[Gluster-users] VMs blocked for more than 120 seconds

Tue May 21 10:05:07 UTC 2019

Hi Martin,

Glad it worked! And yes, 3.7.6 is really old! :)

So the issue is occurring when the vm flushes outstanding data to disk. And
this
is taking > 120s because there's lot of buffered writes to flush, possibly
followed
by an fsync too which needs to sync them to disk (volume profile would have
been helpful in confirming this). All these two options do is to truly
honor O_DIRECT flag
(which is what we want anyway given the vms are opened with 'cache=none'
qemu option).
This will skip write-caching on gluster client side and also bypass the
page-cache on the
gluster-bricks, and so data gets flushed faster, thereby eliminating these
timeouts.

-Krutika

On Mon, May 20, 2019 at 3:38 PM Martin <snowmailer at gmail.com> wrote:

> Hi Krutika,
>
> Also, gluster version please?
>
> I am running old 3.7.6. (Yes I know I should upgrade asap)
>
> I’ve applied firstly "network.remote-dio off", behaviour did not changed,
> VMs got stuck after some time again.
> Then I’ve set "performance.strict-o-direct on" and problem completly
> disappeared. No more stucks at all (7 days without any problems at all).
> This SOLVED the issue.
>
> Can you explain what remote-dio and strict-o-direct variables changed in
> behaviour of my Gluster? It would be great for later archive/users to
> understand what and why this solved my issue.
>
> Anyway, Thanks a LOT!!!
>
> BR,
> Martin
>
> On 13 May 2019, at 10:20, Krutika Dhananjay <kdhananj at redhat.com> wrote:
>
> OK. In that case, can you check if the following two changes help:
>
> # gluster volume set $VOL network.remote-dio off
> # gluster volume set $VOL performance.strict-o-direct on
>
> preferably one option changed at a time, its impact tested and then the
> next change applied and tested.
>
> Also, gluster version please?
>
> -Krutika
>
> On Mon, May 13, 2019 at 1:02 PM Martin Toth <snowmailer at gmail.com> wrote:
>
>> Cache in qemu is none. That should be correct. This is full command :
>>
>> /usr/bin/qemu-system-x86_64 -name one-312 -S -machine
>> pc-i440fx-xenial,accel=kvm,usb=off -m 4096 -realtime mlock=off -smp
>> 4,sockets=4,cores=1,threads=1 -uuid e95a774e-a594-4e98-b141-9f30a3f848c1
>> -no-user-config -nodefaults -chardev
>> socket,id=charmonitor,path=/var/lib/libvirt/qemu/domain-one-312/monitor.sock,server,nowait
>> -mon chardev=charmonitor,id=monitor,mode=control -rtc base=localtime
>> -no-shutdown -boot order=c,menu=on,splash-time=3000,strict=on -device
>> piix3-usb-uhci,id=usb,bus=pci.0,addr=0x1.0x2
>>
>> -device virtio-scsi-pci,id=scsi0,bus=pci.0,addr=0x4
>> -device virtio-serial-pci,id=virtio-serial0,bus=pci.0,addr=0x5
>> -drive file=/var/lib/one//datastores/116/312/*disk.0*
>> ,format=raw,if=none,id=drive-virtio-disk1,cache=none
>> -device
>> virtio-blk-pci,scsi=off,bus=pci.0,addr=0x6,drive=drive-virtio-disk1,id=virtio-disk1
>> -drive file=gluster://localhost:24007/imagestore/
>> *7b64d6757acc47a39503f68731f89b8e*
>> ,format=qcow2,if=none,id=drive-scsi0-0-0-0,cache=none
>> -device
>> scsi-hd,bus=scsi0.0,channel=0,scsi-id=0,lun=0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0
>> -drive file=/var/lib/one//datastores/116/312/*disk.1*
>> ,format=raw,if=none,id=drive-ide0-0-0,readonly=on
>> -device ide-cd,bus=ide.0,unit=0,drive=drive-ide0-0-0,id=ide0-0-0
>>
>> -netdev tap,fd=26,id=hostnet0
>> -device e1000,netdev=hostnet0,id=net0,mac=02:00:5c:f0:e4:39,bus=pci.0,addr=0x3
>> -chardev pty,id=charserial0 -device
>> isa-serial,chardev=charserial0,id=serial0
>> -chardev socket,id=charchannel0,path=/var/lib/libvirt/qemu/channel/target/domain-one-312/org.qemu.guest_agent.0,server,nowait
>> -device
>> virtserialport,bus=virtio-serial0.0,nr=1,chardev=charchannel0,id=channel0,name=org.qemu.guest_agent.0
>> -vnc 0.0.0.0:312,password -device
>> cirrus-vga,id=video0,bus=pci.0,addr=0x2 -device
>> virtio-balloon-pci,id=balloon0,bus=pci.0,addr=0x7 -msg timestamp=on
>>
>> I’ve highlighted disks. First is VM context disk - Fuse used, second is
>> SDA (OS is installed here) - libgfapi used, third is SWAP - Fuse used.
>>
>> Krutika,
>> I will start profiling on Gluster Volumes and wait for next VM to fail.
>> Than I will attach/send profiling info after some VM will be failed. I
>> suppose this is correct profiling strategy.
>>
>
> About this, how many vms do you need to recreate it? A single vm? Or
> multiple vms doing IO in parallel?
>
>
>> Thanks,
>> BR!
>> Martin
>>
>> On 13 May 2019, at 09:21, Krutika Dhananjay <kdhananj at redhat.com> wrote:
>>
>> Also, what's the caching policy that qemu is using on the affected vms?
>> Is it cache=none? Or something else? You can get this information in the
>> command line of qemu-kvm process corresponding to your vm in the ps output.
>>
>> -Krutika
>>
>> On Mon, May 13, 2019 at 12:49 PM Krutika Dhananjay <kdhananj at redhat.com>
>> wrote:
>>
>>> What version of gluster are you using?
>>> Also, can you capture and share volume-profile output for a run where
>>> you manage to recreate this issue?
>>>
>>> https://docs.gluster.org/en/v3/Administrator%20Guide/Monitoring%20Workload/#running-glusterfs-volume-profile-command
>>> Let me know if you have any questions.
>>>
>>> -Krutika
>>>
>>> On Mon, May 13, 2019 at 12:34 PM Martin Toth <snowmailer at gmail.com>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> there is no healing operation, not peer disconnects, no readonly
>>>> filesystem. Yes, storage is slow and unavailable for 120 seconds, but why,
>>>> its SSD with 10G, performance is good.
>>>>
>>>> > you'd have it's log on qemu's standard output,
>>>>
>>>> If you mean /var/log/libvirt/qemu/vm.log there is nothing. I am looking
>>>> for problem for more than month, tried everything. Can’t find anything. Any
>>>> more clues or leads?
>>>>
>>>> BR,
>>>> Martin
>>>>
>>>> > On 13 May 2019, at 08:55, lemonnierk at ulrar.net wrote:
>>>> >
>>>> > On Mon, May 13, 2019 at 08:47:45AM +0200, Martin Toth wrote:
>>>> >> Hi all,
>>>> >
>>>> > Hi
>>>> >
>>>> >>
>>>> >> I am running replica 3 on SSDs with 10G networking, everything works
>>>> OK but VMs stored in Gluster volume occasionally freeze with “Task XY
>>>> blocked for more than 120 seconds”.
>>>> >> Only solution is to poweroff (hard) VM and than boot it up again. I
>>>> am unable to SSH and also login with console, its stuck probably on some
>>>> disk operation. No error/warning logs or messages are store in VMs logs.
>>>> >>
>>>> >
>>>> > As far as I know this should be unrelated, I get this during heals
>>>> > without any freezes, it just means the storage is slow I think.
>>>> >
>>>> >> KVM/Libvirt(qemu) using libgfapi and fuse mount to access VM disks
>>>> on replica volume. Can someone advice  how to debug this problem or what
>>>> can cause these issues?
>>>> >> It’s really annoying, I’ve tried to google everything but nothing
>>>> came up. I’ve tried changing virtio-scsi-pci to virtio-blk-pci disk
>>>> drivers, but its not related.
>>>> >>
>>>> >
>>>> > Any chance your gluster goes readonly ? Have you checked your gluster
>>>> > logs to see if maybe they lose each other some times ?
>>>> > /var/log/glusterfs
>>>> >
>>>> > For libgfapi accesses you'd have it's log on qemu's standard output,
>>>> > that might contain the actual error at the time of the freez.
>>>> > _______________________________________________
>>>> > Gluster-users mailing list
>>>> > Gluster-users at gluster.org
>>>> > https://lists.gluster.org/mailman/listinfo/gluster-users
>>>>
>>>> _______________________________________________
>>>> Gluster-users mailing list
>>>> Gluster-users at gluster.org
>>>> https://lists.gluster.org/mailman/listinfo/gluster-users
>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20190521/389f526e/attachment.html>