Re: [linux-lvm] [lvmlockd] "VG xxx lock faied: error -221" while lv operation sucess

[David Teigland <teigland@redhat.com>]

On Tue, Jul 10, 2018 at 03:52:26PM +0800, Damon Wang wrote:
> Hi dave,
> 
> Thank you for your reply, unfortunately I have rebooted that host last stat :(
> 
> But good news is I have saved some logs, but I don't know weather it
> helps since I only saved "lvmlockctl -i", "sanlock client status",
> sanlock.log and messages.
> 
> I pasted "sanlock client status" on https://pastebin.com/M9TRKeBk
> "lvmlockctl -i ": https://pastebin.com/v9E1Atig
> 
> Related messages and sanlock.log:
> 
> "messages": https://pastebin.com/1tqtQvDS
> "sanlock.log": https://pastebin.com/ESA4Wz3B

That's all that's needed.  sanlock was getting io timeouts (-202) from the
storage (probably related to the qla errors you also see.)  sanlock uses a
10 second io timeout by default (although using a non default value has
not been as thoroughly tested.)

Here's the first timeout, while trying to acquire a lock:

Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 ballot 2 dblock read2 error -202
Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 ballot 2 retract error -210
Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 paxos_acquire 2 ballot error -210
Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 acquire_token disk error -210 RETRACT_PAXOS
Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 paxos_release already free last lver 1 owner 31 1 0 writer 31 1 20359 disk lver 1 owner 31 1 0 writer 31 1 20359
Jul  5 15:27:49 iscsi sanlock[19403]: 2018-07-05 15:27:49 20369 [19408]: r2136 cmd_acquire 3,10,19384 acquire_token -210 lease io error
Jul  5 15:27:49 iscsi lvmlockd[19384]: 1530775669 S lvm_f68643c145454594871f6f51b69bf8ba R 4bPTER-Q21t-VedL-UM8w-dPgn-6S13-lPoaB4 lock_san acquire error -210

Here's the second, while trying to release a lock:

Jul  5 15:27:59 iscsi sanlock[19403]: 2018-07-05 15:27:59 20379 [19407]: r2135 write_host_block host_id 31 flags 0 gen 0 rv -202
Jul  5 15:27:59 iscsi sanlock[19403]: 2018-07-05 15:27:59 20379 [19407]: r2135 release_token shared write_host_block -202
Jul  5 15:27:59 iscsi sanlock[19403]: 2018-07-05 15:27:59 20379 [19407]: r2135 release_token timeout r_flags 1
Jul  5 15:27:59 iscsi lvmlockd[19384]: 1530775679 S lvm_f68643c145454594871f6f51b69bf8ba R VGLK unlock_san release error -202
Jul  5 15:27:59 iscsi lvmlockd[19384]: 1530775679 S lvm_f68643c145454594871f6f51b69bf8ba R VGLK res_unlock lm error -221
Jul  5 15:28:02 iscsi kernel: qla2xxx [0000:05:00.1]-801c:7: Abort command issued nexus=7:4:3 --  1 2002.
Jul  5 15:28:04 iscsi kernel: qla2xxx [0000:05:00.1]-801c:7: Abort command issued nexus=7:4:0 --  1 2002.

The io timeout error during acquire is usually not a problem, because the
command will retry and succeed, or fail cleanly.  The io timeout error
during release is a bigger problem, because it leaves a stray lock that
later interferes with trying to acquire the lock again (the -17 errors you
get after this.)  It looks like lvmlockd needs to better handle io timeout
errors during unlock.  Thanks for the info,
Dave