This patch reverts 13d80020. The help message should be added
in oslo-incubator since generate_sample.sh is synced from there also.
Resolves bug #1204204
9f2b93b Sample config file generator clean up
33f9f78 Replaces relative path to absolute path in conf file generator
c97e17b Refactors if statement in config generator
8b11fda Provide useful defaults for config generator
4c02e0a Unset OS_xx variable before generate configuration
Change-Id: I96fcb39c0c620c147fea5c96416e506a758619ce
system_metadata is still being lazy loaded as a part of healing the
instance info_cache. A previous attempt to fix this issue missed the
more important call.
It's actually more efficient to not load the system_metadata when getting
the list and allow the 1 instance to lazy load. For the cached instances,
we will ask for system_metadata when getting a fresh copy from the DB.
Fixes bug 1223969
Change-Id: Ib95e868135ed2b36af7ddbcfb8d5905bfe14d1fe
When resize revert one VM instance, if the NFS is slow, the NFS
mounted on source host might not respond quickly. This will cause
that even if the VM instance configuration files was deleted from
resize target host, the NFS on source host may still think the
files were not deleted. Then in driver.py::finish_revert_migration,
the driver try to delete the file again, this will lead nova compute
throw exception.
The fix was ignore exception for such case: if failed to delete the
instance with error no as 2 (No such file or directory)
Fix bug 1206312
Change-Id: I89990c1750a7fd7f421cab03780e7704ee661de1
Before objects, a notification would be sent for any change to an
instance and the notifications.send_update method would handle any
filtering that needed to be done prior to sending the event.
Currently, the instance objects code is pre-filtering the notification
by only checking if vm_state or task_state changed which is a
regression.
This patch removes the filter in the instance objects code so that
notifications are again sent for any changes to the instance and lets
the notifications.send_update method handle the event as it did before.
Closes-Bug: #1220186
Change-Id: Ib6ceedfe55b021be0fe7fa910e97d46c0cd8f0cd
The help message for cinder_ca_certificates_file and
neutron_ca_certificates_file spelled certificates wrong.
As this is user facing, we should fix it.
Fixes bug 1227780
Change-Id: If9372fffb655d21069c005351ebb593d8e16e056
The nova.tests.objects.test_quotas module had a print statement which is
causing a hacking H233 check failure. It was probably just left there
from debugging the code while it was under development, so this patch
simply removes it.
Closes-Bug: #1227655
Change-Id: I25296d7f4a9d2242bfe3ed2e00df6d840a7b19cd
if there are no enabled_services, then zone_hosts is very likely
to be empty, which means zone_hosts['nova'] will fail. So let's
use get with a default of empty array to cover this case. Added
a fix/test for v3 API as well.
Fixes LP# 1226791
Change-Id: I602c96242291fd9a4bdf67af6445aa935ec7f7e3
NoDBTestCase is faster because it skips initializing a database for
tests which don't actually talk to the database. Its also nicer
because its more explicit about the dependancies of a given test.
This is a first pass at moving test cases over to the database
lacking version, but further work is required here.
Change-Id: Ia28de4ebea4f354cba06bcdc66bb01d4b332943a
Partial-Bug: 1227473
MultiprocessWSGITest.test_terminate_sigterm
fails because the wait for workers to die is unreliable.
Parent process can now wait on child or grand-child
processes to complete before moving on. This allows
for grand-child (aka worker) processes to take as much
time as they need to shutdown. NOTE: If the processes
still take too long, they will be killed by the test
framework throwing the "TIMEOUT" exception.
Reverts to using a ping signal to child process
to determine if the child process is live. Waits for
the signal call to fail, signaling process death.
closes bug #1218133
Change-Id: I08e17005508fe412f462773f9ab212cf0f10382f
This commit addresses the situation in which a compute node is deleted, but
its compute node stats still remain in the database as **not** deleted. This
causes ill side effects in compute_node_get_all when it's retrieving host stats
as it doesn't expect there to be compute node stats for which there is no
corresponding compute node (i.e., causing some nodes' stats to be empty).
As such, when a compute node is deleted, its stats should also be implicitly
deleted.
The new test case that's been created fails without the code changes, which
illustrates the problem that compute node stats are empty when they should
not be.
Also included is a simple DB migration script that will update old stats that
were not marked soft-deleted as they should have been.
Change-Id: Ief0f7cf1a506e71898b5a45a0513d34167432d67
Closes-Bug: #1224712
If rescue setup fails, detach the root disk from the rescue VM so it
doesn't get nuked by the undo stack logic in spawn()
bug 1227898
Change-Id: I8864b4c524f04f7b387be1018fea9919a461c48b
The default controller for a Windows 7 virtual machine is the
LSI Logic SAS. When the disk adapter type is specified as 'lsiLogic'
via glance image metadata, VMware driver defaults to LSI Logic Parallel
controller. With this fix the driver will honor a property 'lsiLogicsas'
to distinguish the right adapter and enable booting Windows instances.
Fixes: bug #1187853
Change-Id: Idb36a784a31e4938e8951adeb83335b3890dac85
Zhongyue Luo