http://web.archive.org/web/20151107061126/https://openzfsonosx.org/forum/viewtopic.php?f=26&t=2373 for a snapshot of the topic that preceded this post. (Some of that content disappeared, so I continued here.)
2009 MacBookPro5,2 with 8 GB memory and Mavericks:
- first with 1.4.5 (2015-10-19) from https://openzfsonosx.org/wiki/Downloads#1.4.5
- then with 1.4.5 from the installer offered at https://github.com/openzfsonosx/zfs/issues/425#issuecomment-154332440
Maybe related: ARC Duelling with OS X File Cache · Issue #292 · openzfsonosx/zfs · GitHub
Also remarkable, although I never experimented with a ZVOL: After heavy reads on zvols and datasets, the system hangs with 1.4.5 · Issue #419 · openzfsonosx/zfs
Re https://openzfsonosx.org/wiki/Performance I experimented with a tune away from the defaults. If I recall correctly:
- 2 GB ARC
- 1 GB ARC meta.
Then, for a pool named zhandy on a Transcend StoreJet 25M (25 mobile) (probably TS640GSJ25M) with a USB 2.0 connection, I used PC-BSD 11.0-CURRENTNOV2015 on a different notebook to:
- begin a scrub
- unmount the one file system of the pool
- export the pool.
Saturday 2015-11-07 sometime around 06:30 I reconnected that mobile hard disk drive to the MacBookPro5,2 and commanded:
sudo zpool import zhandy
Soon after entering the password, I could no longer see an on-screen pointer so, suspecting a problem, I keyed Control-T (SIGINFO) then, at 06:39:11, the chord for sysdiagnose(1).
The on-screen clock stopped at 06:40:54 but activity (the scrub, presumably) remained visible at the mobile hard disk drive. So at the nearby notebook with PC-BSD, I began writing these notes and attempted a connection to the Mac:
ssh bbsadmin-l@192.168.1.3
The GUI of the Mac was almost completely at a standstill – very slow blinking of the rectangular cursor at the command line – and at 07:30 in real time, the clock in the menu bar of the Mac appeared stuck at 06:43:40.
No response to the ssh command so I could not tell, remotely, whether spindump had begun.
At the mobile HDD, activity remained visible.
At the command line, still, the result of SIGINFO:
load: 130.55 cmd: zpool 8006 running 0.00u 45.04s
I keyed Control-T once more, no response. I continued drafting this post, then with the real time around 07:32 I saw that the clock on the Mac had progressed to 07:31:45.
Around 08:00, with the clock still stuck at 07:31:45, I keyed Command-Control-Power to force a restart of the OS.
Soon afterwards, mobile HDD activity ceased to be visible but the forced restart was unsuccessful, so after 08:05 I forced off the Mac and then used single user mode, hoping to find a directory –
/private/var/tmp/sysdiagnose_…
– if found, I would have moved it to a non-volatile area before proceeding with multi-user mode.
Unfortunately, the unresponsive state of the OS had not allowed the stackshot(1) daemon to run the sysdiagnose tool – so there was no spindump, and so on. A single user mode view of /private/var/log showed that the most recent write was to system.log at 06:38 – before the key chord for sysdiagnose was used.
Alongside system.log one other file – org.openzfsonosx.zed.err – was also modified at 06:38.
Later, I made a copy of that .err file but there are no time stamps; I can’t guess which lines were the last before the OS stopped responding. I might paste the content of that file to a pastebin (and link from this post), but I doubt that it’ll help towards a bug report …