Bug 1292046

Summary:	Renames/deletes failed with "No such file or directory" when few of the bricks from the hot tier went offline
Product:	[Community] GlusterFS	Reporter:	Krutika Dhananjay <kdhananj>
Component:	replicate	Assignee:	Krutika Dhananjay <kdhananj>
Status:	CLOSED CURRENTRELEASE	QA Contact:
Severity:	high	Docs Contact:
Priority:	unspecified
Version:	3.7.6	CC:	bugs, kdhananj, spandura, storage-qa-internal
Target Milestone:	---	Keywords:	Triaged
Target Release:	---
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:	glusterfs-3.7.7	Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:	1291701	Environment:
Last Closed:	2016-04-19 07:51:02 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:	1291560, 1291701
Bug Blocks:

Description Krutika Dhananjay 2015-12-16 10:36:25 UTC

+++ This bug was initially created as a clone of Bug #1291701 +++

+++ This bug was initially created as a clone of Bug #1291560 +++

Description of problem:
===========================
On a tiered volume, with 2x2 cold tier and 2x3 dis-rep hot tier was performing 
renames|deletes on files/dirs. When the bricks went offline, the renames/deletes  failed with "No such file or directory". The bricks from cold tier were all online as quorum was set. Only one of the bricks from the each sub-volume of the hot-tier went offline. 

Version-Release number of selected component (if applicable):
===============================================================
glusterfs 3.7.5 built on Dec  3 2015 11:30:45


How reproducible:
====================
Often

Steps to Reproduce:
======================
1. Create 2x2 dis-rep cold-tier and 2x3 dis-rep hot-tier volume. Start the volume. Mount the volume. 

2. From mount, create files/dirs. 

3. rename few of the files/dirs created to different name

4. While rename is in progress, crash the mounts of the bricks using "godown" utility.(available in xfsprogs). 


Actual results:
================
On mount, the rename fails with "No such file or directory"

Expected results:
====================
Renames/deletes shouldn't fail

Additional info:
===================
 
Volume Name: testvol
Type: Tier
Volume ID: 5a2f042d-ee04-4b3d-b5d5-d36e29cea325
Status: Started
Number of Bricks: 10
Transport-type: tcp
Hot Tier :
Hot Tier Type : Distributed-Replicate
Number of Bricks: 2 x 3 = 6
Brick1: rhsauto020.lab.eng.blr.redhat.com:/bricks/brick2/testvol_tier5
Brick2: rhsauto017.lab.eng.blr.redhat.com:/bricks/brick2/testvol_tier4
Brick3: rhsauto038.lab.eng.blr.redhat.com:/bricks/brick1/testvol_tier3
Brick4: rhsauto021.lab.eng.blr.redhat.com:/bricks/brick1/testvol_tier2
Brick5: rhsauto020.lab.eng.blr.redhat.com:/bricks/brick1/testvol_tier1
Brick6: rhsauto017.lab.eng.blr.redhat.com:/bricks/brick1/testvol_tier0
Cold Tier:
Cold Tier Type : Distributed-Replicate
Number of Bricks: 2 x 2 = 4
Brick7: rhsauto017.lab.eng.blr.redhat.com:/bricks/brick0/testvol_brick0
Brick8: rhsauto020.lab.eng.blr.redhat.com:/bricks/brick0/testvol_brick1
Brick9: rhsauto021.lab.eng.blr.redhat.com:/bricks/brick0/testvol_brick2
Brick10: rhsauto038.lab.eng.blr.redhat.com:/bricks/brick0/testvol_brick3
Options Reconfigured:
diagnostics.brick-log-level: DEBUG
diagnostics.client-log-level: DEBUG
performance.readdir-ahead: on
features.ctr-enabled: on
cluster.tier-mode: cache
cluster.watermark-low: 75
cluster.watermark-hi: 90

Error messages seen in client log:
=================================
[2015-12-14 07:45:12.156546] E [MSGID: 114031] [client-rpc-fops.c:251:client3_3_mknod_cbk] 0-testvol-client-9: remote operation failed. Path: (null) [Input/output error]
[2015-12-14 07:45:12.159452] W [MSGID: 114031] [client-rpc-fops.c:664:client3_3_unlink_cbk] 0-testvol-client-0: remote operation failed [Device or resource busy]
[2015-12-14 07:45:12.159480] W [MSGID: 114031] [client-rpc-fops.c:664:client3_3_unlink_cbk] 0-testvol-client-1: remote operation failed [Device or resource busy]
[2015-12-14 07:45:12.160293] I [MSGID: 109069] [dht-common.c:1159:dht_lookup_unlink_stale_linkto_cbk] 0-testvol-tier-dht: Returned with op_ret -1 and op_errno 16 for /E_file_44

--- Additional comment from Vijay Bellur on 2015-12-15 08:28:17 EST ---

REVIEW: http://review.gluster.org/12973 (cluster/afr: During name heal, propagate EIO only on gfid or type mismatch) posted (#1) for review on master by Krutika Dhananjay (kdhananj)

--- Additional comment from Vijay Bellur on 2015-12-16 05:33:42 EST ---

COMMIT: http://review.gluster.org/12973 committed in master by Pranith Kumar Karampuri (pkarampu) 
------
commit 7ba6469eee3118cc4ece905d2538ef778320ae63
Author: Krutika Dhananjay <kdhananj>
Date:   Tue Dec 15 18:48:20 2015 +0530

    cluster/afr: During name heal, propagate EIO only on gfid or type mismatch
    
    When the disk associated with a brick returns EIO during lookup, chances are
    that name heal would return an EIO because one of the syncop_XXX() operations
    as part of it returned an EIO. This is inherently treated by afr_lookup_selfheal_wrap()
    as a split-brain and the lookup is aborted prematurely with EIO even if it
    succeeded on the other replica(s).
    
    Change-Id: Ib9b7f2974bff8e206897bb4f689f0482264c61e5
    BUG: 1291701
    Signed-off-by: Krutika Dhananjay <kdhananj>
    Reviewed-on: http://review.gluster.org/12973
    Tested-by: NetBSD Build System <jenkins.org>
    Tested-by: Gluster Build System <jenkins.com>
    Reviewed-by: Pranith Kumar Karampuri <pkarampu>

Comment 1 Vijay Bellur 2015-12-16 10:38:18 UTC

REVIEW: http://review.gluster.org/12979 (cluster/afr: During name heal, propagate EIO only on gfid or type mismatch) posted (#1) for review on release-3.7 by Krutika Dhananjay (kdhananj)

Comment 2 Vijay Bellur 2015-12-17 12:12:12 UTC

REVIEW: http://review.gluster.org/12979 (cluster/afr: During name heal, propagate EIO only on gfid or type mismatch) posted (#2) for review on release-3.7 by Krutika Dhananjay (kdhananj)

Comment 3 Vijay Bellur 2015-12-18 08:54:00 UTC

COMMIT: http://review.gluster.org/12979 committed in release-3.7 by Pranith Kumar Karampuri (pkarampu) 
------
commit 1f93e47c8777f7df3e8b1e1a6f42a4c33cb1c513
Author: Krutika Dhananjay <kdhananj>
Date:   Tue Dec 15 18:48:20 2015 +0530

    cluster/afr: During name heal, propagate EIO only on gfid or type mismatch
    
            Backport of: http://review.gluster.org/#/c/12973/
    
    When the disk associated with a brick returns EIO during lookup, chances are
    that name heal would return an EIO because one of the syncop_XXX() operations
    as part of it returned an EIO. This is inherently treated by afr_lookup_selfheal_wrap()
    as a split-brain and the lookup is aborted prematurely with EIO even if it
    succeeded on the other replica(s).
    
    Change-Id: I754fa59c585712b8037f98a8c3c1737a2167fa1b
    BUG: 1292046
    Signed-off-by: Krutika Dhananjay <kdhananj>
    Reviewed-on: http://review.gluster.org/12979
    Tested-by: NetBSD Build System <jenkins.org>
    Tested-by: Gluster Build System <jenkins.com>
    Reviewed-by: Pranith Kumar Karampuri <pkarampu>

Comment 4 Kaushal 2016-04-19 07:51:02 UTC

This bug is getting closed because a release has been made available that should address the reported issue. In case the problem is still not fixed with glusterfs-3.7.7, please open a new bug report.

glusterfs-3.7.7 has been announced on the Gluster mailinglists [1], packages for several distributions should become available in the near future. Keep an eye on the Gluster Users mailinglist [2] and the update infrastructure for your distribution.

[1] https://www.gluster.org/pipermail/gluster-users/2016-February/025292.html
[2] http://thread.gmane.org/gmane.comp.file-systems.gluster.user