|
 |
d1681e |
From 737d077a44899f6222822408c400fcd91939ca5b Mon Sep 17 00:00:00 2001
|
|
|
d1681e |
From: Kotresh HR <khiremat@redhat.com>
|
|
|
d1681e |
Date: Thu, 12 Jul 2018 04:31:01 -0400
|
|
|
d1681e |
Subject: [PATCH 321/325] geo-rep: Fix symlink rename syncing issue
|
|
|
d1681e |
|
|
|
d1681e |
Problem:
|
|
|
d1681e |
Geo-rep sometimes fails to sync the rename of symlink
|
|
|
d1681e |
if the I/O is as follows
|
|
|
d1681e |
|
|
|
d1681e |
1. touch file1
|
|
|
d1681e |
2. ln -s "./file1" sym_400
|
|
|
d1681e |
3. mv sym_400 renamed_sym_400
|
|
|
d1681e |
4. mkdir sym_400
|
|
|
d1681e |
|
|
|
d1681e |
The file 'renamed_sym_400' failed to sync to slave
|
|
|
d1681e |
|
|
|
d1681e |
Cause:
|
|
|
d1681e |
Assume there are three distribute subvolume (brick1, brick2, brick3).
|
|
|
d1681e |
The changelogs are recorded as follows for above I/O pattern.
|
|
|
d1681e |
Note that the MKDIR is recorded on all bricks.
|
|
|
d1681e |
|
|
|
d1681e |
1. brick1:
|
|
|
d1681e |
-------
|
|
|
d1681e |
|
|
|
d1681e |
CREATE file1
|
|
|
d1681e |
SYMLINK sym_400
|
|
|
d1681e |
RENAME sym_400 renamed_sym_400
|
|
|
d1681e |
MKDIR sym_400
|
|
|
d1681e |
|
|
|
d1681e |
2. brick2:
|
|
|
d1681e |
-------
|
|
|
d1681e |
|
|
|
d1681e |
MKDIR sym_400
|
|
|
d1681e |
|
|
|
d1681e |
3. brick3:
|
|
|
d1681e |
-------
|
|
|
d1681e |
|
|
|
d1681e |
MKDIR sym_400
|
|
|
d1681e |
|
|
|
d1681e |
The operations on 'brick1' should be processed sequentially. But
|
|
|
d1681e |
since MKDIR is recorded on all the bricks, The brick 'brick2/brick3'
|
|
|
d1681e |
processed MKDIR first before 'brick1' causing out of order syncing
|
|
|
d1681e |
and created directory sym_400 first.
|
|
|
d1681e |
|
|
|
d1681e |
Now 'brick1' processed it's changelog.
|
|
|
d1681e |
|
|
|
d1681e |
CREATE file1 -> succeeds
|
|
|
d1681e |
SYMLINK sym_400 -> No longer present in master. Ignored
|
|
|
d1681e |
RENAME sym_400 renamed_sym_400
|
|
|
d1681e |
While processing RENAME, if source('sym_400') doesn't
|
|
|
d1681e |
present, destination('renamed_sym_400') is created. But
|
|
|
d1681e |
geo-rep stats the name 'sym_400' to confirm source file's
|
|
|
d1681e |
presence. In this race, since source name 'sym_400' is
|
|
|
d1681e |
present as directory, it doesn't create destination.
|
|
|
d1681e |
Hence RENAME is ignored.
|
|
|
d1681e |
|
|
|
d1681e |
Fix:
|
|
|
d1681e |
The fix is not rely only on stat of source name during RENAME.
|
|
|
d1681e |
It should stat the name and if the name is present, gfid should
|
|
|
d1681e |
be same. Only then it can conclude the presence of source.
|
|
|
d1681e |
|
|
|
d1681e |
>upstream patch : https://review.gluster.org/#/c/20496/
|
|
|
d1681e |
|
|
|
d1681e |
Backport of:
|
|
|
d1681e |
> BUG: 1600405
|
|
|
d1681e |
> Change-Id: I9fbec4f13ca6a182798a7f81b356fe2003aff969
|
|
|
d1681e |
> Signed-off-by: Kotresh HR <khiremat@redhat.com>
|
|
|
d1681e |
|
|
|
d1681e |
BUG: 1601314
|
|
|
d1681e |
Change-Id: I9fbec4f13ca6a182798a7f81b356fe2003aff969
|
|
|
d1681e |
Signed-off-by: Kotresh HR <khiremat@redhat.com>
|
|
|
d1681e |
Reviewed-on: https://code.engineering.redhat.com/gerrit/144104
|
|
|
d1681e |
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>
|
|
|
d1681e |
Tested-by: RHGS Build Bot <nigelb@redhat.com>
|
|
|
d1681e |
---
|
|
|
d1681e |
geo-replication/syncdaemon/resource.py | 11 +++++++++--
|
|
|
d1681e |
1 file changed, 9 insertions(+), 2 deletions(-)
|
|
|
d1681e |
|
|
|
d1681e |
diff --git a/geo-replication/syncdaemon/resource.py b/geo-replication/syncdaemon/resource.py
|
|
|
d1681e |
index 00e62b7..0d5462a 100644
|
|
|
d1681e |
--- a/geo-replication/syncdaemon/resource.py
|
|
|
d1681e |
+++ b/geo-replication/syncdaemon/resource.py
|
|
|
d1681e |
@@ -674,8 +674,14 @@ class Server(object):
|
|
|
d1681e |
collect_failure(e, EEXIST)
|
|
|
d1681e |
elif op == 'RENAME':
|
|
|
d1681e |
en = e['entry1']
|
|
|
d1681e |
- st = lstat(entry)
|
|
|
d1681e |
- if isinstance(st, int):
|
|
|
d1681e |
+ # The matching disk gfid check validates two things
|
|
|
d1681e |
+ # 1. Validates name is present, return false otherwise
|
|
|
d1681e |
+ # 2. Validates gfid is same, returns false otherwise
|
|
|
d1681e |
+ # So both validations are necessary to decide src doesn't
|
|
|
d1681e |
+ # exist. We can't rely on only gfid stat as hardlink could
|
|
|
d1681e |
+ # be present and we can't rely only on name as name could
|
|
|
d1681e |
+ # exist with differnt gfid.
|
|
|
d1681e |
+ if not matching_disk_gfid(gfid, entry):
|
|
|
d1681e |
if e['stat'] and not stat.S_ISDIR(e['stat']['mode']):
|
|
|
d1681e |
if stat.S_ISLNK(e['stat']['mode']) and \
|
|
|
d1681e |
e['link'] is not None:
|
|
|
d1681e |
@@ -699,6 +705,7 @@ class Server(object):
|
|
|
d1681e |
[ENOENT, EEXIST], [ESTALE])
|
|
|
d1681e |
collect_failure(e, cmd_ret)
|
|
|
d1681e |
else:
|
|
|
d1681e |
+ st = lstat(entry)
|
|
|
d1681e |
st1 = lstat(en)
|
|
|
d1681e |
if isinstance(st1, int):
|
|
|
d1681e |
rename_with_disk_gfid_confirmation(gfid, entry, en)
|
|
|
d1681e |
--
|
|
|
d1681e |
1.8.3.1
|
|
|
d1681e |
|