To revisit the DEFINED, UNDEFINED topic - I have been giving this some thought, and it seems like it is redundant information to me. Doesn't the existence of a config file define the difference between a transient and a persistent domain? Therefore this could be tracked solely with the enum values we have? If a transient domain becomes persistent via adding a config file, the ADDED event would be emitted, denoting a config file came into being, and by definition changing this domain from transient, to persistent. I'm not sure I understand the need for these events. Moving on to the shutdown reason discussion - I can certainly see the need to know why a domain stopped. That said - would it not be more consistent with the API to introduce an event "sub-type" enum to be passed alongside of the event-type, passed as a second integer? This would reduce the size of the wire protocol, not needing to pass the full character string, as well. This would trivially allow us finer grained notifications within a particular event. A downside is that it may become a bit noisy for clients who may not be interested in all of the sub-events. Perhaps we should add some sort of filtering mechanism? -----Original Message----- From: libvir-list-bounces(a)redhat.com on behalf of Daniel P. Berrange Sent: Fri 11/14/2008 6:18 AM To: libvir-list(a)redhat.com Subject: [libvirt] RFC: Add a 'reason' argument for domain events ? We currently have a set of domain lifecycle events VIR_DOMAIN_EVENT_ADDED = 0, VIR_DOMAIN_EVENT_REMOVED = 1, VIR_DOMAIN_EVENT_STARTED = 2, VIR_DOMAIN_EVENT_SUSPENDED = 3, VIR_DOMAIN_EVENT_RESUMED = 4, VIR_DOMAIN_EVENT_STOPPED = 5, VIR_DOMAIN_EVENT_SAVED = 6, VIR_DOMAIN_EVENT_RESTORED = 7, And a couple more I proposed last week to allow distinguishing of new domains which are transient vs persistent VIR_DOMAIN_EVENT_DEFINED = 8, VIR_DOMAIN_EVENT_UNDEFINED = 9, Previously Stefan has suggested we should consider having an event for migration, and though I rejected that at the time, I'm now inclined to agree that this info would be useful here. I'm also thinking I'd like to have more information about STOPPED & STARTED events in general. eg, there are a number of reasons why an domain may have started - explicitly booted on the host - restored from a saved image - incoming migration operation and there are a number of reasons why a domain might have stopped - forcably destroyed by host admin - shutdown by host admin - shutdown by guest admin - host emulator process crashed - killed by mgmt after host emulation hung - migrated to another host - saved to a memory image We have explicit events for the SAVED/RESTORED reasons, but what should we do about the other reasons ? One option is to add alot more events VIR_DOMAIN_MIGRATED_IN (migrated to another node) VIR_DOMAIN_MIGRATED_OUT (migrated from another node) VIR_DOMAIN_SHUTDOWN (graceful shutdown by host admin) VIR_DOMAIN_DESTROYED (force destroyed by host admin) VIR_DOMAIN_CRASHED (guest kernel crashed) VIR_DOMAIN_HUNG (host emulator hung) leaving STOPPED to just be a generic stop event, with no particular reason. The downside with this, is if an application just wants to know about whether a domain shutdown, not why, then they have to track lots and lots of events. Also, not every driver would be able to provide all of these events, so would often have to fallback on a generic STOPPED event So one alternative is to provide a generic 'char * reason' with each event with provides scope on the cause of the lifecycle operation. So you'd get VIR_DOMAIN_STOPPED ("crashed", "shutdown", "destroyed", "quit", "hung", "migrated", "saved") VIR_DOMAIN_STARTED ("booted", "migrated", "restored") nb, we could remove the explicit SAVED and RESTORED events in this style. With such a 'reason' arg, we could possibly avoid adding the DEFINED and UNDEFINED events I suggested. Instead, adding a reason for ADDED and REMOVED events, VIR_DOMAIN_ADDED ("started", "defined") VIR_DOMAIN_REMOVED ("shutdown", "undefined") which lets you distinguish transient from persistent domains. The downside of this though, is that we can't explicitly track when a configuration file is 're-defined', eg adding a config file for an existing running guest. 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(a)redhat.com https://www.redhat.com/mailman/listinfo/libvir-list

On Fri, Nov 14, 2008 at 11:18:48AM +0000, Daniel P. Berrange wrote: [...] yes that makes using the API way harder than it should, though enumerating the cases: in a switch is not that hard... I like this, I wonder if a free form part to the message might make sense, even if we don't have currently a way to pass this say from virsh command line. Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/