[LU-9104] Unknown config param in llog fails mounting target

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.10.1, Lustre 2.11.0
• Severity: 3
• Assignee: James A Simmons
• Reporter: Rahul Deshmukh (Inactive)

Description

If we have unknown config parameter in llog then it fails the target mounting.

Steps to re-create it are as :

$>sh llmount.sh
$>export NOFORMAT=yes
$>for OST in $(lctl get_param mgs.MGS.live.lustre | grep OST); do echo "max_rpcs_in_flight=50 on $OST" ; lctl conf_param $OST.ost.max_rpcs_in_flight=50 ; done <--- Note here ost.max.. is mention instead of osc.max..
$>sh llmountcleanup.sh
$>sh llmount.sh <------- this hangs, as ost is not avilable 

Issue Links

• is related to: LU-8066 Move lustre procfs handling to sysfs and debugfs.

Comments

[Peter Jones]

great - thanks James

[James A Simmons]

I can fix this under LU-8066 since we will be moving to sysfs which handles these error codes differently.

[Peter Jones]

IMHO, given that this bug has existed in multiple shipped releases, it would be better to track this issue in a new ticket so that we can more easily track getting it into future releases.

[Li Xi]

I think we should fix this problem by ignoring ENOSYS in osd_process_config() so reopening the ticket.

[Li Xi]

The patch of https://review.whamcloud.com/25368/ might still has a bug in osd_process_config()

When a llog config of type MDT can not be understood by class_process_proc_param(PARAM_MDT) or any other prefix (e.g. PARAM_HSM), it is passed to osd_process_config() and class_process_proc_param() is called for it for twice.

The problem is why the -ENOSYS failure of class_process_proc_param(PARAM_OST) is not ignored in osd_process_config()?

I am not sure about the llog processing codes, but I guess the process is: class_process_proc_param(PARAM_MDT) -> class_process_proc_param(PARAM_HSM) -> class_process_proc_param(PARAM_HSM) -> class_process_proc_param(PARAM_OSD) -> class_process_proc_param(PARAM_OST). So if osd_process_config() doesn't understand the llog record, it should ignore the -ENOSYS. Right?

[Taizeng Wu (Inactive)]

I am trying to use `mount -t lustre -o nosvc xxx` which only start mgs, then execute `

lctl conf_param -d public4-MDT0000.mdt.job_cleanaup_interval` to remove wrong config.

[Taizeng Wu (Inactive)]

I am using lustre-2.10.5, when i mount mgs&mdt, it report unkown config parameter, how i remove wrong parameter.

 

```

[Thu Nov 29 15:00:14 2018] LDISKFS-fs (sdb): mounted filesystem with ordered data mode. Opts: errors=remount-ro
[Thu Nov 29 15:00:17 2018] LDISKFS-fs (sdb): mounted filesystem with ordered data mode. Opts: errors=remount-ro
[Thu Nov 29 15:00:18 2018] LDISKFS-fs (sdb): mounted filesystem with ordered data mode. Opts: user_xattr,errors=remount-ro,no_mbcache,nodelalloc
[Thu Nov 29 15:00:19 2018] Lustre: MGS: Connection restored to 0b413a06-995f-b510-6aba-9810cdf71a17 (at 0@lo)
[Thu Nov 29 15:00:19 2018] Lustre: Found index 0 for public4-MDT0000, updating log
[Thu Nov 29 15:00:19 2018] LustreError: 11-0: public4-OST0000-osc-MDT0000: operation ost_connect to node 10.10.1.11@o2ib failed: rc = -114
[Thu Nov 29 15:00:19 2018] LustreError: Skipped 4 previous similar messages
[Thu Nov 29 15:00:19 2018] LustreError: 19825:0:(obd_config.c:1361:class_process_proc_param()) public4-MDT0000: unknown config parameter 'mdt.job_cleanaup_interval=60'
[Thu Nov 29 15:00:19 2018] LustreError: 19825:0:(obd_config.c:1682:class_config_llog_handler()) MGC10.10.1.14@o2ib: cfg command failed: rc = -38
[Thu Nov 29 15:00:19 2018] Lustre: cmd=cf00f 0:public4-MDT0000 1:mdt.job_cleanaup_interval=60

[Thu Nov 29 15:00:19 2018] LustreError: 15c-8: MGC10.10.1.14@o2ib: The configuration from log 'public4-MDT0000' failed (-38). This may be the result of communication errors between this node and the MGS, a bad configuration, or other errors. See the syslog for more information.
[Thu Nov 29 15:00:19 2018] LustreError: 19761:0:(obd_mount_server.c:1386:server_start_targets()) failed to start server public4-MDT0000: -38
[Thu Nov 29 15:00:19 2018] LustreError: 19761:0:(obd_mount_server.c:1879:server_fill_super()) Unable to start targets: -38
[Thu Nov 29 15:00:19 2018] Lustre: Failing over public4-MDT0000
[Thu Nov 29 15:00:25 2018] Lustre: 19761:0:(client.c:2114:ptlrpc_expire_one_request()) @@@ Request sent has timed out for slow reply: [sent 1543474819/real 1543474819] req@ffff99bab7d6a100 x1618446900006864/t0(0) o251->MGC10.10.1.14@o2ib@0@lo:26/25 lens 224/224 e 0 to 1 dl 1543474825 ref 2 fl Rpc:XN/0/ffffffff rc 0/-1
[Thu Nov 29 15:00:26 2018] Lustre: server umount public4-MDT0000 complete
[Thu Nov 29 15:00:26 2018] LustreError: 19761:0:(obd_mount.c:1582:lustre_fill_super()) Unable to mount (-38)

```

[Gerrit Updater]

John L. Hammond ([e-mail]) merged in patch https://review.whamcloud.com/28232/
Subject: LU-9104 obd: Ignore unknown config param while mounting
Project: fs/lustre-release
Branch: b2_10
Current Patch Set:
Commit: 542815f43c2fbbc29e528fc1de496e54851ca720

[Gerrit Updater]

Minh Diep ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28232
Subject: LU-9104 obd: Ignore unknown config param while mounting
Project: fs/lustre-release
Branch: b2_10
Current Patch Set: 1
Commit: 902a5fba10096e3196cfed25a2bfc7db0bfe692a

[Peter Jones]

Landed for 2.11