lxc directory mountpoint migration don't work

[grin]

mp0: /mnt/remote_nfs_share,mp=/mnt/extstorage,size=1T

Oct 06 15:32:07 ERROR: migration aborted (duration 00:00:00): can't determine assigned storage for mountpoint 'mp0'

it is caused by the storage subsystem since it wants to be sure that the storage is managed by proxmox. However there seem to be no way to do it: directory and nfs (since it's an nfs remote dir mounted in the host) want to create a .raw disk instead of using the directory.

Logically I would create a "shared directory", then mount it by mountpoint, but I cannot guess how.

(defining a directory on '.mnt/remote_nfs_share' and refer it as
mp0: nfsshare:/,mp=/mnt/extstorage,size=1T
won't work.)

 

[fabian]

you did not post whether you are running a current version of PVE, but if you are, there is a "-force" flag for "pct migrate" which allows migration with bind and/or device mount points. note that you are responsible yourself for making sure that those source paths/devices are available under the configured name on the target node.

pct migrate <vmid> <target> [OPTIONS]

Migrate the container to another node. Creates a new migration task.

<vmid> integer (1 - N)
The (unique) ID of the VM.

<target> string
Target node.

-force boolean
Force migration despite local bind / device mounts. WARNING: identical bind / device mounts need
to be available on the target node.

 

[grin]

you did not post whether you are running a current version of PVE, but if you are, there is a "-force" flag for "pct migrate" which allows migration with bind and/or device mount points.

We are running the latest and greatest (in fact the question have arisen while upgrading).
How do you want me to convey this information to ha-manager? Or does it automagically "force" now?

[UPDATE] I have checked, and 'pct migrate <ctid> <node> --force --online' (also with '-force' with one dash, there seem to be a bit discrepancy between the manpage and pct's error outputs) fails, since it starts a HA migrate which doesn't seem to use --force and fail with the usual error.

 

[grin]

You possibly meant:

--- PVECT.pm  2016-09-15 13:24:38.000000000 +0200
+++ /tmp/PVECT.pm  2016-10-07 11:25:25.156919665 +0200
@@ -92,6 +92,7 @@
  vmid => $id,
  target => $target,
  online => 0, # we cannot migrate CT (yet) online, only relocate
+  force => 1,  # we want to pull over local mount points if possible
  };
  # always relocate container for now

(doesn't work though, I'd need more time to reverse engineer it, but you see the point)

 

[fabian]

yes, this does not work (yet) for HA managed containers. could you file a tracking bug at https://bugzilla.proxmox.com ? I am afraid it will need a bit of discussion on how to implement it, because it's not that trivial for the HA case (we need to configure somewhere which bind mounts are available on which nodes, but they don't fit into the regular storage model, ...)

 

[grin]

https://bugzilla.proxmox.com/show_bug.cgi?id=1147