Delta backup fails for specific vm with VDI chain error

olivierlambert

Don't do migrate in every direction, it's useless. This is not an error but a protection. You need to really understand what coalesce means.

First, it happens automatically in the host every time a snapshot is removed. Manually or by a XO backup job. XO doesn't have any info or control on that process. So it's not a XO "thing". In XO, we decided to control the VDI chain length to avoid having a chain growing faster than your SR can coalesce (hence the protection).

Coalesce isn't instantly done. You can see the disk needing coalesce in XO, SR view, advanced. Then you just have to wait.

Especially in your case, coalesce seems to run. So just be patient and check the "Advanced" tab. As soon it's empty, it means the chain is clean and ready for a next run.

In other word, your SR can't keep up with the pace of your backup schedule. Triggering the protection avoid the creation of a new snapshot and it's not a bad thing. It means it will be skipped until next time.

mbt

Hi Olivier,

thank your for taking the time to explain what's going on under the hood - very interesting stuff!
So I basically tried to fix a problem that needed no fixing and maybe thereby prolonged the whole process.

Is there a way to monitor the progress of a leaf-coalesce and is there a way to speed it up, other than shtting the VM down and doing it offline?
The delta backup job runs daily and I assumed the servers would be fast enough. They've recently been upgraded from 10k SAS drives to SSD. It's only SATA and just RAID 5, so I don't expect miracles but thus far it had been running well the past couple of years.

All stats are basically telling me the sr load is low - it rarely jumps to 200-300 IOPS, low single digit MiB/s, low 0.x/ IOwait, 37ms write latency, ...

Regards,
Alex

olivierlambert

No way to track the coalesce speed, there's absolutely no API whatsoever to monitor this You can have a rough idea by knowing your average time to coalesce (remove a snapshot and watch the Advanced view in XO). Bigger the snapshot (old or a lot of writes since then), longer the coalesce.
Coalesce is low prio to avoid competing with needed IOPS. However, IDK how they set this up. It's deep in the middle of OCaml code…

You can rescan the SR to try to trigger it faster. When it started, you'll have a vhd process running, so you can use htop to watch it and saw the priority and disk R/W of the process.

mbt

OCaml!? Really!? Brrrr....

OK, it seems there's a vhd-util process constantly consuming a single CPU core, rarely reading or writing any data at all. Probably going through larger chunks of data. After a couple of minutes the process ends and another one starts.

Let's assume it's doing its thing.

olivierlambert

Again, you can check the "result" on the "Advanced" tab of your SR in XO

Anonabhar

Hi mbt,

Recently I have had this problem a lot.

in XOA, bring up "Storages" and locate the SR. Click on Advanced and see if that VHD is showing up as needing to be coalessed. Does the depth oscillate between 1,2,1,2,1,2,etc.?

If so, shutdown the VM, and force a rescan of the SR.

There will be a 5 minute pause between the time the GC goes quiet to when it goes active. Let the system do a offline coalesce on the VHD and see if that clears it.

~Peg

Anonabhar

I also find:

tail -f /var/log/SMlog | grep SMGC

to be my friend... Worrying... But hypnotic to watch at 3AM 8-)

mbt

Hi Peg,

thanks for the information!
The situation stayed the same, so the online coalescing for this paricular vm still doesn't work right now and therefore it gets skipped by the daily delta backup job.
I think yesterday I saw the depth oscillate for a splitsecond down to 1 and then back to 2.

Unfortunately the VM is in access 24/7 by users in all kinds of time zones around the world. I'll have to schedule a maintenance window with the customer and hope this gets the job done. Not being able to backup a vm makes me nervous.

-alex

olivierlambert

The only reason I know for this case would be: creating more new blocks than the SR can coalesce. You can isolate this VM on another SR to see if the situation is improved.

mbt

Now something strange happened:

To have at least some sort of recent backup I started a vm copy to the other host. It is currently only 10% through but while it was working I saw some activity in the SMLog and one after another, the stalled coalesce jobs disappeared.

I'll wait for the copy to finish and see if backup is now working again.

I'm with you Olivier, on "creating more new blocks than the SR can coalesce", but there's only double digit or low triple digit IOPS on the sr, barely throughput, etc. So there's very low load. Whenever some of the other vms' vdis get coalesced or when they are backupped stats pick up and we get several thousand IOPS, etc.
That's why I'd say that bare metal performance can't be the issue here.

olivierlambert

It's not bare metal perf, it's how coalesce prio is set with hard limits in the code. Feel free to dig inside it, I never had the time to do it myself

Anonabhar

Hi mbt,

Sadly, when thing get that messy for me, I find that the only solution is to export the VM and then import. Thats one the (longest but most reliable) ways of de-leafing the tree.

Now, a trick I use, if downtime is an issue, is to do a Disaster Recovery backup to the same machine or SR. Once the backup is complete, I copy the MAC addresses for the network cards to the new VM (it will complain, but its OK), clear the blocked operations on the new VM, shutdown the original and then start the new.

If you use the above method, you can clear the blocked operation by using CLI:

xe vm-param-clear param-name=blocked-operations uuid=<NEW UUID>

If you are using the "enterprise" or open source version, you can eliminate the amount of data that will be missing from the point of starting the DR backup and the starting up of the new server by performing a Continuous Replication while the server is still running and after the replication is finished, shut the server down, do a second Continuous Replication to sync the images and then perform the steps I mentioned above.

If you do the above, your downtime is in minutes and not longer.

I hope that helps a bit,
Peg

olivierlambert

We are not aware of that kind of coalesce issues, what kind of SRs do you have?

Anonabhar

Hi Oliver,

The problem is of my own making and in no way blame XO for this. Its a perfect storm of events that caused my problem. Basically, I had a schedual that did delta backups to a local NFS repo. Later that night, I have 2 other jobs that do Cont Replication to 2 other DC's. Lastly, I have a monthly backup that spools to another NFS Repo that we backup to DLT tapes.. Basically what happened is some of the jobs got delayed finishing and they over-ran and were running on top of each other. The GC got overloaded and started to throw "Exception Timeout" and the situation just kept building and got worse. As a consequence, I started to get "Weird" things happening in the running VM's (long delays, sluggishness, crashing of daemons, etc..) Once I realized what was going on, it was to late and I am doing the above to about 200+ VM's to get it fixed.

We are using LLVMoiSCSI on Equal Logic PS6000's as the backend

If you want a laugh, this is what one of my VDI tree's looks like:

vhd=VHD-036594ed-6814-4afa-a063-c702d575316e capacity=21474836480 size=11580473344 hidden=1 parent=none
   vhd=VHD-0bf17661-b95f-4b5f-adf6-ef5b2ff0a039 capacity=21474836480 size=8388608 hidden=0 parent=VHD-036594ed-6814-4afa-a063-c702d575316e
   vhd=VHD-40df70e3-25b9-4875-a995-0f7497924afd capacity=21474836480 size=901775360 hidden=1 parent=VHD-036594ed-6814-4afa-a063-c702d575316e
      vhd=VHD-8b409ba8-77f4-4270-a36d-6d4c0091ed3c capacity=21474836480 size=8388608 hidden=0 parent=VHD-40df70e3-25b9-4875-a995-0f7497924afd
      vhd=VHD-5ec35406-34c0-4c9c-b92d-80df45f5e6c1 capacity=21474836480 size=1094713344 hidden=1 parent=VHD-40df70e3-25b9-4875-a995-0f7497924afd
         vhd=VHD-e039102b-ac33-487a-b38b-ccd60cf83573 capacity=21474836480 size=8388608 hidden=0 parent=VHD-5ec35406-34c0-4c9c-b92d-80df45f5e6c1
         vhd=VHD-ccdfe0c5-6044-4131-8ce5-3b2129f083c2 capacity=21474836480 size=843055104 hidden=1 parent=VHD-5ec35406-34c0-4c9c-b92d-80df45f5e6c
1
            vhd=VHD-26560aef-677c-498f-81e0-91cd791ec2ee capacity=21474836480 size=8388608 hidden=0 parent=VHD-ccdfe0c5-6044-4131-8ce5-3b2129f083
c2
            vhd=VHD-91abb814-214b-4fe7-aac6-20f0177348cb capacity=21474836480 size=977272832 hidden=1 parent=VHD-ccdfe0c5-6044-4131-8ce5-3b2129f0
83c2
               vhd=VHD-78cac789-3a46-4d79-87ff-0e8d28c17916 capacity=21474836480 size=8388608 hidden=0 parent=VHD-91abb814-214b-4fe7-aac6-20f0177
348cb
               vhd=VHD-aab84809-2165-4af3-9f47-a1b1ff4044f8 capacity=21474836480 size=1019215872 hidden=1 parent=VHD-91abb814-214b-4fe7-aac6-20f0
177348cb
                  vhd=VHD-cdb89a9e-a583-461c-adc6-7b68c0e88556 capacity=21474836480 size=8388608 hidden=0 parent=VHD-aab84809-2165-4af3-9f47-a1b1
ff4044f8
                  vhd=VHD-78d28f42-9327-440c-bd72-46b0716252a8 capacity=21474836480 size=268435456 hidden=1 parent=VHD-aab84809-2165-4af3-9f47-a1
b1ff4044f8
                     vhd=VHD-76360be2-10b7-4be1-8cb6-b2595ed8f5df capacity=21474836480 size=8388608 hidden=0 parent=VHD-78d28f42-9327-440c-bd72-4
6b0716252a8
                     vhd=VHD-492d0a9f-8a1d-469d-8bbc-9bbdd910a83e capacity=21474836480 size=994050048 hidden=1 parent=VHD-78d28f42-9327-440c-bd72
-46b0716252a8
                        vhd=VHD-9276a1dc-185b-4299-b958-415dc52ee2f8 capacity=21474836480 size=8388608 hidden=0 parent=VHD-492d0a9f-8a1d-469d-8bb
c-9bbdd910a83e
                        vhd=VHD-c5017f51-309a-4e49-bf42-ce6a838212d2 capacity=21474836480 size=1111490560 hidden=1 parent=VHD-492d0a9f-8a1d-469d-
8bbc-9bbdd910a83e
                           vhd=VHD-22a8f4fb-11ca-4caa-8439-ac8082482ca3 capacity=21474836480 size=8388608 hidden=0 parent=VHD-c5017f51-309a-4e49-
bf42-ce6a838212d2
                           vhd=VHD-b0151f77-e694-437b-8ff2-f352871ad37d capacity=21474836480 size=243269632 hidden=1 parent=VHD-c5017f51-309a-4e4
9-bf42-ce6a838212d2
                              vhd=VHD-25fffec1-fb4d-4717-8bff-c2050eb49aa2 capacity=21474836480 size=8388608 hidden=0 parent=VHD-b0151f77-e694-43
7b-8ff2-f352871ad37d
                              vhd=VHD-a0e6091b-3560-40bd-85ec-eb251e47366c capacity=21474836480 size=918552576 hidden=1 parent=VHD-b0151f77-e694-
437b-8ff2-f352871ad37d
                                 vhd=VHD-c40b1169-7d28-464c-9a6f-5309c7e2382d capacity=21474836480 size=8388608 hidden=0 parent=VHD-a0e6091b-3560
-40bd-85ec-eb251e47366c
                                 vhd=VHD-1038bb56-d4d9-49ef-b2fb-10319423cbd1 capacity=21474836480 size=264241152 hidden=1 parent=VHD-a0e6091b-35
60-40bd-85ec-eb251e47366c
                                    vhd=VHD-85b6bd0c-da46-41ab-8483-a831beeb0a0a capacity=21474836480 size=8388608 hidden=0 parent=VHD-1038bb56-d
4d9-49ef-b2fb-10319423cbd1
                                    vhd=VHD-52b51deb-1c26-4101-a6b0-c2fde959c66b capacity=21474836480 size=830472192 hidden=1 parent=VHD-1038bb56
-d4d9-49ef-b2fb-10319423cbd1
                                       vhd=VHD-4aa2d7d6-cf34-449d-8ee1-98e9b6a19bf5 capacity=21474836480 size=8388608 hidden=0 parent=VHD-52b51de
b-1c26-4101-a6b0-c2fde959c66b
                                       vhd=VHD-57304603-4a25-4723-93cf-24564d3e96f3 capacity=21474836480 size=264241152 hidden=1 parent=VHD-52b51
deb-1c26-4101-a6b0-c2fde959c66b
                                          vhd=VHD-e28412b8-af76-4a72-85af-7780a97c7ae4 capacity=21474836480 size=8388608 hidden=0 parent=VHD-5730
4603-4a25-4723-93cf-24564d3e96f3
                                          vhd=VHD-dedb2e9b-78e3-4067-92c9-42e3266c72cd capacity=21474836480 size=1946157056 hidden=1 parent=VHD-5
7304603-4a25-4723-93cf-24564d3e96f3
                                             vhd=VHD-4b697f4c-2169-4d7c-a41a-718fd427745b capacity=21474836480 size=8388608 hidden=0 parent=VHD-d
edb2e9b-78e3-4067-92c9-42e3266c72cd
                                             vhd=VHD-9ffc4f7c-c27a-4976-8c93-b1522822fab4 capacity=21474836480 size=21525168128 hidden=0 parent=V
HD-dedb2e9b-78e3-4067-92c9-42e3266c72cd

~Peg

olivierlambert

IIRC, we got a protection on the same job to re-run on itself. But not between jobs.

I'm a bit surprised that our VDI chain protection didn't prevent this scenario.

Anonabhar

<looks sheepishly as he admits> It probably would have.... The Cont replication jobs were running on a opensource version (different appliance) while the incremental and full (monthly) exports were being run on a Enterprise version.

Not sure if the VDI protection would pick up the other XOA jobs.

~Peg

olivierlambert

Mixing jobs in various XO isn't a good idea in any case.

mbt

For the records:

After the vm copy that somehow seems to have triggered the coalescing the delta backup job and another weekly full backup job ran without any issues.

Today is going to be interesting. The daily deltas only run weekdays (Mo-Fr), the full backup on Sundays. The deltas ran fine, the full backup afterwards, too. But now I again see stuck vdi coalesce jobs on the sr's advanced tab.
Looks the same as last Monday. If it doesn't clear over the day I'll try the same trick from Friday and see if that works again.

Interesting, that now twice in a row this happens after the full backup job.

mbt

Same as Friday. I started a copy of the vm and immediately the coalescing began and finished successfully shortly after.

Anonabhar

mbt,

How are you copying the VM? Are you using XC or XO and are you doing a fast clone or a full copy? If you have the same problem as I do at the moment, using fast clone will not help your situation as it will not create a new tree for the VHD.

It does not surprise me that you are seeing the SMGC kick in. Some operations will ask the SM to re-scan the SR and this will also cause the SMGC to start. If you previously had a stuck job that "Exceptioned" the GC will try again, and if the VM is shutdown, it will be able to do a offline coalesce.

See if you can get a tree view of your VDI's and look for something that has a lot of indents in it.. What type of SR do you use? iSCSI or NFS?

~Peg