New SDM verb: SDM.copy_data
by Adam Litke
Hi oVirt!
I wanted to take a moment to update you on a new storage verb we have
in vdsm called SDM.copy_data. The purpose of this verb is to allow
copying of data between various endpoints* on hosts that are not
necessarily the SPM. This will allow us in the future to spread the
load created by this data-intensive operation across multiple hosts
while providing a better user experience than the current SPM-driven
flow can.
Current status (Sprint 1):
- Basic SDM.copy_data is merged to vdsm
- We still use the SPM host for creating the destination disk and to
serially copy all volumes.
- The engine Move Disk flow is being adapted to use the new verb.
Patches should be merged in the next day or two.
https://gerrit.ovirt.org/#/c/61087/
- New code paths must be activated by setting a config variable:
"update vdc_options set option_value = true \
where option_name = 'DataOperationsByHSM';"
- Currently positive flows are implemented
Next steps:
- Copy progress reporting in UI
- Support negative flows (eg. lost contact with host)
- Use this verb for live storage migration flows
- Vdsm-side locking
- Select a non-SPM host (when available) for copying
Farther out:
- Support for copy in parallel
- Implement other data path verbs (eg. wipe_volume, sparsify_volume)
--
Adam Litke
8 years, 4 months
Vdsm/virt conversion to etree
by Milan Zamazal
We have been planning switching from minidom to etree in Vdsm for long
time. There are several good reasons why to make the change:
- We currently use both minidom and etree, we should use a single
library.
- etree has apparently better performance in the sense of CPU time and
RAM usage.
- etree has a simpler and more convenient API (e.g. limited XPath
support).
After various discussions and trying different draft approaches, there
is now available a complete series of patches migrating virt usage of
minidom to etree. It starts with patch
https://gerrit.ovirt.org/55769
and you can follow the whole vm-etree topic from there. Please read the
commit message in the initial patch, it explains the overall approach.
I tried to make the series reviewer friendly, for patch-by-patch review.
It's still draft, not finished, missing many tests, etc., but it already
basically works. If you have any serious problems with the approach to
the conversion or its general directions, tell me before we start
discussing details. Just don't forget we must perform the conversion in
some reasonably safe way and it must be possible to review it.
Please note the series doesn't try to solve all the world/Vdsm problems,
it focuses just on the etree conversion. Further changes may be
implemented in followup patches. I already implemented some
post-refactoring in vm-etree-refact topic starting with
https://gerrit.ovirt.org/60940.
Thanks to Francesco Romani for cooperation and help with the work.
8 years, 4 months
Engine 4.0.2 stable branch created
by Tal Nisan
Hi everyone,
A branch for 4.0.2 was created today (ovirt-engine-4.0.2).
Note that all patches for 4.0.2 bugs are to be backported twice (4.0 and
then 4.0.2), 4.0.3 bugs can now be backported as well to 4.0 only.
T.
8 years, 4 months
[VDSM] vdsm_master_verify-error-codes_created running on master - why?
by Nir Soffer
Hi all,
Since yesterday, vdsm_master_verify-error-codes_created job is running
on master.
I guess that this was a unintended change in jenkins - please revert this
change.
If someone want to add a job for vdsm master, it must be approved by
vdsm maintainers first.
The best would be to run everything from the automation scripts, so
vdsm maintainers have full control on the way patches are checked.
Thanks,
Nir
8 years, 4 months
Arrays.asList and GWT
by Roman Mohr
Hi List,
Seems like GWT can't handle Arrays.asList not very well. Sharing this
in case others have similar issues:
I was working on a very strange deserialization bug in combination
with GWT [1]. We had a deserialization problem after loading a VM and
trying to send the VM back to the backend.
The error looked like this:
2016-06-28 12:17:36,206 ERROR [io.undertow.servlet] (default
task-58) Exception while dispatching incoming RPC call:
java.lang.NumberFormatException: For input string: "undefined"
It turned out that GWT could serialize an ArrayList backed by an
Integer[] array but it could not deserialize it afterwards.
In my case a list returned by ResultSet#getArray in the DAO layer was
sent to GWT:
entity.setCpuIds(Arrays.asList((Integer[])
rs.getArray("cpu_core_ids").getArray()));
After copying the content over into a regular ArrayList everything
worked fine again.
entity.setCpuIds(new ArrayList<Integer>(Arrays.asList((Integer[])
rs.getArray("cpu_core_ids").getArray())))
Here is also the gerrit patch set [2].
So if you see strange GWT RPC errors like the one above look out for
"Arrays.asList" in you code :).
Best Regards,
Roman
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1350861
[2] https://gerrit.ovirt.org/#/c/61038/2
8 years, 4 months
oVirt 4.0.2 RC1 merge / branch / tag / bugzilla reminder
by Sandro Bonazzola
All stable branch maintainers, please make sure to
- merge all relevant open bugs until Wednesday morning 11:00 AM TLV time.
For each package that need to be built (i.e oVirt product) please make sure
every bug in MODIFIED has the right Target Release and Target Milestone.
A Target release should state the version of the package you're building
and should include the same version you used for the tag you just used for
this build. (e.g. for ovirt-engine, tag: ovirt-engine-4.0.2, tr: 4.0.2)
A list of bugs that require attention is here:
-
https://bugzilla.redhat.com/buglist.cgi?quicksearch=target_milestone%3A4....
--
Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com
8 years, 4 months
oVirt 4.0.2 RC1 build planned
by Sandro Bonazzola
Fyi oVirt developers,
An oVirt build is planned for this Wednesday 11:00 AM TLV time (10:00 AM
CEST).
Taking into consideration the time it takes for Jenkins to run a full CI
everything need to be backported by Tuesday 11PM.
Please make sure to mark as verified and CR +2 so it will be ready for
merging Wednesday morning.
A new branch ovirt-engine-4.0.2 will be created just before the build, an
announce will be sent to remind to backport to that branch after its
creation.
A list of pending blockers is available here:
https://bugzilla.redhat.com/buglist.cgi?quicksearch=target_milestone%3A4....
--
Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com
8 years, 4 months
[VDSM] all check-patch jobs fail on fedora
by Nir Soffer
Hi infra,
Vdsm tests fails now on fedora since they are running now on fedora 24,
while vdsm automation was not updated yet to fedora 24.
See http://jenkins.ovirt.org/job/vdsm_master_check-patch-fc24-x86_64/
The problem patch seems to be:
commit 4a40459191b39bafb6d14c34d6346bb152bc55da
Author: Sandro Bonazzola <sbonazzo(a)redhat.com>
Date: Thu Jul 14 13:58:44 2016 +0200
vdsm: move from fc22 to fc24
- dropped fc22 gone EOL
- moved master to fc24
Change-Id: Ifce815087c649404b44f657911702735c53f75c3
Signed-off-by: Sandro Bonazzola <sbonazzo(a)redhat.com>
Please revert this patch.
We must add support for fedora 24 to vdsm first. When the tests
are stable we can switch to fedora 24.
Thanks,
Nir
8 years, 4 months
