[LU-7004] fix "lctl set_param -P" to allow deprecation of "lctl conf_param"

• Type: Improvement
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.13.0
• Affects Version/s: Lustre 2.9.0
• Labels: llnl patch
• Assignee: James A Simmons
• Reporter: Andreas Dilger

Description

The lctl set_param -P functionality was landed via LU-3155 for 2.5.0 and deprecates lctl conf_param usage. There is a deprecation message printed in jt_lcfg_mgsparam() since 2.7.53 that conf_param is obsolete and set_param should be used instead.

The test scripts need to be changed over to use "lctl set_param -P" exclusively, both to quiet the deprecation warning, as well as verify that the "lctl set_param -P" functionality is working correctly.

For now, I'm going to bump the deprecation warning to 2.8.53 to give us time to make this change in the 2.9 release cycle since there isn't enough time to do it for 2.8.0 anymore.

Issue Links

• is related to: LU-10937 Use sysfs to fix up sptlrpc handling.
• is related to: LU-8482 auster run of sanity trying to use deprecated lctl conf_param in 2.8
• is related to: LU-9431 class_process_proc_param can't handle sysfs
• is related to: LU-7031 lctl set_param -P behaves differently than lctl conf_param
• is related to: LU-9399 lctl conf_param and lctl set_param -P are broken for llite parameters
• is related to: LU-10626 utils/tests: lctl set_param -P does not appear to do anything
• is related to: LU-10756 Send Uevents for interesting Lustre changes
• is related to: LU-8414 DNE: Setting of remote_dir_gid parameter not persistent
• is related to: LU-3155 Permanent parameters with lctl set_param -P

Comments

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/32784
Subject: LU-7004 obd: handle fsname-MDT0000-**MDT0000 for obdname2fsname
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 29b45b5dd7c87dbf4b6168ca07d8bbd0acbafe84

[James A Simmons]

looks like this work is coming to completion

[James A Simmons]

So the good news is that I found lctl conf_param lustre-OST*.failover.node=10.0.0.1@tcp can be done with set_param P with already existing code. The command to do it is lctl set_param -P osc.lustre-OST*.import=connection=10.0.0.1@tcp. So its just a matter of updating the test in this case. Only sptlrpc is not functional.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/31081/
Subject: LU-7004 quota: make lctl set_param -P functional for quota
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 6d9df3b2c0d54b16d00b6e419c7bb958e6b54844

[Gerrit Updater]

John L. Hammond ([e-mail]) merged in patch https://review.whamcloud.com/31207/
Subject: LU-7004 obd: make LCFG_SET_PARAM functional
Project: fs/lustre-release
Branch: b2_10
Current Patch Set:
Commit: d7f9f2130c5d8c74004f36aa68938df57778a6b1

[Gerrit Updater]

Minh Diep ([e-mail]) uploaded a new patch: https://review.whamcloud.com/31207
Subject: LU-7004 obd: make LCFG_SET_PARAM functional
Project: fs/lustre-release
Branch: b2_10
Current Patch Set: 1
Commit: 14d209c516d710585d4c492bfdfa6d2eb1add25d

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/31081
Subject: LU-7004 quota: make lctl set_param -P functional for quota
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: e68d0a91625c64dd66dc2f7395cf7dd102dba57b

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/28590/
Subject: LU-7004 obd: make LCFG_SET_PARAM functional
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: dfe60d0b98a1a888ca4ffce14788938c192b1520

[James A Simmons]

I also got this working. Only one test fails with lctl set_param -P which I tracked down to the problem that the lcfg changes are only being transmitted to the clients. So trying to deactivate an OST/MDT only appears to happen on the client and the servers don't see the same change. It appears the choice of where to send the lcfg info is done in the mgc layer. Does anyone know the exact method of how this done?

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/30087
Subject: LU-7004 tests: move from lctl conf_param to lctl set_param -P
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 57e99d4dd495dd0e69daee9dc52931c33fa45395

[James A Simmons]

We have 4 classes of parameters.

Class one are the global values i.e PARAM_TIMEOUT, PARAM_JOBID_VAR. The patch for this ticket adds support for this class for set_param -P.

The second class are the ones settable at mount or initial format i.e PARAM_MGSNODE, PARAM_NETWORK. Those should never be changed with set_param -P so they are filtered out. The patch for this ticket adds that filtering. Currently you can stomp on those values with set_param -P

The 3rd class is your typical obd proc mapping. This somewhat worked for set_param -P.

Lastly we have your "virtual" devices. Some of these that were virtual can now be mapped to real proc/sysfs entries. The two for this case are PARAM_ACTIVATE and PARAM_ID_UPCALL. Its PARAM_FAILNODE and PARAM_SRPC that are currently virtual. I have managed to allow setting the failnode node using set_param -P but it doesn't transmit the change to the client for some reason. As for PARAM_SRPC I have no idea what the goal was there. In lustre_param.h it list srpc.flavor as a proc entry but it doesn't exist. Anyways I will have to cover that in LU-7183.