qmp command 'guest-fstrim' failed - got timeout

[janos]

Hi,

On a CentOS 7 VM with CEPH backend, i can run fstrim inside the VPS without issue:

~ # time fstrim -a -v
/data: 0 B (0 bytes) trimmed
/boot: 817.9 MiB (857587712 bytes) trimmed
/: 13 GiB (13946646528 bytes) trimmed

real    0m15.663s
user    0m0.013s
sys     0m0.026s

I try to do this via the agent:

qm guest cmd 144 fstrim

But this is the result:

VM 144 qmp command 'guest-fstrim' failed - got timeout

I see this in the VM syslog:

Dec  7 20:54:58 proxy qemu-ga: info: guest-ping called
Dec  7 20:54:58 proxy qemu-ga: info: guest-fstrim called

I think, qm try to wait for the respond, and after few seconds it run to timeout, what is normal because fstrim can run several sec or min. However i do not see timeout settings for "guest cmd" command.

Any idea?

 

[karnz]

Also found this issue on the latest PVE 5.3-5.

 

[janos]

Still exist ing 6.1. Not only with CEPH. I think fstrim have to an async commands.

 

[thiagotgc]

Still exist ing 7.2.

 

[Neobin]

Couldn't find any related ticket on Bugzilla [1].
So maybe someone who is affected should open one there.

[1] https://bugzilla.proxmox.com

 

[fiona]

Hi,
FYI, I sent a patch now to increase the timeout.

 

[rcd]

Now I get this in 7.4-3

# qm guest exec 106 fstrim -- -a
timeout reached, returning pid
{
   "pid" : 30791
}

 

[fiona]

Hi,

Now I get this in 7.4-3

# qm guest exec 106 fstrim -- -a
timeout reached, returning pid
{
   "pid" : 30791
}

how large is your disk? What kind of underlying storage do you have (network storage)? The timeout is 10 minutes now, but maybe that still is not enough for you.

 

[rcd]

Storage is local zpool. 350G with 200G used.

I didn't time it, but I think it was under a minute, certainly not ten.

Just tried again, this time it succeeded after 26 seconds. Seems it's a bit hit and miss.

 

[fiona]

Storage is local zpool. 350G with 200G used.

I didn't time it, but I think it was under a minute, certainly not ten.

Just tried again, this time it succeeded after 26 seconds. Seems it's a bit hit and miss.

Ah, you're right. There is a different timeout depending on whether the command is issued via API or via CLI. And the default in the CLI is 30 seconds. But it's not a big deal, you get the PID and the command will just continue inside the VM. If you do want to wait for it, just use qm guest exec 106 fstrim --timeout <seconds> -- -a with a large timeout.

 

[rcd]

Ok I'll try that thanks.