[ovirt-devel] Push notifications in 4.0 backend

[Martin Betak]

Hi All,

I have created a PoC patch [1] demonstrating the idea of annotating
basic CRUD commands to publish CDI events. It is not meant as 100%
solution, but as a simplification of the common use cases when
one would inject CDI event with given qualifier and fire it after
successful completion of transaction.

The usage of this annotations is demonstrated on several basic CRUD
commands in [2] on StoragePool, VDS, VDSGroup, .etc

As always, comments and suggestions are very welcome!

Thank you and best regards,

Martin 

[1] infra: https://gerrit.ovirt.org/#/c/48696/
[2] usage: https://gerrit.ovirt.org/#/c/48697/



----- Original Message -----
> From: "Vojtech Szocs" <vszocs at redhat.com>
> To: "Martin Betak" <mbetak at redhat.com>, "Einav Cohen" <ecohen at redhat.com>
> Cc: "engine-devel at ovirt.org" <devel at ovirt.org>, "Roy Golan" <rgolan at redhat.com>, "Roman Mohr" <rmohr at redhat.com>,
> "Michal Skrivanek" <mskrivan at redhat.com>, "Piotr Kliczewski" <pkliczew at redhat.com>, "Tomas Jelinek"
> <tjelinek at redhat.com>, "Alexander Wels" <awels at redhat.com>, "Greg Sheremeta" <gshereme at redhat.com>, "Scott
> Dickerson" <sdickers at redhat.com>
> Sent: Friday, November 13, 2015 9:31:45 PM
> Subject: Re: Push notifications in 4.0 backend
> 
> Hi everyone,
> 
> assuming that 4.0 UI will be based on the existing GWT technology,
> I'd like to improve two things which I believe are very important:
> 
> #1 goal: improve GWT compilation times
>    - don't use standard GWT i18n mechanism which yields separate
>      permutation vector, but use our own i18n mechanism instead
>    - in practice, compiling for X browsers and Y languages should
>      result in GWT compiler processing X permutations (not X * Y)
>    - this will also directly impact GWT debug performance, making
>      GWT debugging experience less painful for developers
> 
> #2 goal: improve UX related to backend operations
>    - replace periodic polling with push notifications that inform
>      UI of changes in oVirt "system" as they happen
>    - in practice, UI becomes reactive instead of proactive, which
>      has several benefits (reduced HTTP load on Engine being the
>      most obvious one)
> 
> So what Martin wrote in email below directly relates to #2 goal.
> 
> Push notifications improve user experience regardless of specific
> UI technology, regardless of whether we improve existing REST API
> (e.g. introduce data aggregations) or not.
> 
> For me, it's a big +1.
> 
> Having BLL commands firing CDI events upon execution makes sense.
> That said, I'd suggest to start with a simple implementation and
> proceed from there.
> 
> What Martin suggested:
> 
>   void onVmChanged(@Observes @Updated VM vm)
> 
> could be even simplified into:
> 
>   void onCommandExecuted(@Observes @CommandExecuted UpdateVmCommand cmd)
> 
> and still it would bring value to the general idea, which is the
> ability to detect changes in oVirt "system" as they happen along
> with the ability to react upon such changes.
> 
> I'm interested what Engine backend maintainers' thoughts are.
> 
> Regards,
> Vojtech
> 
> 
> ----- Original Message -----
> > From: "Martin Betak" <mbetak at redhat.com>
> > To: "engine-devel at ovirt.org" <devel at ovirt.org>
> > Cc: "Roy Golan" <rgolan at redhat.com>, "Roman Mohr" <rmohr at redhat.com>,
> > "Michal Skrivanek" <mskrivan at redhat.com>,
> > "Piotr Kliczewski" <pkliczew at redhat.com>, "Vojtech Szocs"
> > <vszocs at redhat.com>, "Tomas Jelinek" <tjelinek at redhat.com>
> > Sent: Wednesday, November 11, 2015 4:34:11 PM
> > Subject: Push notifications in 4.0 backend
> > 
> > Hi All,
> > 
> > I would like to take this opportunity to start a discussion
> > about the possibility of implementing a user facing change notifications.
> > 
> > The benefit of this would be to remove the need for periodic polling
> > from frontends and other services that consume our REST API.
> > 
> > Also implmenting a common infrastructure on the backend for event
> > notifications (e.g. CDI events) would further reduce the internal
> > need for polling the DB by the backend itself, maybe even reducing
> > the need to use DB for some things and just keep them in memory and
> > updated by CDI event observers.
> > 
> > There are many solutions how to provide the user-facing part of the
> > notifications:
> > Doctor Rest, MQTT, websocket, server-sent events, ... .  Ideally these
> > notifications
> > should be consumable both by web browser (webadmin/userportal) but also by
> > other services (such as ManageIQ), or other REST clients such as moVirt
> > android client.
> > 
> > But regardless of the chosen user-facing transport, I believe a common
> > infrastructure
> > can be implemented on the BLL layer with the usage of CDI events fired from
> > commands.
> > I see 2 major sources of changes in the engine (please correct me if this
> > is
> > wrong):
> > 1) CRUD & management commands
> > 2) Vms/Hosts monitoring
> > 
> > the changes originating from 2) are AFAIK very localized and not so
> > numerous
> > so a manual
> > firing of appropriate events for VMs and Hosts could be added here.
> > 
> > The 1) case is more extensive in terms of required code changes. While a
> > manual solution
> > would still be feasible, I believe there is place for a more
> > automated/declarative way.
> > 
> > One solution for 1) that comes to my mind are simple command-level
> > annotations covering the
> > Created, Updated, Removed (C, U, and D from CRUD) cases. The goal here is
> > to
> > fire the
> > appropriate CDI events when an entity is created/updated/deleted. Since
> > commands usually
> > contain getters for entities they work with (such as getVm(), getVds(),
> > getStorageDomain() ...)
> > It should be sufficient for the most common simple cases (of course this
> > will
> > not cover
> > everything) to use annotation @Creates, @Updates, @Removes on the commands
> > classes, where
> > parameters of the annotation should specify the getter method that returns
> > the affected entity
> > (VM/VDS/StorageDomain...). This could be specified by the entity class
> > token
> > or method name
> > (depending on the level of "magic" one prefers :-) and the CommandBase
> > infrastructure would
> > then collect those annotations and upon successful completion of the
> > command
> > fire the
> > appropriate events.
> > 
> > Example #1:
> > @Updates('getVm') // or @Updates(VM.class)?
> > public class UpdateVmCommand<...>  extends VmManagementComandBase ...
> > 
> > Note that since Java 8 we have repeatable annotations so we can have more
> > complex commands
> > that affect more entities.
> > 
> > Example #2:
> > @Updates(Vm.class)
> > @Updates(VmTemplate.class)
> > // possibly also some @Creates and @Removes annotations or their
> > combination
> > public class ContrivedExampleCommand extends SomeCommandBase
> > 
> > the infrastructure would then look upon successful completion of the
> > command
> > on the getVm()
> > and getVmTemplate() methods, invoke them, determine the resulting types of
> > entities VM and VmTemplate
> > and since the annotations used were @Updates fire CDI event equivalent to
> > 
> >   @Inject
> >   @Updated // our custom CDI qualifier
> >   Event<VM> vmChangedEvent;
> > 
> > and anologously for VmTemplate.
> > 
> > But regardless of the exact implementation of the CDI event firing: whether
> > manual, the above
> > proposal, or some crazy usage of AspectJ - the interface for the rest of
> > the
> > code should always
> > be the like this:
> > 
> > public void onVmChanged(@Observes @Updated VM vm) {
> >     // ....
> > }
> > 
> > On top of this, I believe, we can build the user-facing part of push
> > notifications and also
> > improve our existing backend code.
> > 
> > Thank you for reading this long email and I welcome any comments or
> > counter-proposals you
> > might have on this topic :-)
> > 
> > Best regards,
> > 
> > Martin
> > 
>