[LU-14590] add output aggregation for "lctl get_param"

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.17.0
• Assignee: Fred Dilger
• Reporter: Andreas Dilger

Description

When running "lctl get_param" on a system with hundreds of OSTs and MDTs, and thousands of clients, often there are are many lines of output that are identical for all devices. For example:

# lctl get_param ldlm.namespaces.*.*
ldlm.namespaces.testfs-OST0000-osc-MDT0000.contended_locks=32
ldlm.namespaces.testfs-OST0000-osc-MDT0000.contention_seconds=2
ldlm.namespaces.testfs-OST0000-osc-MDT0000.ctime_age_limit=10
ldlm.namespaces.testfs-OST0000-osc-MDT0000.dirty_age_limit=10
ldlm.namespaces.testfs-OST0000-osc-MDT0000.early_lock_cancel=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.lock_count=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.lock_timeouts=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.lock_unused_count=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.lru_max_age=3900000
ldlm.namespaces.testfs-OST0000-osc-MDT0000.lru_size=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.max_nolock_bytes=0
ldlm.namespaces.testfs-OST0000-osc-MDT0000.max_parallel_ast=1024
ldlm.namespaces.testfs-OST0000-osc-MDT0001.contended_locks=32
ldlm.namespaces.testfs-OST0000-osc-MDT0001.contention_seconds=2

:
:
ldlm.namespaces.testfs-OST0102-osc-MDT0013.contended_locks=32
ldlm.namespaces.testfs-OST0102-osc-MDT0013.contention_seconds=2
ldlm.namespaces.testfs-OST0102-osc-MDT0013.ctime_age_limit=10
ldlm.namespaces.testfs-OST0102-osc-MDT0013.dirty_age_limit=10
ldlm.namespaces.testfs-OST0102-osc-MDT0013.early_lock_cancel=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.lock_count=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.lock_timeouts=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.lock_unused_count=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.lru_max_age=3900000
ldlm.namespaces.testfs-OST0102-osc-MDT0013.lru_size=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.max_nolock_bytes=0
ldlm.namespaces.testfs-OST0102-osc-MDT0013.max_parallel_ast=1024

but it would be much more convenient if the output was aggregated with wildcards in a manner similar to how multiple parameters are specified with "lctl set_param":

ldlm.namespaces.*.contended_locks=32
ldlm.namespaces.*.contention_seconds=2
ldlm.namespaces.*.ctime_age_limit=10
ldlm.namespaces.*.dirty_age_limit=10
ldlm.namespaces.*.early_lock_cancel=0
ldlm.namespaces.*.lock_count=0
ldlm.namespaces.*.lock_timeouts=0
ldlm.namespaces.*.lock_unused_count=0
ldlm.namespaces.testfs-OST00[00-14]-osc-MDT00[00-13].lru_max_age=3900000
ldlm.namespaces.testfs-OST00[15-63]-osc-MDT00[00-11,13].lru_max_age=3900000
ldlm.namespaces.testfs-OST00[15-63]-osc-MDT0012.lru_max_age=100000
ldlm.namespaces.*.lru_size=0
ldlm.namespaces.*.max_nolock_bytes=0
ldlm.namespaces.*.max_parallel_ast=1024

or something similar (it would be implementation dependent how disjoint regions of identifiers would be shown). This would not only allow reducing the amount of output, but also make it much more obvious in cases where there are differences in the settings (e.g. lru_max_age above).

Something like "lctl merge_param" would take the output of "lctl get_param" as input and merge the lines, either by optionally (with new '-m' option) forking a process to pipe the output of "lctl get_param -m" into, or allowing the output of previously-captured "get_param" output to be aggregated (possibly from dsh or clush running on multiple nodes at once).

There would have to be some implementation-specific smarts in the aggregation, for example to understand that client instance identifiers can be aggregated.

Issue Links

• is related to: LU-18689 recovery-small test 57 crashes in osp and sptlrpc
• is related to: LU-14442 lctl list_param/get_param '-w' option to dump only tunable/writeable parameters

Comments

[Alex Zhuravlev]

Alex, it isn't clear how this patch is causing the kernel crashes? It is not touching the kernel or recovery-small.sh?

I checked the original patch right after bisection and noticed it didn't change code in the kernel, but it does change how lfs get_param forms requests (slightly), so I guess this change have increased likelyhood?
will test with the patch James mentioned above.

[Andreas Dilger]

Alex, it isn't clear how this patch is causing the kernel crashes? It is not touching the kernel or recovery-small.sh?

[James A Simmons]

Alex patch https://review.whamcloud.com/c/fs/lustre-release/+/57948 attempts to address your issue. Please try it out.

[Alex Zhuravlev]

# MDSCOUNT=2 ONLY=57 ONLY_REPEAT=20  bash recovery-small.sh
...
[  162.131670] Lustre: DEBUG MARKER: == recovery-small test 57: read procfs entries causes kernel crash (repeat 8/20 iter, 1/ min) ========================================================== 10:12:28 (1746526348)
[  163.541999] systemd[1]: mnt-lustre.mount: Succeeded.
[  163.565883] LustreError: 12317:0:(obd_class.h:479:obd_check_dev()) Device 34 not setup
[  163.574210] LustreError: 12317:0:(obd_class.h:479:obd_check_dev()) Skipped 31 previous similar messages
[  163.574380] LustreError: 12318:0:(obd_class.h:479:obd_check_dev()) Device 34 not setup
[  163.575759] BUG: unable to handle kernel NULL pointer dereference at 0000000000000118
[  163.575821] PGD 14585a067 P4D 14585a067 PUD 153768067 PMD 0 
[  163.575851] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[  163.575872] CPU: 0 PID: 12318 Comm: lctl Tainted: G        W  O     --------- -  - 4.18.0 #12
[  163.575907] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-3.fc41 04/01/2014
[  163.575947] RIP: 0010:do_raw_read_lock+0x1/0x50
[  163.576013] Code: 85 92 5b 00 85 c0 74 8f 48 c7 c6 ef 61 e2 a9 48 89 df e8 c2 fd ff ff e9 7b ff ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 53 <81> 7f 08 ed 1e af de 48 89 fb 75 12 b8 00 02 00 00 f0 0f c1 03 a9
[  163.576076] RSP: 0018:ffff9b5bc457fde0 EFLAGS: 00010296
[  163.576138] RAX: ffffffffc08b0754 RBX: 0000000000000000 RCX: 0000000000000000
[  163.576317] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000110
[  163.576343] RBP: 0000000000000110 R08: 0000000000000001 R09: 0000000000000000
[  163.576369] R10: ffff9b5bc457fe40 R11: 0000000000000000 R12: 00000000ffffffff
[  163.576395] R13: 0000000000000000 R14: ffff9b5b84991400 R15: ffff9b5b9778dc30
[  163.576422] FS:  00007f801f4f3dc0(0000) GS:ffff9b5c0b000000(0000) knlGS:0000000000000000
[  163.576452] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  163.576491] CR2: 0000000000000118 CR3: 0000000145de6000 CR4: 0000000000350eb0
[  163.576517] Call Trace:
[  163.576527]  sptlrpc_import_sec_ref+0x14/0x40 [ptlrpc]
[  163.576666]  srpc_sptlrpc_sepol_seq_show+0x31/0x230 [ptlrpc]
[  163.576787]  seq_read+0x14e/0x3e0
[  163.576813]  full_proxy_read+0x4b/0x70
[  163.576921]  vfs_read+0xa1/0x150
[  163.576941]  ksys_read+0x3d/0xa0
[  163.577013]  do_syscall_64+0x4b/0x1b0

bisection results:

COMMIT          TESTED  PASSED  FAILED          COMMIT DESCRIPTION
e6172ff4af      3       2       1       BAD     LU-18904 tests: skip sanity/230k for older servers
fd2f0d3960      2       2       0               LU-18876 osp: rename conflicting ping_show
69aa1bf646      3       3       0               LU-18687 doc: move man pages to Documentation [1]
8837fce6d3      3       2       1       BAD     LU-18863 nodemap: warning for inconsistencies with offset
6f03f976c1      3       3       0               LU-18814 tests: run extra prog, execute NLOOPS mpi workloads
f251864767      3       3       0               LU-18862 ldiskfs: update for RHEL 9.6
8dbf049479      3       3       0               LU-17427 mdt: reduce hold time for BFL rename lock
9fe3951c44      3       2       1       BAD     LU-16350 ldiskfs: update for kernel 6.12
df2b5d99ad      11      10      1       BAD     LU-17933 target: do not break grants on RPC failure
a2e3a2f5a3      12      10      2       BAD     LU-14590 utils: merge similar list_param params
33e6c43e4c      20      20      0       GOOD    LU-12885 llite: add enum ll_file_flags for clarity
d33e96cdd4      20      20      0       GOOD    LU-18515 build: fix configure checks for ZFS 2.2.3
2db2c9dceb      20      20      0       GOOD    LU-13814 clio: add coo_dio_pages_init
8c0d073c17      20      20      0       GOOD    LU-13814 clio: add cl_dio_pages_init
a9099de5d5      20      20      0       GOOD    LU-15935 tests: ignore replay-dual/33 cleanup error

[Peter Jones]

Merged for 2.17

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/55724/
Subject: LU-14590 utils: merge similar list_param params
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: a2e3a2f5a3a891fc3fed391023b3cdb65af2d427