:
Some data structure revamp is done in
http://review.gluster.com/#/c/5106/ fix and this patch is
based on the above fix. Now we can store which particular self-heal
got failed i.e GFID_OR_MISSING_ENTRY_SELF_HEAL, METADATA, DATA,
ENTRY. And we can do two types of self heal failure check.
1. Individual type check: We can check which among all four
(Metadata, Data, Gfid or missing entry, entry self heal)
got failed.
2. In afr_self_heal_completion_cbk, we need to make check
based on the fact that if any specific self heal got failed treat
the complete self heal as failure so that it will populate
corresponding circular buffer of event history accordingly.
Change-Id: Icb91e513bcc752386fc8a78812405cfabe5cac2d
BUG: 977797
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/5253
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
As the end of the self heal, message logged by
"afr_self_heal_completion_cbk" is inadequate to determine what exactly failed
during the course of afr self heal. It is worth to have knowledge of what all
types of self heal got triggered for an entity and whether the status is success
or failure.
Fix:
At the end of self heal, it will log information about out of 4 types of self
heal (gfid or missing entry self heal, metadata, data and entry self heal),
who all got triggered and who all got failed or successful at the end.
Change-Id: I5360762fbd7d391ac4c6af6706b4835c5801835a
BUG: 968301
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/5106
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add support for the DISCARD file operation. Discard punches a hole
in a file in the provided range. Block de-allocation is implemented
via fallocate() (as requested via fuse and passed on to the brick
fs) but a separate fop is created within gluster to emphasize the
fact that discard changes file data (the discarded region is
replaced with zeroes) and must invalidate caches where appropriate.
BUG: 963678
Change-Id: I34633a0bfff2187afeab4292a15f3cc9adf261af
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/5090
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Implement support for the fallocate file operation. fallocate
allocates blocks for a particular inode such that future writes
to the associated region of the file are guaranteed not to fail
with ENOSPC.
This patch adds fallocate support to the following areas:
- libglusterfs
- mount/fuse
- io-stats
- performance/md-cache,open-behind
- quota
- cluster/afr,dht,stripe
- rpc/xdr
- protocol/client,server
- io-threads
- marker
- storage/posix
- libgfapi
BUG: 949242
Change-Id: Ice8e61351f9d6115c5df68768bc844abbf0ce8bd
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4969
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Today there is a non-obvious dependence of eager-locking on
write-behind. The reason is that eager-locking works as long
as the inheriting transaction has no overlaps with any of the
transactions already in progress. While write-behind provides
non-overlapping writes as a side-effect most of times (and only
guarantees it when strict-write-ordering option is enabled,
which is not on by default) eager-lock needs the behavior
as a guarantee. This is leading to complex and unwanted checks
for the presence of write-behind in the graph, for the simple
task of checking for overlaps.
This patch removes the interdependence between eager-locking
and write-behind by making eager-locking do its own overlap checks
with in-progress writes.
Change-Id: Iccba1185aeb5f1e7f060089c895a62840787133f
BUG: 912581
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4782
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Introduce AFR_CALL_RESUME macro which cleans up frame->local, like
how AFR_STACK_UNWIND etc. do.
Therefore fix leak in afr_fsync() path.
Change-Id: I3855d8e7e84dbc44e05f507563b7f722bf9621b8
BUG: 927146
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4745
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
AFR now provides a stronger guarantee that fsync() returns only
after completely finishing all the deferred/delayed POST-OP on that
open file.
To acheive this we make a stub out of the returning fsync and
register it with the "delayed" frame in afr_changelog_wake_resume().
The delayed frame, after getting woken up and finishing the POST-OP
will call_resume() the registered stub (which UNWINDs the fsync) at
the time of frame destruction.
This provides a guarantee that an application's (or FUSE) fsync()
returns only after finishing up all the previous transactions,
including delayed POST-OPs and UNLOCK.
Change-Id: Iaa955457e2f25088a144fde37ad0444277b5cf49
BUG: 927146
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4737
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The changelogging scheme of AFR stores information about the state
of all replicas in all replicas (in the extended attribute of the
respective files on each server) in the form of 'pending counts'
of operations (effectively "dirty flags"). These xattrs are blindly
trusted while performing self-heal, and therefore utmost care has
to be taken while updating and maintaing them.
The most critical updation is the clearing of the pending counts
corresponding to the *other* server in the changelog of a given
server. Before clearing the pending count, we need durability
guarantee of the write which was performed on the other server.
To obtain such a guarantee, it may be necessary to explicitly
introduce an fsync() phase (if the file itself wasn't already
opened with O_SYNC).
This patch introduces the detection of unstable stable writes on
a file and issues explicit fsync() on the servers before performing
the POST-OP clearing of pending flags.
Change-Id: I2171b86a74ec91e40e5877eef0a4e7379578ecf7
BUG: 927146
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4721
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Reviewed-by: Krishnan Parthasarathi <kparthas@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before Anonymous fds are available, afr had to queue up
transactions if the file is not opened on one of its
subvolumes. This happens until the attempt to open the
file either succeeds or fails. These attempts happen
until the file is successfully opened on the subvolume.
Now client xlator uses anonymous fds to perform the fops
if the fd used for the fop is not 'opened'.
Fops will be successful even when the file is not opened
so there is no need to queue up the transactions anymore in afr.
Open is attempted on the subvolume where it is not
opened independent of the fop.
Change-Id: Id1a4b4ebe6f89f9efe8f6a8247918b91247d0819
BUG: 913051
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4568
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
fd based operations such as readv checked only for data split brain
instead of complete split-brain (i.e both data + metadata) assuming that
open would have done the complete split-brain check. However open-behind
would have unwound open, without winding to afr thus preventing the complete
split-brain check and some appliations will be able to read the contents
of the file even though the file has metadata split-brain. So let all
the fd based fops do a defensive check of complete split-brain.
Change-Id: Ia90b35f2b08426dfcad804b7f8105278c86fbd2d
BUG: 846240
Signed-off-by: Raghavendra Bhat <raghavendra@redhat.com>
Reviewed-on: http://review.gluster.org/4548
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
| |
Change-Id: I7049c0c64e36a9dfa4cc0e0b34de7ec111d2f6c1
BUG: 908302
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4076
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As of http://review.gluster.org/2828, the blocking lock code
path's condition for checking completion of locking atempt is
broken. The condition -
if ((child_index == priv->child_count) || ...)
and
if ((child_index == priv->child_count) && ...)
which is retained to check completion of blocking lock attempts
for DATA/METADATA transaction will _always_ fail because a few
lines above we have -
child_index = cookie % priv->child_count;
So child_index will never equal priv->child_count. This leaves
the correctness at the mercy of the next part of the
conditional -
.. (int_lock->lock_count == int_lock->lk_expected_count) ..
This "works" as long as no server went down during the transaction.
If a server goes down in the middle of the transaction, then this
condition also fails, and the code wraps around and starts a
blocking lock attempt loop all the way again from from the first server.
This results in double locks getting acquired on those servers, and
eventually the second condition gets hit (first condition is _never_
hit) and we come out of locking phase.
During unlock phase we perform only one unlock per server leaving the
other lock "leaked" forever.
Change-Id: I7189cdf3f70901b04647516fe1d1e189f36cc8dd
BUG: 765564
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4433
Reviewed-by: Krishnan Parthasarathi <kparthas@redhat.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* There are upto 3 entry lockees that may be needed to perform
entrylk'ing in posix dir-write operations.
* For eg, rmdir ("/a/b") needs to acquire locks on two entities,
- entrylk ("/a", "b")
- entrylk ("/a/b", null)
* Changed existing entrylk/rename/selfheal (entrylk) transactions
to use the new book-keeping structures
* Fixed few issues in afr_trace_entry_lk{in,out} functions. Tracing is now
aware of the new entry lockee structure.
Implementation notes:
* Changed 'cookie' sent in stack_wind to encode lockee_entity_no
and subvol_no.
cookie is a non-negative integer such that 0 <= cookie < replica_count,
When more than one lock is being acquired across the subvolumes,
cookie % replica_count gives the subvol_no
cookie / replica_count gives the lockee_entity_no.
Change-Id: Idbf41803387a7d59a0f7fcb1453d91cea74da153
BUG: 765564
Signed-off-by: Krishnan Parthasarathi <kp@gluster.com>
Reviewed-on: http://review.gluster.org/2828
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
Afr prevents opens on a file in split-brian but the
fd that is already open still has the capability to perform
both reads and writes to the file.
Fix:
Fail readvs on a file with EIO.
Change-Id: I8e07f24c36fab800499b36ab374f984b743332cd
BUG: 873962
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4199
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The most important errno logic historically only prioritized ESTALE
over ENOENT. Commit c8c0942d added EIO prioritization over ENOENT
to ensure that split-brain was reported when it occurs in
conjunction with bricks missing the file entry. The unintended side
effect of this change is that (non split-brain) EIO errors reported
from the bricks themselves are now reported to the client when the
expectation is that afr should squash said errors in favor of
marking the file inconsistent.
The high-level problem is that EIO is overloaded with different
meanings from different contexts. This commit adds an eio parameter
to the errno priority logic to conditionally flag when EIO is of
higher priority and should be propagated to the client.
BUG: 892730
Change-Id: Ib692a8a1f1737ef190d57894f392ec53ffb33aab
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4376
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
afr_more_important_error() is written to return whether a new errno
should override an existing errno for high-level operations that
could span multiple sub-operations. It specifically prioritizes
ESTALE over EIO over ENOENT, and otherwise defaults to the latest
error passed having priority.
This change preserves current behavior, but rewrites the logic to
return the higher priority error of the existing and new errno. The
purpose of the change is to make the logic a bit more clear and set
the stage for future changes to make the logic flexible based on
context.
BUG: 892730
Change-Id: Id1aa48855dfb0507abc9d1ef22f2259b30472576
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4375
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Along with this change, fixed the race of setting the
split-brain status in inode-ctx after unwinding the fop from
self-heal in case of back-ground self-heal.
Change-Id: Ifc829300df485f50f139443802e8b6dc7038b4ad
BUG: 873962
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4198
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
When create/mknod fails on some of the nodes, appropriate pending
data/metadata changelogs are not assigned. This was not considered
to be an issue because entry self-heal would do the assigning of
appropriate changelog after creating new entries. But using
the combination of rebalance and remove brick we can construct a
case where a file with same name and gfid can be created in a dir
with different data and link-to xattr without any changelog.
Fix:
When a create/mknod failure is observed mark the appropriate
changelog on the new file created.
Change-Id: I4c32cbf5594a13fb14deaf97ff30b2fff11cbfd6
BUG: 858212
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4207
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Introduce a block flag to support an optional blocking or
non-blocking mode in the self-heal data locking mechanism. All
callers are modified to use blocking mode, which is the current
default behavior (no change in behavior is introduced by this
commit).
BUG: 874045
Change-Id: Ib7ff9984578fa11de4e3b6981508100cdddd37cd
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4257
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In a replica pair unlike files, directories may not have their
content in same order, so readdir for same (offset, size) may
not give same entries on both the sobvolumes of replica pair.
Switching over from one subvolume to another may not be a good
idea sometimes. It may lead to duplicate entries or fewer entries
or both. This patch provides a way to disable readdir-failover
so that applications like rebalance can retry if they want to.
Change-Id: I2b23eb224a2e84016a561362932613ac824c11a0
BUG: 859387
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4159
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current failure to handle short writes on writev fops leaves
us open to file corruption. A short write on a user request is
ignored and leaves replicas in an inconsistent state. A short write
during a self-heal is ignored and incorrectly marks the files as
consistent if the heal completes.
Modify user writev handling to return the best case return value
from each of the replicas. Short writes that occur relative to this
value are marked as failed and will require a heal. Modify
self-heal to set an error on a short write and abort the heal.
BUG: 853690
Change-Id: I18b30f58702326249230eeebb361b29e40b535f5
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4150
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
Whenever gluster volume heal vol full command is executed, the entries
stored in the circual buffer for sh->healed are added in the dictionary
in the _crawl_post_sh_action function irrespective of whether actual self heal
(due to non-zero values in chage log) takes place or not.
Fix:
Value of key (actual-sh-done) will be set to 1 whenever self heal takes place
due to non-zero change log values and if for some FOP self heal daemon finds
that no self heal required after examining the pending matrix, the value will
be 0.
Change-Id: I11fd0b9ee76759af17c5bca6bfafbaf66bcaacbc
BUG: 863068
Signed-off-by: Venkatesh Somyajula <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/4181
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem:
Entry self-heal does lookups on all the entries that are read
in readdir. More the size of readdir more number of lookups happen
in parallel. It is observed that it leads to HUGE cpu spikes
rendering everything else on the system unusable.
Fix:
Provided the option self-heal-readdir-size to configure the size.
Default value is at 1KB.
Tests:
Checked that the readdirs are happening with the configured value
in entry-self-heal.
Change-Id: Icaa937ad88857e6f9a12375b1e7f6a49192bc8b1
BUG: 860895
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.org/4002
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RCA:
In case of dir fops create, mknod, mkdir, link, symlink, rename
if the fop fails on read-child then unwinds are happening with
all-zero pre/post iatt-bufs. The bug occurs because the parent
bufs are not saved if the response is not from read-child.
Fix:
Save the pre/post-bufs for the first response. If the response
comes from read-child, overwrite whatever we have cached.
Tests:
Attached the mount process to gdb.
Tested that the unwinds happen with proper pre/post iatt bufs in
the following cases:
1) All success case
2) Failure on read-child
3) Failure on non-read-child
4) Failure on all children.
Tested soft-link self-heal to test the change made in that.
Tested errno ENOTEMPTY for rmdir, rename fops.
Change-Id: I82882423d2d766b4f4a3044203bcb5dbcaee1755
BUG: 845242
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3775
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RCA
The bug is observed because the decision to mark
a file in split-brain is taken outside appropriate locks.
Lookup gathers xattrs outside any lock. The xattrs being
in split-brain in lookup should only be taken as a hint.
Appropriate inodelks should be taken before confirming
a split-brain. Self-heal confirms this at the moment.
If data/metadata self-heal is turned off, inspecting of
xattrs could not be performed so split-brain behavior
does not work correctly if the self-heal options are turned off.
Fix
Self-heals are launched to inspect xattrs even when the
data/metadata self-heal options are turned off. The decision
to heal data/metadata after the xattrs are inspected is based
on whether the options are turned on/off. So decision to set/reset
split-brain flag is taken inside appropriate locks.
Testcases:
tests 33-36 in
https://github.com/pranithk/gluster-tests/blob/master/afr/self-heal.sh
Change-Id: Ia8aeab08208b50c06609ad35a9d72f3d553ee343
BUG: 833727
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3626
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
post-op-delay introduces an artificial delay between the OP and
POST-OP-CHANGELOG phases of a write transaction to increase the
probability of changelog-piggyback and eager-locking to work
more efficiently.
Also enable eager-locking by default.
Change-Id: I865ca4b68512c44818719c7e388952f15d53e6c2
BUG: 836033
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.com/3621
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pranithk@gluster.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Historically PID (frame->root->pid) was used by the locks translator
to identify a locker (and make decisions about which locks contend
or cooperate/merge). Since the introduction of lock_owner parameter
the usage of PID (for locks) was deprecated and is now unused. This
patch nukes the usage of PID in AFR
The usage of lk_owner has also ended up being a mess, because of the
differentiation required between ->lk() and ->inodelk(), (->lk() needs
to be identified by the process (roughly) and ->inodelk() needs to be
identified by the transaction) and also because of optimizations like
eager locking (locks are no more identified by the transaction as they
now get inherited by the next transaction).
The scheme (and technique) now is:
- All FOPs (the third phase of the transaction) happen with the lk_owner
which is set by the topmost layer (FUSE, NFS etc.)
- All entrylks are issued with lk_owner set to the frame->root address.
- Inodelks which will not be subject to eager locking are issued with
lk_owner set to frame->root.
- Inodelks which are subject to eager locking are issued with lk_owner
set to the address of fd_t (which are the only type of frames which
get subject to the eager locking optimization)
- At the start of the transaction, the transaction frame's lk_owner is
set to the either frame->root or fd_t (and never unmodified) depending
on the type of transaction.
- Just before the third phase (FOP phase) the set lk_owner is "saved"
away and overwritten by the lk_owner submitted by the top layer (FUSE
or NFS)
- Right after the third phase, the saved lk_owner is "restored" to resume
the transaction into the POST-OP and eventually UNLOCK using the same
lk_owner which was used during the LOCK phase.
Change-Id: I6ab8e4d6b65ae4185fa85ad3fded8e9188b2f929
BUG: 836033
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.com/3620
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pranithk@gluster.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Controlled by the "choose-local" option (on by default).
Change-Id: I560f27c81703f2c9c62fdb51532c8eb763826df7
BUG: 806462
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: http://review.gluster.com/3005
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Both the first-to-respond method and the round-robin method are susceptible
to clients repeatedly choosing the same servers across a series of opens,
creating hot spots. Also, the code to handle a replica being down will
ignore both methods and just choose the first remaining (which is not an
issue for two-way but can be otherwise). The hashed method more reliably
avoids such hot spots. There are three values/modes.
0: use the old (broken) methods.
1: select a read-child based on a hash of the file's GFID, so all clients
will choose the same subvolume for a file (ensuring maximum consistency)
but will distribute load for a set of files.
2: select a read-child based on a hash of the file's GFID plus the client's
PID, so different children will distribute load even for one file.
Mode 2 will probably be optimal for most cases. Using response time when we
open the file is problematic, both because a single sample might not have
been representative even then and because load might have shifted in the
hours or days since (for long-lived files). Trying to use more current load
information can lead to "herd following" behavior which is just as bad.
Pseudo-random distribution is likely to be the best we can reasonably do,
just as it is for DHT.
Change-Id: I798c2760411eacf32e82a85f03bb7b08a4a49461
BUG: 802513
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: http://review.gluster.com/2926
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The xattrop order in pre/post op on all the subvols
is client-0, client-1... client-n where n is (replica-count - 1).
This order can lead to invalid split-brains if the brick
dies in the middle of xattrops.
Example: transaction completed pre-op, so on all the subvolumes
xattrs have '1' changelog. Now post-op is sent to both the subvols.
On subvol-0 change-log of client-0 is decremented to 0, before
decrementing change-log of client-1 to 0 the brick dies.
This change-log status on subvol-0 gives the meaning that a
change is done on subvol-0 successfully but on subvol-1 it failed.
Which is not what happened.
Changes done when the subvol-0 was down will lead to pending
change-log on subvol-1 for subvol-0. Which is correct.
When the subvol-0 is brought back up, the change-log will be in
split-brain state even when it is not a legitimate split-brain.
If the brick dies in the middle of xattrops it should remain fool.
Pre-op should perform xattrop of the local change-log first and
post-op should perform xattrop of the local change-log last.
In case of optimistic changelogs txn_changelog should be done
last on local if it succeeds, first if it fails.
Change-Id: Ib6eeb20cdc49b0b1fd2f454f25a9c8e08388c6e7
BUG: 765194
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3226
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|
|
|
|
|
|
|
|
|
| |
Change-Id: Ifab37db2af8d489cd516e992b7423c765dcabc4f
BUG: 765587
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3088
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
|