[LU-17385] sanity-lfsck test_26a: only 3 of 4 MDTs are in completed

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.16.0
• Affects Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Alexander Zarochentsev
• Reporter: Maloo

Description

This issue was created by maloo for Andreas Dilger <[e-mail]>

This issue relates to the following test suite run: https://testing.whamcloud.com/test_sets/e80cc085-ac08-4f47-b354-22551a7da132

test_26a failed with the following error:

(7) only 3 of 4 MDTs are in completed

Test session details:
clients: https://build.whamcloud.com/job/lustre-master-patchless/840 - 4.18.0-425.10.1.el8_7.x86_64
servers: https://build.whamcloud.com/job/lustre-master-patchless/840 - 4.18.0-425.10.1.el8_7.x86_64

<<Please provide additional information about the failure here>>

First started on 2023-12-20 for full runs, may be related to recent patch landing.

VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanity-lfsck test_26a - (7) only 3 of 4 MDTs are in completed

Issue Links

• is related to: LU-16826 MDS nodes panicked running lfsck repair create lost objects: (osd_handler.c:6260:osd_index_declare_ea_insert()) ASSERTION( fid != ((void *)0) ) failed

Comments

[Peter Jones]

Merged for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/53591/
Subject: LU-17385 tests: sanity-lfsck 23d fix and enable
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 028ed64d90cfdeb908fb5574aacf2f71c259e2c2

[Gerrit Updater]

"Alexander Zarochentsev <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53591
Subject: LU-17385 tests: sanity-lfsck 23d fix and enable
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 2f4f656947703d8b44a7ea49a8b2c84020591307

[Alexander Zarochentsev]

hongchao.zhang, thanks for the analysis, I am submitting a patch to test the idea.

[Andreas Dilger]

The sanity-lfsck test_23d is now skipped, so any patch that is fixing it needs to remove the always_except line.

[Hongchao Zhang]

The LFSCK command used in test_23d to start layout LFSCK on MDT0000 uses option "-o", which will broadcast LFSCK to all MDTs

int jt_lfsck_start(int argc, char **argv)
{
                ...
                case 'o':
                        start.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST |         
                                          LPF_OST_ORPHAN;
                        break;
                ...
}

the LFSCK command used in test_24 to start namespace LFSCK is also sent to all MDTs, but some MDT could not complete
the previous layout LFSCK yet, then it will find the two LFSCKs are different and return -EOPNOTSUPP

int lfsck_start(const struct lu_env *env, struct dt_device *key,
                struct lfsck_start_param *lsp)
{
        ...
        if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
                rc = -EALREADY;
                if (unlikely(start == NULL)) {
                        spin_unlock(&lfsck->li_lock);
                        GOTO(out, rc);
                }

                while (start->ls_active != 0) {
                        if (!(type & start->ls_active)) {
                                type <<= 1;
                                continue;
                        }

                        com = __lfsck_component_find(lfsck, type,
                                                     &lfsck->li_list_scan);
                        if (com == NULL)
                                com = __lfsck_component_find(lfsck, type,
                                                &lfsck->li_list_double_scan);
                        if (com == NULL) {
                                rc = -EOPNOTSUPP;      <--------- return with error -EOPNOTSUPP
                                break;
                        }
                        ...
                }
                ...
        }
        ...
}

the corresponding logs

00000020:00000001:1.0:1703243105.771867:0:217993:0:(tgt_handler.c:1621:tgt_handle_lfsck_notify()) Process entered
00100000:00000001:0.0:1703243105.771867:0:228721:0:(lfsck_lib.c:3489:lfsck_in_notify()) Process entered
00100000:00000001:1.0:1703243105.771868:0:217993:0:(lfsck_lib.c:3489:lfsck_in_notify()) Process entered
00100000:00000001:0.0:1703243105.771868:0:228721:0:(lfsck_lib.c:3104:lfsck_start()) Process entered
00100000:00000001:1.0:1703243105.771869:0:217993:0:(lfsck_lib.c:3104:lfsck_start()) Process entered
00100000:00000001:1.0:1703243105.771873:0:217993:0:(lfsck_bookmark.c:107:lfsck_bookmark_store()) Process entered
00080000:00000001:1.0:1703243105.771874:0:217993:0:(osd_handler.c:1912:osd_trans_create()) Process entered
00080000:00000010:1.0:1703243105.771877:0:217993:0:(osd_handler.c:1927:osd_trans_create()) kmalloced '(oh)': 288 at 00000000a80bb82f.
00100000:00000001:0.0:1703243105.771880:0:228721:0:(lfsck_lib.c:3174:lfsck_start()) Process leaving via out (rc=18446744073709551521 : -95 : 0xffffffffffffffa1)
00100000:00000001:0.0:1703243105.771886:0:228721:0:(lfsck_lib.c:3546:lfsck_in_notify()) Process leaving (rc=18446744073709551521 : -95 : ffffffffffffffa1)
00000020:00000001:0.0:1703243105.771887:0:228721:0:(tgt_handler.c:1629:tgt_handle_lfsck_notify()) Process leaving (rc=18446744073709551521 : -95 : ffffffffffffffa1)
00080000:00000001:1.0:1703243105.771888:0:217993:0:(osd_handler.c:1955:osd_trans_create()) Process leaving (rc=18446619811807463424 : -124261902088192 : ffff8efc05778800)
00010000:00000040:0.0:1703243105.771889:0:228721:0:(ldlm_lib.c:3238:target_committed_to_req()) last_committed 0, transno 0, xid 1785973155960256
00010000:00000001:0.0:1703243105.771890:0:228721:0:(ldlm_lib.c:3307:target_send_reply()) Process entered
00010000:00000200:0.0:1703243105.771892:0:228721:0:(ldlm_lib.c:3295:target_send_reply_msg()) @@@ sending reply  req@00000000c24bd5b0 x1785973155960256/t0(0) o1101->lustre-MDT0000-mdtlov_UUID@10.240.38.25@tcp:111/0 lens 320/224 e 0 to 0 dl 1703243116 ref 1 fl Interpret:/200/0 rc -95/0 job:'lctl.0' uid:0 gid:0

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/53544/
Subject: LU-17385 tests: always_except sanity-lfsck/24
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 955e38051765609fe3a661035c0fab2cfca733ce

[Andreas Dilger]

It looks like MDT0000 has "finished" the LFSCK run, but with an error:

 

status: partial
flags: incomplete

It isn't clear from the test output why it is "partial". It doesn't look like waiting longer (600s) in wait_all_targets_blocked() is better than using "-w" in this case, because the LFSCK threads are all finished, but with an error.

[Andreas Dilger]

Sorry, a typo on my part, and fixed in my comment. It is the new test that landed which caused the problem. 

[Alexander Zarochentsev]

> It looks like this test failure was introduced by patch https://review.whamcloud.com/50998 "LU-16826 tests: lfsck to repair a dangling remote entry" landing on 2023-12-20 which added sanity-lfsck.sh test_23c, but used:

no, the patch adds test_23d, test_23c is an old one with similar name:

run_test 23c "LFSCK can repair dangling name entry (3)
run_test 23d "LFSCK can repair a dangling name entry to a remote object

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53545
Subject: LU-17385 revert: LU-16826 tests: lfsck to repair a dangling remote entry
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: fb6c848ef816ecb17f02ac461c2352ced320c593