How to remove broken/empty/unexistent ZFS pool?

[interduo]

I was changing one of empty zpool drives.

So I did:

• run command: zpool export zfs-empty-pool,
• remove drive 1 from server,
• remove drive 2 from server,

zpool status

	zfs2-vps1                         UNAVAIL      0     0     0  insufficient replicas
	  mirror-0                        UNAVAIL      0     0     0  insufficient replicas
	    ata-ADATA_SU900_2J0520038666  UNAVAIL      0     0     0
	    ata-ADATA_SU900_2J0220067777  UNAVAIL      0     0     0

ok - that's fine - but how to remove this? I would like to create new zpool with the same name (I don't want to reboot whole host).

This situation is reproducable - generates OOPS in kernel when I run zpool clear zfs2-vps1

[1287844.270966] INFO: task z_wr_iss:756814 blocked for more than 362 seconds.
[1287844.272353]       Tainted: P           O      5.15.35-3-pve #1
[1287844.273590] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[1287844.274900] task:z_wr_iss        state:D stack:    0 pid:756814 ppid:     2 flags:0x00004000
[1287844.276208] Call Trace:
[1287844.277496]  <TASK>
[1287844.278780]  __schedule+0x33d/0x1750
[1287844.280039]  ? spl_kmem_cache_alloc+0x79/0x790 [spl]
[1287844.281302]  ? zfs_btree_insert_into_leaf+0x24a/0x2c0 [zfs]
[1287844.282779]  schedule+0x4e/0xb0
[1287844.283976]  cv_wait_common+0xfd/0x130 [spl]
[1287844.285200]  ? wait_woken+0x70/0x70
[1287844.286448]  __cv_wait+0x15/0x20 [spl]
[1287844.287664]  metaslab_load+0x46/0x8a0 [zfs]
[1287844.289067]  ? __kmalloc_node+0x166/0x3a0
[1287844.290301]  ? __cond_resched+0x1a/0x50
[1287844.291503]  ? recalibrate_cpu_khz+0x10/0x10
[1287844.292709]  metaslab_activate+0x4f/0x2c0 [zfs]
[1287844.294136]  metaslab_alloc_dva+0x2cd/0x14b0 [zfs]
[1287844.295543]  metaslab_alloc+0xbe/0x270 [zfs]
[1287844.297222]  zio_dva_allocate+0xe6/0x9a0 [zfs]
[1287844.298971]  ? spl_kmem_alloc_impl+0xb5/0x100 [spl]
[1287844.300489]  ? zio_io_to_allocate+0x68/0x90 [zfs]
[1287844.302252]  ? tsd_hash_search+0x5c/0xc0 [spl]
[1287844.303774]  zio_execute+0x95/0x160 [zfs]
[1287844.305518]  taskq_thread+0x29b/0x4c0 [spl]
[1287844.307061]  ? wake_up_q+0x90/0x90
[1287844.308574]  ? zio_gang_tree_free+0x70/0x70 [zfs]
[1287844.310331]  ? taskq_thread_spawn+0x60/0x60 [spl]
[1287844.311849]  kthread+0x12a/0x150
[1287844.313355]  ? set_kthread_struct+0x50/0x50
[1287844.314890]  ret_from_fork+0x22/0x30
[1287844.316411]  </TASK>

[rincebrain]

If the pool is showing up in zpool status, it was not exported at the time you ran zpool status.

I seem to recall people reporting issues trying to export things themselves on Proxmox and being surprised that it imported them again behind their back, so that might be burning you. Docker in particular, if I recall, does such things if it thinks it's in control of a pool.

That said, once you have yanked the disks to a live pool, you can either put them back, hope zpool destroy removes it and doesn't hang until reboot, or reboot. #11082 tracks the work for trying to allow you to export a pool that's stuck in the middle of holding some locks, but it turns out that the dance to drop every lock you might conceivably be holding when a pool goes away is...very hard.

[interduo]

I seem to recall people reporting issues trying to export things themselves on Proxmox and being surprised that it imported them again behind their back, so that might be burning you. Docker in particular, if I recall, does such things if it thinks it's in control of a pool

That was that!

zpool destroy or zpool clear or other commands just make process stalled and make uptime bigger.

Ok - I am waiting for merging #11082.