[Bug 277992] mpr and possible trim issues

From: <bugzilla-noreply_at_freebsd.org>
Date: Tue, 09 Apr 2024 12:15:08 UTC
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277992

--- Comment #7 from mike@sentex.net ---
I let the job continue over night and it continues to get slower on writes. i.e
after doing a trim -f /dev/da[disk in pool]/p1 one by one, the tests start out
fast (~ 300MB of files in the zfs dataset) for 4 iterations. 
# zfs get used,referenced,compressratio quirk-test/bull1
NAME              PROPERTY       VALUE  SOURCE
quirk-test/bull1  used           371G   -
quirk-test/bull1  referenced     369G   -
quirk-test/bull1  compressratio  3.86x  -
disks are 1TB each

Apr  8 15:00:59 r-14mfitest LOOP[30713]: ending job ran 1272 seconds (0:21:12)
Apr  8 15:28:54 r-14mfitest LOOP[42451]: ending job ran 1285 seconds (0:21:25)
Apr  8 15:53:22 r-14mfitest LOOP[52716]: ending job ran 1287 seconds (0:21:27)
Apr  8 16:18:03 r-14mfitest LOOP[63083]: ending job ran 1300 seconds (0:21:40)
Apr  8 16:55:39 r-14mfitest LOOP[78807]: ending job ran 2075 seconds (0:34:35)
Apr  8 17:56:56 r-14mfitest LOOP[4539]: ending job ran 3496 seconds (0:58:16)
Apr  8 19:09:44 r-14mfitest LOOP[35043]: ending job ran 4187 seconds (1:9:47)
Apr  8 20:27:37 r-14mfitest LOOP[67633]: ending job ran 4493 seconds (1:14:53)
Apr  8 21:47:30 r-14mfitest LOOP[1103]: ending job ran 4613 seconds (1:16:53)
Apr  8 23:11:06 r-14mfitest LOOP[36135]: ending job ran 4836 seconds (1:20:36)
Apr  9 00:51:15 r-14mfitest LOOP[78070]: ending job ran 5829 seconds (1:37:9)
Apr  9 02:46:06 r-14mfitest LOOP[26187]: ending job ran 6710 seconds (1:51:50)
Apr  9 04:57:16 r-14mfitest LOOP[81635]: ending job ran 7690 seconds (2:8:10)**


*** daily runs at 3am so will slow that one down

This morning, I rebooted the box so trim values are the same. I then forced 6
TRIMs on the pool (zpool trim -w quirk-test) and it sort of seems to make a
difference ? But not the same as doing a full trim -f

Apr  9 08:02:32 r-14mfitest LOOP[53380]: ending job ran 4802 seconds (1:20:2)

so instead of a 2hr run it took 1:20m

--- Comment #8 from mike@sentex.net ---
I let the job continue over night and it continues to get slower on writes. i.e
after doing a trim -f /dev/da[disk in pool]/p1 one by one, the tests start out
fast (~ 300MB of files in the zfs dataset) for 4 iterations. 
# zfs get used,referenced,compressratio quirk-test/bull1
NAME              PROPERTY       VALUE  SOURCE
quirk-test/bull1  used           371G   -
quirk-test/bull1  referenced     369G   -
quirk-test/bull1  compressratio  3.86x  -
disks are 1TB each

Apr  8 15:00:59 r-14mfitest LOOP[30713]: ending job ran 1272 seconds (0:21:12)
Apr  8 15:28:54 r-14mfitest LOOP[42451]: ending job ran 1285 seconds (0:21:25)
Apr  8 15:53:22 r-14mfitest LOOP[52716]: ending job ran 1287 seconds (0:21:27)
Apr  8 16:18:03 r-14mfitest LOOP[63083]: ending job ran 1300 seconds (0:21:40)
Apr  8 16:55:39 r-14mfitest LOOP[78807]: ending job ran 2075 seconds (0:34:35)
Apr  8 17:56:56 r-14mfitest LOOP[4539]: ending job ran 3496 seconds (0:58:16)
Apr  8 19:09:44 r-14mfitest LOOP[35043]: ending job ran 4187 seconds (1:9:47)
Apr  8 20:27:37 r-14mfitest LOOP[67633]: ending job ran 4493 seconds (1:14:53)
Apr  8 21:47:30 r-14mfitest LOOP[1103]: ending job ran 4613 seconds (1:16:53)
Apr  8 23:11:06 r-14mfitest LOOP[36135]: ending job ran 4836 seconds (1:20:36)
Apr  9 00:51:15 r-14mfitest LOOP[78070]: ending job ran 5829 seconds (1:37:9)
Apr  9 02:46:06 r-14mfitest LOOP[26187]: ending job ran 6710 seconds (1:51:50)
Apr  9 04:57:16 r-14mfitest LOOP[81635]: ending job ran 7690 seconds (2:8:10)**


*** daily runs at 3am so will slow that one down

This morning, I rebooted the box so trim values are the same. I then forced 6
TRIMs on the pool (zpool trim -w quirk-test) and it sort of seems to make a
difference ? But not the same as doing a full trim -f

Apr  9 08:02:32 r-14mfitest LOOP[53380]: ending job ran 4802 seconds (1:20:2)

so instead of a 2hr run it took 1:20m

-- 
You are receiving this mail because:
You are the assignee for the bug.