[Linux-cluster] Parameter snapshot in vm.sh
Dirk H. Schulz
dirk.schulz at kinzesberg.de
Thu Jan 14 17:07:37 UTC 2010
Hi Thomas,
Thomas Sjolshagen schrieb:
>
> Quoting "Dirk H. Schulz" <dirk.schulz at kinzesberg.de>:
>
>> Hi folks,
>>
>> I do not understand this as the virtual machine image is stored
>> wherever it is configured in the virtual machine config file - and
>> what does it have to do with snapshots (i. e. what is a "snapshot
>> directory")?
>>
>
> The snapshot directory is the directory into which vm.sh will "virsh
> save" the memory contents of the guest when shutting it down (when
> use_virsh="1" at least). As I understand it, during rgmanager (or
> clusvcadm -d) activities, the vm is saved to the snapshot directory
> and restored from the file in that directory - if it exists.
Thanks for pointing to the right direction. I should have looked more
deeply into vm.sh - you are right, it is used only in conjunction with
use_virsh=1 so I can look up there what is done exactly (when I need it,
at the moment it is all xm).
>
> Somewhat unrelated; I'd love to have a parameter that defines the
> "stop" action as either "snapshot" or "shutdown". It takes a long time
> to snapshot 10 guests to a gfs2 file system and the restore
> functionality does a valiant effort, but the Linux kernel does not
> seem to like being restored as of yet (the guest usually hangs after a
> seemingly arbitrary amount of time). As a result, my innodb based
> tables get a little cranky from the interruption/crash.
Is far as I can see in vm.sh the standard action is shutdown:
> case $1 in
> start)
> validate_all || exit $OCF_ERR_ARGS
> do_start
> exit $?
> ;;
> stop)
> validate_all || exit $OCF_ERR_ARGS
> do_stop shutdown destroy
> exit $?
> ;;
In the "do_stop" line the method is set as a parameter to the do_stop
function (which hands it to the xm/virsh stop functions). If you need
both you could seperate a vm-stop.sh and vm-snap.sh script and use
relevant <vm-stop name=...> containers in cluster.conf (I did not test
that but it should work like this, should it not?).
Dirk
More information about the Linux-cluster
mailing list