Re: [libvirt] [RFC]: Volume allocation progress reporting

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Mar 24, 2009 at 05:20:29PM -0400, Cole Robinson wrote:
> Currently, the volume creation process looks like:
> 
> - Grab the pool lock
> - Fully allocated the volume
> - 'define' the volume (so it shows up in 'virsh vol-list', etc)
> - Lookup the volume object to return
> - Drop the pool lock
> 
> The new sequence is:
> 
> - Grab the pool lock
> - 'define' the volume (even though nothing is on disk yet)
> - Drop the pool lock
> - Allocate the volume as needed
> - Regrab the pool lock
> - Lookup the volume object to return
> - Drop the pool lock (again)
> 
> Since the volume is 'defined', the user can fetch an object reference in
> a separate thread, and continually poll the 'info' command for up to
> date numbers. This also has the benefit of dropping the pool lock during
> the potentially lengthy allocation, as currently 'virsh pool-list' etc.
> will block while any volume is being allocated.
> 
> Non file volumes maintain existing behavior.
> 
> I tried to make the implementation resistant to user error: say if the
> pool is deactivated or deleted while the volume is being allocated. The
> creation process may bail out, but I couldn't produce any bad errors
> (crashing).

This really looks like a pretty good approach / implementation. It is
an effective incremental step which doesn't require new public APIs,
though I still think it'd be worth us adding an async job API to let
you make the 'create' operation itself non-blocking so you could query
status & also more importantly cancel jobs which take too long.

The one think I think it is worth doing though, would be to get
rid of the 'pool' parameter from the createVol method - I don't
like that we're passing in an unlocked object there, because it
leaves open possiblity of someone using it by accident in a future
patch. Of course you still need to support the other pool impls
which don't have this two step process which I obviously why you
left 'pool' in as a parameter.

I think we can avoid that if we switch the 2 ops around. Instead
of define + create. 

Have a method used for all pools:

  - create(conn, pool, vol)

The 'pool' is locked for entire duration of this call, adds the
vol to the list for this pool. Optionally allocates the vol in 
underlying storage. This basically unchanged from current impl

And a second optional method

  - build (conn, vol)

This is always called with 'pool' unlocked, and thus no pool is
passed in. 

So, for all the non-FS based pools no change is required - they
just carry on implementing 'create'  as per current code.

For the FS based pool, you have the same split of responsbilities
that your patch has. 'create' merely adds the vol to the list.
'build' does the actal allocation work.


As for the question of stoping/destroying a pool while a create
operation is still in progress. I think I'd be more comfortable
if we explicitly blocked this ability.

eg, in virStoragePoolObj  struct, we could have a field 

 'unsigned int asyncJobs'

and before calling 'build', increment this. after 'build' finishes,
decrement it again.

Then if a 'destroy' or 'stop' operation is done on a pool and the
asyncJobs field is non-zero, then refuse to allow it, returning a
message saying that it cannot be performed until all outstanding
jobs have completed. 

You may also want to have a boolean flag on the virStorageVolDefPtr
itself indicating whether it is still being created, and thus 
disallow the 'delete' operation.


> +    if (backend->defineVol) {
> +        virStorageVolDefPtr tmpvoldef = NULL;
> +        if (VIR_ALLOC(tmpvoldef) < 0) {
> +            virReportOOMError(obj->conn);
> +            goto cleanup;
> +        }
> +
> +        // FIXME: free on error
> +        /* Make a copy of the 'defined' volume definition, since the
> +         * original's allocation value will change as the user polls 'info',
> +         * but we only care about the initial requested values
> +         */
> +        memcpy(tmpvoldef, voldef, sizeof(*voldef));

Unless you have the boolean flag on the vol too, preventing the delete
operation, I think you'd need to do a proper deep copy here. The delete
operation would free any of the 'char *' fields which you've got a 
reference to here. 

Daniel
-- 
|: Red Hat, Engineering, London   -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org  -o-  http://virt-manager.org  -o-  http://ovirt.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-  F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

--
Libvir-list mailing list
Libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list

[Index of Archives]     [Virt Tools]     [Libvirt Users]     [Lib OS Info]     [Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Big List of Linux Books]     [Yosemite News]     [KDE Users]     [Fedora Tools]