Re: Self Heal Confusion

Ashish Pandey <aspandey@xxxxxxxxxx> · Fri, 28 Dec 2018 01:00:27 -0500 (EST)

Hi Brett,

First the answers of all your questions - 

1.  If a self-heal deamon is listed on a host (all of mine show one with 
a volume status command) can I assume it's enabled and running?

For your volume, projects self heal daemon is UP and running

2.  I assume the volume that has all the self-heals pending has some 
serious issues even though I can access the files and directories on 
it.  If self-heal is running shouldn't the numbers be decreasing?

It should heal the entries and the number of entries coming in "gluster v heal volname info" command should be decreasing.

It appears to me self-heal is not working properly so how to I get it to 
start working or should I delete the volume and start over?

As you can access all the files from mount point, I think the volume and the files are in good state as of now.
I don't think you should think of deleting your volume before trying to fix it.
If there is no fix or the fix is taking time you can go ahead with that option.

-----------------------
Why all these options are off? 

performance.quick-read: off
 performance.parallel-readdir: off
 performance.readdir-ahead: off
 performance.write-behind: off
 performance.read-ahead: off

Although this should not matter to your issue but I think you should enable all the above unless you have a reason to not to do so.
--------------------

I would like you to perform following steps and provide some more information - 

1 - Try to restart self heal and see if that works. 
"gluster v start volume force" will kill and restart the self heal processes.

2 - If step 1 is not fruitful, get the list of entries need to be healed and pick one of the entry to heal. I mean we should focus on one entry to find out why it is 
not getting healed instead of all the 5900 entries. Let's call it entry1.

3 -  Now access the entry1 from mount point, read, write on it and see if this entry has been healed. Check for heal info. Accessing file from mount point triggers client side heal
which could also heal the file.

4 - Check for the logs in /var/log/gluster, mount logs and glustershd logs should be checked and provided.

5 -  Get the external attributes of entry1 from all the bricks.

If the path of the entry1 on mount point is /a/b/c/entry1 then you have to run following command on all the nodes - 

getfattr -m. -d -e hex <path of the brick on the node>/a/b/c/entry1

Please provide the output of above command too.

---
Ashish

From: "Brett Holcomb" <biholcomb@xxxxxxxxxx>
To: gluster-users@xxxxxxxxxxx
Sent: Friday, December 28, 2018 3:49:50 AM
Subject: Re:  Self Heal Confusion

Resend as I did not reply to the list earlier.  TBird responded to the poster and not the list.
On 12/27/18 11:46 AM, Brett Holcomb wrote:
Thank you. I appreciate the help  Here is the information.  Let me know if you need anything else.  I'm fairly new to gluster.
Gluster version is 5.2
1. gluster v info
Volume Name: projects
 Type: Distributed-Replicate
 Volume ID: 5aac71aa-feaa-44e9-a4f9-cb4dd6e0fdc3
 Status: Started
 Snapshot Count: 0
 Number of Bricks: 2 x 3 = 6
 Transport-type: tcp
 Bricks:
 Brick1: gfssrv1:/srv/gfs01/Projects
 Brick2: gfssrv2:/srv/gfs01/Projects
 Brick3: gfssrv3:/srv/gfs01/Projects
 Brick4: gfssrv4:/srv/gfs01/Projects
 Brick5: gfssrv5:/srv/gfs01/Projects
 Brick6: gfssrv6:/srv/gfs01/Projects
 Options Reconfigured:
 cluster.self-heal-daemon: enable
 performance.quick-read: off
 performance.parallel-readdir: off
 performance.readdir-ahead: off
 performance.write-behind: off
 performance.read-ahead: off
 performance.client-io-threads: off
 nfs.disable: on
 transport.address-family: inet
 server.allow-insecure: on
 storage.build-pgfid: on
 changelog.changelog: on
 changelog.capture-del-path: on

2.  gluster v status
Status of volume: projects
 Gluster process                             TCP Port  RDMA Port  Online  Pid
 ------------------------------------------------------------------------------
 Brick gfssrv1:/srv/gfs01/Projects           49154     0          Y       7213 
 Brick gfssrv2:/srv/gfs01/Projects           49154     0          Y       6932 
 Brick gfssrv3:/srv/gfs01/Projects           49154     0          Y       6920 
 Brick gfssrv4:/srv/gfs01/Projects           49154     0          Y       6732 
 Brick gfssrv5:/srv/gfs01/Projects           49154     0          Y       6950 
 Brick gfssrv6:/srv/gfs01/Projects           49154     0          Y       6879 
 Self-heal Daemon on localhost               N/A       N/A        Y       11484
 Self-heal Daemon on gfssrv2                 N/A       N/A        Y       10366
 Self-heal Daemon on gfssrv4                 N/A       N/A        Y       9872 
 Self-heal Daemon on srv-1-gfs3.corp.l1049h.
 net                                         N/A       N/A        Y       9892 
 Self-heal Daemon on gfssrv6                 N/A       N/A        Y       10372
 Self-heal Daemon on gfssrv5                 N/A       N/A        Y       10761

 Task Status of Volume projects
 ------------------------------------------------------------------------------
 There are no active volume tasks
3. I've given the summary since the actual list for two volumes is around 5900 entries.
Brick gfssrv1:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 85
 Number of entries in heal pending: 85
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0

 Brick gfssrv2:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 0
 Number of entries in heal pending: 0
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0

 Brick gfssrv3:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 0
 Number of entries in heal pending: 0
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0

 Brick gfssrv4:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 0
 Number of entries in heal pending: 0
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0
Brick gfssrv5:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 58854
 Number of entries in heal pending: 58854
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0

 Brick gfssrv6:/srv/gfs01/Projects
 Status: Connected
 Total Number of entries: 58854
 Number of entries in heal pending: 58854
 Number of entries in split-brain: 0
 Number of entries possibly healing: 0
On 12/27/18 3:09 AM, Ashish Pandey wrote:
Hi Brett,

Could you please tell us more about the setup?

1 - Gluster v info
2 - gluster v status
3 - gluster v heal <volname> info

These are the very basic information to start with debugging or suggesting any workaround.
It should always be included when asking such questions on mailing list so that people can reply sooner. 

Note: Please hide IP address/hostname or any other information you don't want world to see.

---
Ashish

From: "Brett Holcomb" <biholcomb@xxxxxxxxxx>
 To: gluster-users@xxxxxxxxxxx
 Sent: Thursday, December 27, 2018 12:19:15 AM
 Subject: Re:  Self Heal Confusion

Still no change in the heals pending.  I found this reference, https://archive.fosdem.org/2017/schedule/event/glusterselinux/attachments/slides/1876/export/events/attachments/glusterselinux/slides/1876/fosdem.pdf, which mentions the default SELinux context for a brick and that internal operations such as self-heal, rebalance should be ignored. but they do not elaborate on what ignore means - is it just not doing self-heal or something else.
I did set SELinux to permissive and nothing changed.  I'll try setting the bricks to the context mentioned in this pdf and see what happens.

On 12/20/18 8:26 PM, John Strunk wrote:
Assuming your bricks are up... yes, the heal count should be decreasing.
There is/was a bug wherein self-heal would stop healing but would still be running. I don't know whether your version is affected, but the remedy is to just restart the self-heal daemon.
Force start one of the volumes that has heals pending. The bricks are already running, but it will cause shd to restart and, assuming this is the problem, healing should begin...

$ gluster vol start my-pending-heal-vol force

Others could better comment on the status of the bug.

-John

On Thu, Dec 20, 2018 at 5:45 PM Brett Holcomb <biholcomb@xxxxxxxxxx> wrote:
I have one volume that has 85 pending entries in healing and two more 
 volumes with 58,854 entries in healing pending.  These numbers are from 
 the volume heal info summary command.  They have stayed constant for two 
 days now.  I've read the gluster docs and many more.  The Gluster docs 
 just give some commands and non gluster docs basically repeat that.  
 Given that it appears no self-healing is going on for my volume I am 
 confused as to why.

 1.  If a self-heal deamon is listed on a host (all of mine show one with 
 a volume status command) can I assume it's enabled and running?

 2.  I assume the volume that has all the self-heals pending has some 
 serious issues even though I can access the files and directories on 
 it.  If self-heal is running shouldn't the numbers be decreasing?

 It appears to me self-heal is not working properly so how to I get it to 
 start working or should I delete the volume and start over?

 I'm running gluster 5.2 on Centos 7 latest and updated.

 Thank you.

 _______________________________________________
 Gluster-users mailing list
 Gluster-users@xxxxxxxxxxx
 https://lists.gluster.org/mailman/listinfo/gluster-users

 _______________________________________________
 Gluster-users mailing list
 Gluster-users@xxxxxxxxxxx
 https://lists.gluster.org/mailman/listinfo/gluster-users

_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users

_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users