Summary: | zdb(8): zdb -S poolname no longer works | ||
---|---|---|---|
Product: | Base System | Reporter: | Adam Stylinski <kungfujesus06> |
Component: | bin | Assignee: | freebsd-bugs (Nobody) <bugs> |
Status: | Closed Unable to Reproduce | ||
Severity: | Affects Only Me | CC: | gert |
Priority: | Normal | ||
Version: | Unspecified | ||
Hardware: | Any | ||
OS: | Any |
Description
Adam Stylinski
2014-01-22 15:40:00 UTC
I see this as well, and find it highly confusing. This is on FreeBSD 10.1-RELEASE-p24 / amd64. "zpool status" shows two pools. "zdb -s" or "zdb -S" perfectly well work for one of them, but the other one (copy-paste'ed names) is reported as "No such file or directory": $ zpool list NAME SIZE ALLOC FREE FRAG EXPANDSZ CAP DEDUP HEALTH ALTROOT MOEBIUS4 60G 2.26G 57.7G 1% - 3% 1.00x ONLINE - MOEBIUS4_FC 1016G 217G 799G 6% 256K 21% 1.00x ONLINE - $ zdb -s MOEBIUS4 capacity operations bandwidth ---- errors ---- description used avail read write read write read write cksum ... $ zdb -s MOEBIUS4_FC zdb: can't open 'MOEBIUS4_FC': No such file or directory (same thing for "-S") MOEBIUS4 is the root pool, in case it makes a difference. ok, turns out I could solve this on my own, so maybe the result can help either make the tools more useful, or just fix the issue. Turns out that when this pool was created, it was put on /dev/multipath/fc_mpath (because there are two FC paths to the same LUN on a SAN). I made a mistake when creating the multipath label, that is, I only "gmultipath create"'d it, not "gmultipath label" - it was in RAM, everything worked, after reboot, the label was not there and the pool was imported from /dev/da2. And here comes's the catch: "zdb -s" still wanted to access /dev/multipath/fc_mpath, which, of course, did not exist. After fixing the labeling and re-importing the pool it works now. But it's still a bug - at least, it should point out *what* is "No such file or directory", or preferrably, just use the device that "zpool status" knows is used... For bugs matching the following conditions: - Status == In Progress - Assignee == "bugs@FreeBSD.org" - Last Modified Year <= 2017 Do - Set Status to "Open" What do you mean by "Unable to Reproduce"? This should be fairly easy to do - create a pool using a device label, export the pool, change the device label, import it again. See my description on how I ended up there. (In reply to Gert Doering from comment #4) I cannot reproduce the issue in comment #0 which makes no mention of multipath. The issue that you report is not specific to '-s' option, it would affect all zdb usages for that pool. If you are able to reproduce your issue on CURRENT or 11.2 BETA, please file a new bug. Multipath is just the way I ended up at a pool that was created with a different device name than what it found on "zfs import" - I added this is as an explanation how I got into the same situation as the reporter in #0, who did not provide background. If, as you say, all uses of zdb (not just -s/-S) get confused by a name change of the underlying device(s) - shouldn't that be *fixed*? Not running 11.2-BETA or CURRENT anywhere, but I'll see if I can set up a VM to reproduce. OK, tried with a 12.0-CURRENT snapshot as of last week, and could not(!) reproduce this anymore. What I did was: create a ZFS pool on a separate hard disk, and moved that around - by removing a different hard disk (so da2 became da1), by adding and removing GPT labels (so /dev/gpt/zfs2 became /dev/da1p1). In all cases, zdb -s and zdb -S worked perfectly well - so it seems I either hit some really strange corner case, or it already got fixed in the time since. thanks a lot. (In reply to Gert Doering from comment #7) ZFS has got some improvements at updating stored vdev paths over the years. So, it must be that one of them helped with this problem too. Thank you for testing! |