cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alex Ough <alex.o...@sungardas.com>
Subject Re: Control event publishing in multi region setups
Date Sat, 10 May 2014 15:55:11 GMT
I really don't know why you guys are making it complicated.
The class has two different methods, one with 'event' decorator and the
other without it.
So the callers know which method to call depending on their needs.
And the each method will be called accordingly.


On Sat, May 10, 2014 at 6:13 AM, Alex Huang <Alex.Huang@citrix.com> wrote:

>  -1
>
>
>
> I do not believe in the argument that says “since there’s existing bad
> code, then I can check in code that also causes regressions for users.”  If
> that’s the case, what’s the point of the review?
>
>
>
> We’ve offered a path forward already.  Please reconsider that.
>
>
>
> --Alex
>
>
>
> *From:* Alex Ough [mailto:alex.ough@sungardas.com]
> *Sent:* Friday, May 9, 2014 9:14 PM
> *To:* Alex Huang
> *Cc:* Murali Reddy; Alena Prokharchyk; Kishan Kavala;
> dev@cloudstack.apache.org
>
> *Subject:* Re: Control event publishing in multi region setups
>
>
>
> Are we going to rolling this out?
>
>
>
> On Thu, May 8, 2014 at 2:28 PM, Alex Ough <alex.ough@sungardas.com> wrote:
>
>  That's why there are 2 methods, one is that generates events and the
> other not and there are already a few public methods without event
> decoration.
>
>
>
> On Thu, May 8, 2014 at 2:25 PM, Alex Huang <Alex.Huang@citrix.com> wrote:
>
>  Alex,
>
>
>
> I did read this when you first proposed.  I do understand the two
> implementation.
>
>
>
> I understand that #2 is not activated via events but it doesn’t mean #2
> can just don’t generate events.  The blocker is precisely with the last
> sentence in #2 where it states #2 doesn’t generate an event when “it
> creates/updates/removes the resource in the local region”.
>
>
>
> Perhaps an example would make this more clear.
>
>
>
> Someone who deploys CloudStack sets up a process to listen to account
> events.  It is a simple audit process whose job is to verify that an
> account created in CloudStack is actually in their own billing database.
>  The fact that #2 doesn’t generate an event would mean this process would
> be broken for them.  This is the regression that causes the blocker.
>
>
>
> --Alex
>
>
>
>
>
> *From:* Alex Ough [mailto:alex.ough@sungardas.com]
> *Sent:* Thursday, May 8, 2014 11:02 AM
> *To:* Alex Huang
> *Cc:* Murali Reddy; Alena Prokharchyk; Kishan Kavala
>
>
> *Subject:* Re: Control event publishing in multi region setups
>
>
>
> Alex,
>
>
>
> I think you really review the wiki (
> https://cwiki.apache.org/confluence/display/CLOUDSTACK/Domain-Account-User+Sync+Up+Among+Multiple+Regions)
> or the implemented codes.
>
>
>
> To help you understand, there are 2 synchronizations supported in this
> feature.
>
>
>
> 1. real time sync : This is what you may imagine and event based. This is
> sending requests when they are created/updated/removed in the local region
> by subscribing their events.
>
>
>
> 2. full scan : This is NOT related with events and it is to cover when the
> #1 sync is failed with any reason like network failures. With interval, it
> just scans all resources and compare them with ones in remote regions and
> if there is any missing in the local region, it creates/updates/removes the
> resource in the local region and the NEW METHODS I need are called because
> it is local region only and no need to have events.
>
>
>
> I'm hoping you understand the feature a little more and let me know if you
> need more information.
>
> Thanks
>
> Alex Ough
>
>
>
>
>
> On Thu, May 8, 2014 at 1:43 PM, Alex Huang <Alex.Huang@citrix.com> wrote:
>
>  Hi Alex,
>
>
>
> Please know that the contribution is much appreciated.  It is not a case
> of whether or not Alena “wants” or “doesn’t want” to approve the review.
> She can only approve if the design is sound and has no regression for
> existing deployments of CloudStack.
>
>
>
> This is a blocker because not publishing events when an account is
> propagated is actually an “incorrect” behavior for CloudStack.  Any
> functionality that acts on an account creation within the region will face
> regression.  That’s why it is not “an additional feature” and must be
> fixed.  Think of SunGuard itself.  If it was depending on the account
> creation event and the next version of CloudStack suddenly doesn’t generate
> the event consistently, would it not consider this a bug and ask us to fix
> it?
>
>
>
> I do understand the time consuming nature of providing patches and merging
> code.  Alena tells me that she has reviewed the code and she thinks the
> design is fine except for this one item.  If we can commit to fix this
> problem after the code is checked in, we can check it in now just so you
> don’t have to do another round of merge and review for the part that is
> working.  But the fix will need to be in before the code is released or
> else we might have to revert this checkin.  What do you think?
>
>
>
> --Alex
>
> P.S. I’m not sure why this is not on the dev list.  We should bring this
> back.
>
>
>
> *From:* Alex Ough [mailto:alex.ough@sungardas.com]
> *Sent:* Wednesday, May 7, 2014 4:58 PM
> *To:* Murali Reddy
> *Cc:* Alena Prokharchyk; Alex Huang; Kishan Kavala
>
>
> *Subject:* Re: Control event publishing in multi region setups
>
>
>
> All,
>
>
>
> Alena doesn't want to approve my implementation because of this email
> thread, but I'm frustrated and not sure why this is a blocker.
>
> What I did was just created another method without an event tag like the
> one already existing in 'AccountManagerImpl' class as below.
>
>
>
> @Override
>
> public boolean enableAccount(long accountId)
>
>
>
> And if we need this feature, we really need to create a new jira instead
> of adding it to already existing one
>
> so that we can discuss options to find a best solution.
>
>
>
> It's been a really long path mostly because of miscommunications, and I
> really want to wrap this up without adding a new feature that is not
> existing.
>
>
>
> Let me know what you think.
>
> Thanks
>
> Alex Ough
>
>
>
> On Wed, May 7, 2014 at 10:29 AM, Murali Reddy <Murali.Reddy@citrix.com>
> wrote:
>
>  I don’t think we need to bring back reverted changes, as we want all the
> events generated should be published all the time with in the region. I
> agree with Alex Huang, that we could actually add details (originating
> region) to the account indicating source region where account is created.
> Details particular to an event published on the event bus is a JSON object
> so we can add additional details. Also steps listed out by Alex should
> prevent from cyclic propagation.
>
>
>
> Alex Ough,
>
>
>
> Suggested steps below by alex should work for you. Do you see any problem
> with that?
>
>
>
> Thanks,
>
> Murali
>
>
>
> *From: *Alena Prokharchyk <Alena.Prokharchyk@citrix.com>
> *Date: *Wednesday, 7 May 2014 5:56 AM
> *To: *Alex Huang <Alex.Huang@citrix.com>, Alex Ough <
> alex.ough@sungardas.com>, Kishan Kavala <Kishan.Kavala@citrix.com>,
> Murali Reddy <murali.reddy@citrix.com>
>
>
> *Subject: *Re: Control event publishing in multi region setups
>
>
>
> Alex (Huang), thanks for commenting.  As a conclusion – we should never
> omit event firing when submit create/update.
>
>
>
>
>
> Kishan/Murali, can you please help Alex (Ough) to figure out how to
> implement the behavior Kishan reverted. Kishan, can you check with Murali
> how to bring back your reverted changes for the API to make it work with
> the new events framework?
>
>
>
> Thank you,
>
> Alena.
>
> *From: *Alex Huang <Alex.Huang@citrix.com>
> *Date: *Tuesday, May 6, 2014 at 10:17 AM
> *To: *Alena Prokharchyk <alena.prokharchyk@citrix.com>, Alex Ough <
> alex.ough@sungardas.com>
> *Cc: *Kishan Kavala <Kishan.Kavala@citrix.com>
> *Subject: *RE: Control event publishing in multi region setups
>
>
>
> Hey guys,
>
>
>
> I’m not sure we’re all on the same page.
>
>
>
> First, the event must always be published, regardless of if it was
> propagated from another region or created originally in that region.  The
> reason is there may be other code interested in acting on account creation
> in a region.  We just need to provide a way for Alex’s code to determine
> that the account is propagated rather than created originally in the
> region.  You don’t need details in the event for that.
>
>
>
> The propagation code can do the following.  It’s probably already doing
> that.
>
>
>
> 1.       Listen for the account creation event.
>
> 2.       Upon receiving an account creation event, retrieve the account
> to check if the account is propagated or created.
>
> 3.       If propagated, then don’t propagate or maybe even signal back
> that the propagation is done for this region (depending on the propagation
> logic).  If created, then propagate to other regions.
>
>
>
> Now the detail is in how do we know if an account is created or
> propagated.  For that, it can be done in a number of ways.  I’m open to
> which method.  I would suggest that we add two fields to account:
> origination region and original uuid.  The create account API takes an
> optional fields for the origination region and origination account uuid.
>  If these two parameters are not set in the API, the API set the
> origination region to the current region and the original uuid to the uuid
> of the account.
>
>
>
> Sorry for the confusion here.  I had thought Kishan added this but
> apparently it has been reverted.
>
>
>
> --Alex
>
>
>
> *From:* Alena Prokharchyk
> *Sent:* Tuesday, May 6, 2014 9:57 AM
> *To:* Alex Ough
> *Cc:* Kishan Kavala; Alex Huang
> *Subject:* Re: Control event publishing in multi region setups
>
>
>
> Ok, thank you Alex, so looks like there is no other workaround as of now
> rather than introducing the new methods to the managers. Just go ahead and
> submit the rest of the fixes for both review tickets, and I will commit the
> patch after that.
>
>
>
> -Alena.
>
>
>
> *From: *Alex Ough <alex.ough@sungardas.com>
> *Date: *Tuesday, May 6, 2014 at 9:48 AM
> *To: *Alena Prokharchyk <alena.prokharchyk@citrix.com>
> *Cc: *Kishan Kavala <Kishan.Kavala@citrix.com>, Alex Huang <
> Alex.Huang@citrix.com>
> *Subject: *Re: Control event publishing in multi region setups
>
>
>
> I'm afraid it is not possible because the events are set in the method
> level and one of my colleagues implemented to enable/disable events, but
> this is working as globally.
>
>
>
> On Tue, May 6, 2014 at 12:44 PM, Alena Prokharchyk <
> Alena.Prokharchyk@citrix.com> wrote:
>
>  Kishan, any updates from Murali? All we need to know is – if controlling
> events possible when command is fired through CS client APIs today.
>
>
>
> Thank you!
>
> Alena.
>
>
>
> *From: *Kishan Kavala <Kishan.Kavala@citrix.com>
> *Date: *Tuesday, May 6, 2014 at 7:22 AM
> *To: *Alena Prokharchyk <alena.prokharchyk@citrix.com>
> *Cc: *Alex Ough <alex.ough@sungardas.com>, Alex Huang <
> Alex.Huang@citrix.com>
>
>
> *Subject: *Re: Control event publishing in multi region setups
>
>
>
> Alena,
>
>  Events are published using the event framework introduced by Murali. It
> can contain additional details to indicate whether an event should be
> propagated to other regions.
>
>  In the implementation I added using API sync, there was a flag in the API
> params to indicate whether to propagate event or not. I reverted this code
> later when we moved to use event framework.
>
>   I'll check with Murali for more details regarding adding custom details
> / extending event details.
>
>
> On 06-May-2014, at 4:52 am, "Alena Prokharchyk" <
> Alena.Prokharchyk@citrix.com> wrote:
>
>  Alex, I understand that. But if Kishan implemented the way of extending
> the events with the details that can be later on read by events recipient,
> then you should be able to use the API.
>
>
>
> If there is no such support, the I agree that the way you do it now, is
> the only one way to achieve the desired functionality.
>
>
>
> -Alena.
>
>
>
> *From: *Alex Ough <alex.ough@sungardas.com>
> *Date: *Monday, May 5, 2014 at 4:08 PM
> *To: *Alex Huang <Alex.Huang@citrix.com>
> *Cc: *Alena Prokharchyk <alena.prokharchyk@citrix.com>, Kishan Kavala <
> Kishan.Kavala@citrix.com>
> *Subject: *Re: Control event publishing in multi region setups
>
>
>
> That's exactly why I need methods that do NOT generate events when the
> create/update/delete is just for local resources.
>
>
>
> On Mon, May 5, 2014 at 7:06 PM, Alex Huang <Alex.Huang@citrix.com> wrote:
>
>  That’s actually what I said.  Let me clarify.  When Kishan added the
> region feature, we discussed the problem of infinite circular propagation
> because each management server that adds an account will attempt to
> propagate it to all the regions by adding it to that region and so on.  The
> API needs provide a way for that propagation to be terminated.  That
> doesn’t mean we don’t publish the event in the region where the account is
> propagated to.  We still should publish the event because that region did
> add a new account but the event needs to contain enough details for anyone
> listening to the event to determine that they should not propagate the
> account creation.
>
>
>
> --Alex
>
>
>
> *From:* Alena Prokharchyk
> *Sent:* Monday, May 5, 2014 2:39 PM
> *To:* Kishan Kavala; Alex Ough
> *Cc:* Alex Huang
> *Subject:* Control event publishing in multi region setups
>
>
>
> Kishan,
>
>
>
> Have a question to you. Alex Huang mentioned to me that you were planning
> to add support for controlling event publishing in multi regions setup. So
> you can control whether you want to publish the event in a particular
> region when create/update/delete account/domain API call is made. Can you
> please tell us if you’ve implemented it? And what parameters need to be
> passed to the API call to achieve that.
>
>
>
> Alex (Ought), if Kishan didn’t implement this, then I agree with the way
> you’ve added new methods to Account/DomainManagers to do the object update
> w/o the event generation. Lets wait for Kishan’s reply. By now, you can go
> ahead and fix 1) and 2) in https://reviews.apache.org/r/20099/ which is
> not related to event generation.
>
>
>
> Thank you!
>
> -Alena.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message