L2ARC vdev sometimes perfectly hot before sleep (suspend) of FreeBSD, suddenly cold (after wake) #13478
Replies: 2 comments 4 replies
-
|
It would be nice to specify kernel and zfs versions in the bug report. I believe there were some fixes ab8a8f0 in persistent L2ARC, which fixed an issue where the l2arc can loose all the objects due to a race on write and restore during the import. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. This is present in https://cgit.freebsd.org/src/commit/?id=ab8a8f07457831a21db888afcf48910ce3e9214a |
Beta Was this translation helpful? Give feedback.
-
|
Defocusing from sleep/wake, note to self: lines 8337–8343 of zfs/arc.c at master · openzfs/zfs · GitHub |
Beta Was this translation helpful? Give feedback.
-
Sometimes, I'd encounter the plummet affecting all L2ARC devices at the same time. In recent months, I use three devices (14 + 14 + 7.2 G). I might broaden the subject of this discussion, because sometimes (often?) the plummets were also found after starting or restarting the OS. I typically restart only if I upgrade packages in a new ZFS boot environment, then activate the environment before restarting the OS. Until recently, my habit was physical reconnection of devices before waking the computer. Assuming that presence would reduce the likelihood of things plummeting. Maybe a false assumption; recently, I choose to wait until a few seconds after wake, then reconnect the devices, one by one. If I'm not mistaken: fewer plummets 👍 Whilst the sense of things improving is not yet shown in e.g. Netdata:
– I hope for things to be more obviously improved by me also breaking the habit of having the devices connected at computer start (or restart) time. I'll now experiment, for a few days, with not connecting the devices until shortly after the deamon starts … |
Beta Was this translation helpful? Give feedback.
-
First test result: failure. All cold. I'll hide this comment as off-topic, might add further test results without un-hiding the comment. In this case, I connected the three devices shortly after ttyv0 showed the start of SDDM. From
From PostscriptThis is odd:
Compare that with the screenshot in the previous comment, taken around ninety minutes earlier, when the Netdata view of things had no gap at the far right of the timeline. Shutdown was around 11:58 (BOOT was at 12:00:28), now there appears to be a gap of around three hours. Tentative cross-reference: FreeBSD bug 270997 – net-mgmt/netdata: sometimes not running when expected, sometimes running but http://localhost:19999 not responding |
Beta Was this translation helpful? Give feedback.
-
Sorry, I discovered that hiding one's own commentary is a feature of issues but not discussions. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
I have two L2ARC devices on USB, with a script to gracefully offline the two before sleep.
Sometimes, after wake, with onlining automated by zfsd(8), I find one of the two cold.
Yesterday, for example,
/dev/gpt/cache-august(GUID0x6285db6c05d8b7f1):1456blocks restored at 10:14 https://pastebin.com/raw/cqxqs2tD0012blocks restored at 18:37 https://pastebin.com/raw/3i061EAhPlease: is it normal for each L2ARC vdev to occasionally go cold (when taken online)? Or should I investigate further, as a possible bug?
I see a 2020 comment from @behlendorf:
– but (sorry) I'm not sure that I understand. For ring buffer, Wikipedia redirects to circular buffer, with this sentence:
So, I expect and understand overwrite (loss) of the oldest, but I'm surprised by the suddenness of hot to cold (like, losing everything at once).
Side note: in recent months I have
vfs.zfs.l2arc.noprefetch=0in sysctl.conf(5).Beta Was this translation helpful? Give feedback.
All reactions