This changes the thread pool usage of the ComputeManager to go through
the concurrency mode aware util functions.
The concurrent live migration pool had a seemingly unlimited option
when configured with value 0, but in reality GreenThreadPool has a
default worker size of 1000. In reality it is almost never right to
have more than one live migration running concurrently. Also with
native threading having 1000 worker is just too costly. So we
decided to deprecate the value 0 and changed the implementation of
unlimited to mean 5 threads in native threading mode. We kept the 1000
greenthread in eventlet mode for backward compatibility.
The _sync_power_states periodic task also spawn tasks for each instance
to be synced. As it uses a shared data structure across these tasks
and the caller a lock is needed to avoid race conditions.
Also the default pool size is 1000 for these tasks in our configuration.
That would use a lot of memory on a busy host in native threading mode.
So we changed the default value from 1000 to 5.
Change-Id: I9567d5fabdf086b5d0493103d9f6bde4f66af387
Signed-off-by: Balazs Gibizer <gibi@redhat.com>
This is a follow up for the release notes added in the commit
35207ee8b5 that changed the default mode
for the scheduler and the API services. At that time we missed to note
the upgrade impact of such change. So this patch extends the reno with
an upgrade note.
Change-Id: I280e7eb9c1da6eeaf50e96e8b19e296961f2651a
Signed-off-by: Balazs Gibizer <gibi@redhat.com>
Python 3.14 tightened call semantics, and testtools 2.8.2 no longer
tolerates unittest-style kwargs, when Nova tests were mixing
testtools.TestCase with unittest. This patch fixes this by using
positional arguments for assertRaises.
Signed-off-by: Thomas Goirand <zigo@debian.org>
Change-Id: Ic589880d4759336ab6ceec5057eafe223f692caa
The threading Lock object cannot be copied so the ProviderTree cannot be
deep copied in threading mode. This patch adds custom pickling /
copying handlers to the ProviderTree class to ignore the lock from the
source and readd it in the destination object during pickling /
copying.
Note that ProviderTree uses a named lock which means all the instance of
the ProviderTree object in the same process uses a shared lock object.
The copy handlers ensures the same behavior during copying so the copy
will use the same shared lock as well.
Signed-off-by: Balazs Gibizer <gibi@redhat.com>
Change-Id: I084e72ce81dd456d67c9046f37d1ccb01b9fa4ba
This has been on the TODO list for over 8 years. Time to resolve it.
Change-Id: Ic5367a1042ae5da974a25c135b58a1ebf22eb531
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
Our event listener depends on devices having an alias set. We add an
early assertion to prove this is the case. In real-life, this will
always be the case since the devices we can detach - like interfaces and
disks - are among the lists of devices that libvirt will automatically
generate an alias for if the user (nova-compute, in this case) doesn't
provide their own [1]. However, many of our tests were not doing this so
we must update our LibvirtFixture in particular to start doing so.
[1] https://github.com/libvirt/libvirt/blob/v11.10.0/src/qemu/qemu_alias.c#L692-L786
Change-Id: Id98d8029af673ffa89d6472be98f90a6f0975511
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
Rather than setting dunder attributes manually.
Change-Id: I71bf90f5deacb11cbf6e39dfe459edd5a03f3d75
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
There were libvirt disk attach / detach unit test cases that mocked
threading.Event.wait to speed up the test execution as the libvirt event
based device detach has a 20 seconds timeout.
However this mock is very wide. It effects all the threading.Event usage
during the test case. As we are switched to the native threaded backed
of oslo.service we got an LoopingCall implementation that relies on
threading.Event to signal when the call is finished. In these test cases
the wide mock broke this logic and the LoopingCall signalled "done" while
the actual call in the background was still running making that call
leaking out from the test case and therefore under the fixtures the test
case tore down after the test case finished. This caused that later the
call hit non mocked code paths like importing the real libvirt lib
instead of using our libvirt fixture causing late failures.
The fix is to remove the wide mock and instead change the detach timeout
in these test cases.
Later follow ups will check put protection in place to avoid
re-introducing this wide mock.
Closes-Bug: #2136815
Change-Id: I3f40d9dad6ef87e6fa4db0ea4065a77421e8c271
Signed-off-by: Balazs Gibizer <gibi@redhat.com>
Introduce the new Host.supports_mem_encryption which tells whether
the host supports memory encryption. This allows us to centralize
and generalize the check logic for additional mechanisms such as
Arm CCA, planned to be added in the future.
Implements: blueprint generalize-sev-code
Change-Id: If020c71bd4962c6ca96e042592854e57d9a7dcce
Signed-off-by: Taketani Ryo <taketani.ryo@fujitsu.com>
Follow-ups for the original fix to bug #2131960. Includes an update to
make tests match Nova style (ddt) and another to unify state constant
code between Ironic and Nova.
The new ironic_states.py was copied from ironic/common/states.py from
commit 2622864b70bc66ec711045a3a0bf30478b27b9cd in ironic.
Change-Id: I49fba64da039e5c54967b6f5c56cb44927f815c3
Signed-off-by: Jay Faulkner <jay@jvf.cc>
The json format of policy files was deprecated long ago and is being
removed completely. Mention policy.yaml instead.
Change-Id: I9f6f11fd563d39a4fc1b940c67084292de7c56e3
Signed-off-by: Takashi Kajinami <kajinamit@oss.nttdata.com>
Without this, we won't notice errors raised in the operation thread.
Before 1cd1c472bd the unit test actually
forced such errors to be raised even if in the real code it would
never be raised. But that patch fixed the unit test fixture to be more
realistic without realizing that such fixture error also means that we
might have wrong assumptions about the code under test.
Now we know that exception from the live migration thread was
never raised back to the monitor thread. To improve logging we added a
future.result() call after the main monitoring code finished.
Also the code had complex way to signal the monitoring thread that the
migration thread returned early by registering a callback on the
migration thread and setting an event. This can be simplified to just
check the status of the future of the migration thread. So the event and
the callback is removed.
All this was found because commit 25fbf32f22
missed to add the new parallel arg to the mock of guest.migrate()
on master, but the exception was never propagated to the unit test on
master. Backporting that change showed that in the old unit test env
there is a valid exception.
Co-authored-by: Dan Smith <dms@danplanet.com>
Change-Id: I22683ad5118796c6406f80d8726053afa84fff56
Signed-off-by: Dan Smith <dansmith@redhat.com>
Signed-off-by: Balazs Gibizer <gibi@redhat.com>
This was missed in commit 25fbf32f22
because of a bug in our _live_migration_operation() post-eventlet
handling.
Change-Id: I39a7d6ebd72d9938bcb60143dfc50bd6a9c994b0
Signed-off-by: Dan Smith <dansmith@redhat.com>
elajkat