[LU-7799] mdt.*.hsm.actions skips some records

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.9.0
• Affects Version/s: Lustre 2.5.0, Lustre 2.7.0, Lustre 2.8.0
• Labels: hsm llog
• Severity: 3
• Assignee: John Hammond
• Reporter: John Hammond

Description

Note that this is surely responsible for a number of mysterious sanity-hsm test failures.

Due to a bug in (or misuse of) llog_cat_process() the HSM actions proc file will skip some records when read.

~# # mount and setup HSM
~# killall lhsmtool_posix
~# wc -l /proc/fs/lustre/mdt/lustre-MDT0000/hsm/actions
0 /proc/fs/lustre/mdt/lustre-MDT0000/hsm/actions
~# cd /mnt/lustre
lustre# for ((i = 0; i < 20; i++)); do
>   touch f$i
>   lfs hsm_archive f$i
> done

Now there should be 20 records in the actions file but there are only 19:

lustre# wc -l /proc/fs/lustre/mdt/lustre-MDT0000/hsm/actions
19 /proc/fs/lustre/mdt/lustre-MDT0000/hsm/actions

The missing record corresponds to f17:

lustre# lfs path2fid f17
[0x200000401:0x12:0x0]
lustre# grep '0x200000401:0x12:0x0' /proc/fs/lustre/mdt/lustre-MDT0000/hsm/actions
lustre# lfs hsm_action f17
f17: ARCHIVE waiting (from 0 to EOF)

The issue is with how the startidx parameter to llog_cat_process() is handled (see mdt_hsm_actions_proc_show() and hsm_actions_show_cb()). startidx becomes lpd_startidx then lpcd_first_idx which is actually skipped in llog_process_thread().