Description of problem: On an AIO install in fedora 19 using packstack glance-api service fails to start the first time. Version-Release number of selected component (if applicable): openstack-glance-2013.2-0.8.b3.fc20.noarch openstack-packstack-2013.2.1-0.9.dev756.fc20.noarch How reproducible: always Steps to Reproduce: 1. packstack --allinone 2. Check /var/log/glance/api.log 3. systemctl status openstack-glance-api Actual results: The following errors appear in the glance-api.log and the service is dead. 2013-09-10 11:54:13.070 6219 ERROR glance.store.sheepdog [-] Error in store configuration: Unexpected error while running command. Command: collie Exit code: 127 Stdout: '' Stderr: '/bin/sh: collie: command not found\n' 2013-09-10 11:54:13.072 6219 WARNING glance.store.base [-] Failed to configure store correctly: Store sheepdog could not be configured correctly. Reason: Error in store configuration: Unexpected error while running command. Command: collie Exit code: 127 Stdout: '' Stderr: '/bin/sh: collie: command not found\n' Disabling add method. 2013-09-10 11:54:13.105 6219 WARNING glance.store.base [-] Failed to configure store correctly: Store cinder could not be configured correctly. Reason: Cinder storage requires a context. Disabling add method. 2013-09-10 11:54:13.385 6219 CRITICAL glance [-] 'glance.notifier.notify_qpid.QpidStrategy' is not an available notifier strategy. 2013-09-10 13:53:19.514 2424 WARNING glance.store.base [-] Failed to configure store correctly: Store cinder could not be configured correctly. Reason: Cinder storage requires a context. Disabling add method. 2013-09-10 13:53:19.952 2424 WARNING keystoneclient.middleware.auth_token [-] Configuring auth_uri to point to the public identity endpoint is required; clients may not be able to authenticate against an admin endpoint Expected results: Either openstack-glance needs to require the sheepdog package or the service shouldn't crash when one of the stores couldn't be loaded and just log the error (fail if no store can be loaded). Additional info: Installing the sheepdog package and restarting the service will fix the issue. If for any reason packstack installs openstack-nova before openstack-glance the sheepdog package will already be installed and the error won't appear.
I didn't mention it on the description, but the collie command is provided by the sheepdog package. Also, I've noticed that it is not openstack-nova who depends on sheepdog but the package libvirt-daemon-driver-storage which is pulled as a dependency of libvirt package. Besides that, I've found that the CentOS 6.4 libvirt package doesn't pull sheepdog, hence, it must be installed by hand to run glance-api.
Note that not the missing collie/sheepdog is causing your crash but your invalid notifier_strategy. I see the same collie warnings in my EL6.4 setup but Glance is working just fine. Still, the warnings are ugly for people who don't intend to use sheepdog. Btw, in my setup, the notifier_strategy is set to "noop". (I did not use Packstack and it seems to be the default value from the package).
happened to me on RHEL64 too: from /var/log/glance/api.log: 2013-09-15 11:40:51.699 5495 ERROR glance.store.sheepdog [-] Error in store configuration: Unexpected error while running command. Command: collie Exit code: 127 Stdout: '' Stderr: '/bin/sh: collie: command not found\n' 2013-09-15 11:40:51.700 5495 WARNING glance.store.base [-] Failed to configure store correctly: Store sheepdog could not be configured correctly. Reason: Error in store configuration: Unexpected error while running command. Command: collie Exit code: 127 Stdout: '' Stderr: '/bin/sh: collie: command not found\n' Disabling add method. 2013-09-15 11:40:51.738 5495 WARNING glance.store.base [-] Failed to configure store correctly: Store cinder could not be configured correctly. Reason: Cinder storage requires a context. Disabling add method. 2013-09-15 11:40:51.971 5495 CRITICAL glance [-] 'glance.notifier.notify_qpid.QpidStrategy' is not an available notifier strategy.
Sandro is right on that this error is not causing glance-api to die but only adding ugly errors in the log. The real cause for glance-api dying is in bug #1006484 and the fix will be released soon.
Hi I encounter this issue when selinux is set to permissive Are we sure selinux is the issue? restarting glance without !! installing sheepdog package helps I have used rhel 6.4 -selinux-policy-3.7.19-195.el6_4.12.noarch Reproduce: - updated all repos - restart the host in order for new kernel to be used. - set selinux to permissive - installed packtsack
*** Bug 1009002 has been marked as a duplicate of this bug. ***
Seems there are 3 bugs here: 1. erroneous sheepdog warnings 2. packstack setting invalid notifier strategy 3. selinux disallowing glance access to amqp -> bug #1006484
Which notifier strategy should Packstack use? IMHO 'noop' is just a placeholder doing nothing. So I guess having it as strategy seems useless in point of view of ceilometer installation.
For known_stores in glance-api-dist.conf, probably should just have "filesystem" and "http" Others can be enabled through config/installers as required. Re notifier strategy, the "qpid" config for notifier strategy looks correct (though should be in dist.conf). I think the issue here is that glance doesn't depend on the python-qpid package and so there can be an import error.
So as of openstack-glance-2013.2-0.12.rc1.el6 the only issue left is the sheepdog and cinder error/warnings in the logs on startup.
This is something that should be fixed upstream. I opened a bug suggesting to remove all stores but filesystem and http from the default stores. Enabling stores has to be explicit and manually done by the user. https://bugs.launchpad.net/glance/+bug/1255556 Closing and moving upstream.