Re: [openstack-dev] [horizon] REST and Django
for today, so can’t actually try that out at the moment. Thanks, Travis From: Thai Q Tran tqt...@us.ibm.commailto:tqt...@us.ibm.com Reply-To: OpenStack List openstack-dev@lists.openstack.orgmailto: openstack-dev@lists.openstack.org Date: Friday, December 12, 2014 at 11:05 AM To: OpenStack List openstack-dev@lists.openstack.orgmailto: openstack-dev@lists.openstack.org Subject: Re: [openstack-dev] [horizon] REST and Django In your previous example, you are posting to a certain URL (i.e. /keystone/{ver:=x.0}/{method:=update}). client: POST /keystone/{ver:=x.0}/{method:=update} = middleware: just forward to clients[ver].getattr(method)(**kwargs) = keystone: update Correct me if I'm wrong, but it looks like you have a unique URL for each /service/version/method. I fail to see how that is different than what we have today? Is there a view for each service? each version? Let's say for argument sake that you have a single view that takes care of all URL routing. All requests pass through this view and contain a JSON that contains instruction on which API to invoke and what parameters to pass. And lets also say that you wrote some code that uses reflection to map the JSON to an action. What you end up with is a client-centric application, where all of the logic resides client-side. If there are things we want to accomplish server-side, it will be extremely hard to pull off. Things like caching, websocket, aggregation, batch actions, translation, etc What you end up with is a client with no help from the server. Obviously the other extreme is what we have today, where we do everything server-side and only using client-side for binding events. I personally prefer a more balance approach where we can leverage both the server and client. There are things that client can do well, and there are things that server can do well. Going the RPC way restrict us to just client technologies and may hamper any additional future functionalities we want to bring server-side. In other words, using REST over RPC gives us the opportunity to use server-side technologies to help solve problems should the need for it arises. I would also argue that the REST approach is NOT what we have today. What we have today is a static webpage that is generated server-side, where API is hidden from the client. What we end up with using the REST approach is a dynamic webpage generated client-side, two very different things. We have essentially striped out the rendering logic from Django templating and replaced it with Angular. -Tihomir Trifonov t.trifo...@gmail.commailto:t.trifo...@gmail.com wrote: - To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org From: Tihomir Trifonov t.trifo...@gmail.commailto:t.trifo...@gmail.com Date: 12/12/2014 04:53AM Subject: Re: [openstack-dev] [horizon] REST and Django Here's an example: Admin user Joe has an Domain open and stares at it for 15 minutes while he updates the description. Admin user Bob is asked to go ahead and enable it. He opens the record, edits it, and then saves it. Joe finished perfecting the description and saves it. Doing this action would mean that the Domain is enabled and the description gets updated. Last man in still wins if he updates the same fields, but if they update different fields then both of their changes will take affect without them stomping on each other. Whether that is good or bad may depend on the situation… That's a great example. I believe that all of the Openstack APIs support PATCH updates of arbitrary fields. This way - the frontend(AngularJS) can detect which fields are being modified, and to submit only these fields for update. If we however use a form with POST, although we will load the object before updating it, the middleware cannot find which fields are actually modified, and will update them all, which is more likely what PUT should do. Thus having full control in the frontend part, we can submit only changed fields. If however a service API doesn't support PATCH, it is actually a problem in the API and not in the client... The service API documentation almost always lags (although, helped by specs now) and the service team takes on the burden of exposing a programmatic way to access the API. This is tested and easily consumable via the python clients, which removes some guesswork from using the service. True. But what if the service team modifies a method signature from let's say: def add_something(self, request, field1, field2): to def add_something(self, request, field1, field2, field3): and in the middleware we have the old signature: def add_something(self, request, field1, field2): we still need to modify the middleware to add the new field. If however the middleware is transparent and just passes **kwargs, it will pass through whatever
Re: [openstack-dev] [horizon] REST and Django
need the lines: project = api.keystone.tenant_get(request, id) kwargs = _tenant_kwargs_from_DATA(request.DATA, enabled=None) I agree that if you already have all the data it may be bad to have to do another call. I do think there is room for discussing the reasoning, though. As far as I can tell, they do this so that if you are updating an entity, you have to be very specific about the fields you are changing. I actually see this as potentially a protectionary measure against data loss and sometimes a very nice to have feature. It perhaps was intended to *help* guard against race conditions (no locking and no transactions with many users simultaneously accessing the data). Here's an example: Admin user Joe has a Domain open and stares at it for 15 minutes while he updates just the description. Admin user Bob is asked to go ahead and enable it. He opens the record, edits it, and then saves it. Joe finished perfecting the description and saves it. They could in effect both edit the same domain independently. Last man in still wins if he updates the same fields, but if they update different fields then both of their changes will take affect without them stomping on each other. Or maybe it is intended to encourage client users to compare their current and previous to see if they should issue a warning if the data changed between getting and updating the data. Or maybe like you said, it is just overhead API calls. From: Tihomir Trifonov t.trifo...@gmail.commailto:t.trifo...@gmail.com Reply-To: OpenStack List openstack-dev@lists.openstack.orgmailto: openstack-...@lists.openstack.or g Date: Thursday, December 11, 2014 at 7:53 AM To: OpenStack List openstack-dev@lists.openstack.orgmailto: openstack-...@lists.openstack.or g Subject: Re: [openstack-dev] [horizon] REST and Django Client just needs to know which URL to hit in order to invoke a certain API, and does not need to know the procedure name or parameters ordering. That's where the difference is. I think the client has to know the procedure name and parameters. Otherwise we have a translation factory pattern, that converts one naming convention to another. And you won't be able to call any service API if there is no code in the middleware to translate it to the service API procedure name and parameters. To avoid this - we can use a transparent proxy model - direct mapping of a client call to service API naming, which can be done if the client invokes the methods with the names in the service API, so that the middleware will just pass parameters, and will not translate. Instead of: updating user data: client: POST /user/=middleware: convert to /keystone/update/= keystone: update we may use: client: POST /keystone/{ver:=x.0}/{method:=update}= middleware: just forward to clients[ver].getattr(method)(**kwargs) = keystone: update The idea here is that if we have keystone 4.0 client, we will have to just add it to the clients [] list and nothing more is required at the middleware level. Just create the frontend code to use the new Keystone 4.0 methods. Otherwise we will have to add all new/different signatures of 4.0 against 2.0/3.0 in the middleware in order to use Keystone 4.0. There is also a great example of using a pluggable/new feature in Horizon. Do you remember the volume types support patch? The patch was pending in Gerrit for few months - first waiting the cinder support for volume types to go upstream, then waiting few more weeks for review. I am not sure, but as far as I remember, the Horizon patch even missed a release milestone and was introduced in the next release. If we have a transparent middleware - this will be no more an issue. As long as someone has written the frontend modules(which should be easy to add and customize), and they install the required version of the service API - they will not need updated Horizon to start using the feature. Maybe I am not the right person to give examples here, but how many of you had some kind of Horizon customization being locally merged/patched in your local distros/setups, until the patch is being pushed upstream? I will say it again. Nova, Keystone, Cinder, Glance etc. already have stable public APIs. Why do we want to add the translation middleware and to introduce another level of REST API? This layer will often hide new features, added to the service APIs and will delay their appearance in Horizon. That's simply not needed. I believe it is possible to just wrap the authentication in the middleware REST, but not to translate anything as RPC methods/parameters. And one more example: @rest_utils.ajax() def put(self, request, id): Update a single project. The POST data should be an application/json object containing the parameters to update: name (string), description (string), domain_id (string) and enabled
Re: [openstack-dev] [horizon] REST and Django
ts in the patch were very helpful for me to understand your concerns about the ease of customizing without requiring upstream changes. It also reminded me that Ive also previously questioned the python middleman. However, here are a couple of bullet points for Devils Advocate consideration. * Will we take on auto-discovery of API extensions in two spots (python for legacy and JS for new)? * The Horizon team will have to keep an even closer eye on every single project and be ready to react if there are changes to the API that break things. Right now in Glance, for example, they are working on some fixes to the v2 API (soon to become v2.3) that will allow them to deprecate v1 somewhat transparently to users of the client library. * The service API documentation almost always lags (although, helped by specs now) and the service team takes on the burden of exposing a programmatic way to access the API. This is tested and easily consumable via the python clients, which removes some guesswork from using the service. * This is going to be an incremental approach with legacy support requirements anyway. So, incorporating python side changes wont just go away. * Which approach would be better if we introduce a server side caching mechanism or a new source of data such as elastic search to improve performance? Would the client side code have to be changed dramatically to take advantage of those improvements or could it be done transparently on the server side if we own the exposed API? Im not sure I fully understood your example about Cinder. Was it the cinder client that held up delivery of horizon support, the cinder API or both? If the API isnt in, then it would hold up delivery of the feature in any case. There still would be timing pressures to react and build a new view that supports it. For customization, with Richards approach new views could be supported by just dropping in a new REST API decorated module with the APIs you want, including direct pass through support if desired to new APIs. Downstream customizations / Upstream changes to views seem a bit like a bit of a related, but different issue to me as long as their is an easy way to drop in new API support. Finally, regarding the client making two calls to do an update: Do we really need the lines: project = api.keystone.tenant_get(request, id) kwargs = _tenant_kwargs_from_DATA(request.DATA, enabled=None) I agree that if you already have all the data it may be bad to have to do another call. I do think there is room for discussing the reasoning, though. As far as I can tell, they do this so that if you are updating an entity, you have to be very specific about the fields you are changing. I actually see this as potentially a protectionary measure against data loss and sometimes a very nice to have feature. It perhaps was intended to *help* guard against race conditions (no locking and no transactions with many users simultaneously accessing the data). Here's an example: Admin user Joe has a Domain open and stares at it for 15 minutes while he updates just the description. Admin user Bob is asked to go ahead and enable it. He opens the record, edits it, and then saves it. Joe finished perfecting the description and saves it. They could in effect both edit the same domain independently. Last man in still wins if he updates the same fields, but if they update different fields then both of their changes will take affect without them stomping on each other. Or maybe it is intended to encourage client users to compare their current and previous to see if they should issue a warning if the data changed between getting and updating the data. Or maybe like you said, it is just overhead API calls. From: Tihomir Trifonov t.trifo...@gmail.commailto:t.trifo...@gmail.com Reply-To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-...@lists.openstack.or g Date: Thursday, December 11, 2014 at 7:53 AM To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-...@lists.openstack.or g Subject: Re: [openstack-dev] [horizon] REST and Django Client just needs to know which URL to hit in order to invoke a certain API, and does not need to know the procedure name or parameters ordering. That's where the difference is. I think the client has to know the procedure name and parameters. Otherwise we have a translation factory pattern, that converts one naming convention to another. And you won't be able to call any service API if there is no code in the middleware to translate it to the service API procedure name and parameters. To avoid this - we can use a transparent proxy model - direct mapping of a client call to service API naming, which can be done if the client invokes the methods with the names in the service API, so that the middleware will just pass parameters, and will not translate. Instead of: updating user data: client: POST /user/ = middleware: convert to /keystone/update/ = keystone: update we may use: client: P
Re: [openstack-dev] [horizon] REST and Django
Tihomir, Today I added one glance call based on Richard’s decorator pattern[1] and started to play with incorporating some of your ideas. Please note, I only had limited time today. That is passing the kwargs through to the glance client. This was an interesting first choice, because it immediately highlighted a concrete example of the horizon glance wrapper post-processing still being useful (rather than be a direct pass-through with no wrapper). See below. If you have some some concrete code examples of your ideas it would be helpful. [1] https://review.openstack.org/#/c/141273/2/openstack_dashboard/api/rest/glance.py With the patch, basically, you can call the following and all of the GET parameters get passed directly through to the horizon glance client and you get results back as expected. http://localhost:8002/api/glance/images/?sort_dir=descsort_key=created_atpaginate=Truemarker=bb2cfb1c-2234-4f54-aec5-b4916fe2d747 If you pass in an incorrect sort_key, the glance client returns the following error message which propagates back to the REST caller as an error with the message: sort_key must be one of the following: name, status, container_format, disk_format, size, id, created_at, updated_at. This is done by passing **request.GET.dict() through. Please note, that if you try this (with POSTMAN, for example), you need to set the header of X-Requested-With = XMLHttpRequest So, what issues did it immediately call out with directly invoking the client? The python-glanceclient internally handles pagination by returning a generator. Each iteration on the generator will handle making a request for the next page of data. If you were to just do something like return list(image_generator) to serialize it back out to the caller, it would actually end up making a call back to the server X times to fetch all pages before serializing back (thereby not really paginating). The horizon glance client wrapper today handles this by using islice intelligently along with honoring the API_RESULT_LIMIT setting in Horizon. So, this gives a direct example of where the wrapper does something that a direct passthrough to the client would not allow. That said, I can see a few ways that we could use the same REST decorator code and provide direct access to the API. We’d simply provide a class where the url_regex maps to the desired path and gives direct passthrough. Maybe that kind of passthrough could always be provided for ease of customization / extensibility and additional methods with wrappers provided when necessary. I need to leave for today, so can’t actually try that out at the moment. Thanks, Travis From: Thai Q Tran tqt...@us.ibm.commailto:tqt...@us.ibm.com Reply-To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Date: Friday, December 12, 2014 at 11:05 AM To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Subject: Re: [openstack-dev] [horizon] REST and Django In your previous example, you are posting to a certain URL (i.e. /keystone/{ver:=x.0}/{method:=update}). client: POST /keystone/{ver:=x.0}/{method:=update} = middleware: just forward to clients[ver].getattr(method)(**kwargs) = keystone: update Correct me if I'm wrong, but it looks like you have a unique URL for each /service/version/method. I fail to see how that is different than what we have today? Is there a view for each service? each version? Let's say for argument sake that you have a single view that takes care of all URL routing. All requests pass through this view and contain a JSON that contains instruction on which API to invoke and what parameters to pass. And lets also say that you wrote some code that uses reflection to map the JSON to an action. What you end up with is a client-centric application, where all of the logic resides client-side. If there are things we want to accomplish server-side, it will be extremely hard to pull off. Things like caching, websocket, aggregation, batch actions, translation, etc What you end up with is a client with no help from the server. Obviously the other extreme is what we have today, where we do everything server-side and only using client-side for binding events. I personally prefer a more balance approach where we can leverage both the server and client. There are things that client can do well, and there are things that server can do well. Going the RPC way restrict us to just client technologies and may hamper any additional future functionalities we want to bring server-side. In other words, using REST over RPC gives us the opportunity to use server-side technologies to help solve problems should the need for it arises. I would also argue that the REST approach is NOT what we have today. What we have today is a static webpage that is generated server-side, where API is hidden from the client. What we end up with using the REST approach is a dynamic
Re: [openstack-dev] [horizon] REST and Django
deletions. -Tihomir Trifonov t.trifo...@gmail.com wrote: - To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org From: Tihomir Trifonov t.trifo...@gmail.com Date: 12/10/2014 03:04AM Subject: Re: [openstack-dev] [horizon] REST and Django Richard, thanks for the reply, I agree that the given example is not a real REST. But we already have the REST API - that's Keystone, Nova, Cinder, Glance, Neutron etc, APIs. So what we plan to do here? To add a new REST layer to communicate with other REST API? Do we really need Frontend-REST-REST architecture ? My opinion is that we don't need another REST layer as we currently are trying to go away from the Django layer, which is the same - another processing layer. Although we call it REST proxy or whatever - it doesn't need to be a real REST, but just an aggregation proxy that combines and forwards some requests with adding minimal processing overhead. What makes sense for me is to keep the authentication in this layer as it is now - push a cookie to the frontend, but the REST layer will extract the auth tokens from the session storage and prepare the auth context for the REST API request to OS services. This way we will not expose the tokens to the JS frontend, and will have strict control over the authentication. The frontend will just send data requests, they will be wrapped with auth context and forwarded. Regarding the existing issues with versions in the API - for me the existing approach is wrong. All these fixes were made as workarounds. What should have been done is to create abstractions for each version and to use a separate class for each version. This was partially done for the keystoneclient in api/keystone.py, but not for the forms/views, where we still have if-else for versions. What I suggest here is to have different(concrete) views/forms for each version, and to use them according the context. If the Keystone backend is v2.0 - then in the Frontend use keystone2() object, otherwise use keystone3() object. This of course needs some more coding, but is much cleaner in terms of customization and testing. For me the current hacks with 'if keystone.version == 3.0' are wrong at many levels. And this can be solved now. *The problem till now was that we had one frontend that had to be backed by different versions of backend components*. *Now we can have different frontends that map to specific backend*. That's how I understand the power of Angular with it's views and directives. That's where I see the real benefit of using full-featured frontend. Also imagine how easy will be then to deprecate a component version, compared to what we need to do now for the same. Otherwise we just rewrite the current Django middleware with another DjangoRest middleware and don't change anything, we don't fix the problems. We just move them to another place. I still think that in Paris we talked about a new generation of the Dashboard, a different approach on building the frontend for OpenStack. What I've heard there from users/operators of Horizon was that it was extremely hard to add customizations and new features to the Dashboard, as all these needed to go through upstream changes and to wait until next release cycle to get them. Do we still want to address these concerns and how? Please, correct me if I got things wrong. On Wed, Dec 10, 2014 at 11:56 AM, Richard Jones r1chardj0...@gmail.com wrote: Sorry I didn't respond to this earlier today, I had intended to. What you're describing isn't REST, and the principles of REST are what have been guiding the design of the new API so far. I see a lot of value in using REST approaches, mostly around clarity of the interface. While the idea of a very thin proxy seemed like a great idea at one point, my conversations at the summit convinced me that there was value in both using the client interfaces present in the openstack_dashboard/api code base (since they abstract away many issues in the apis including across versions) and also value in us being able to clean up (for example, using project_id rather than project in the user API we've already implemented) and extend those interfaces (to allow batched operations). We want to be careful about what we expose in Horizon to the JS clients through this API. That necessitates some amount of code in Horizon. About half of the current API for keysone represents that control (the other half is docstrings :) Richard On Tue Dec 09 2014 at 9:37:47 PM Tihomir Trifonov t.trifo...@gmail.com wrote: Sorry for the late reply, just few thoughts on the matter. IMO the REST middleware should be as thin as possible. And I mean thin in terms of processing - it should not do pre/post processing of the requests, but just unpack/pack. So here is an example: instead of making AJAX calls that contain instructions: POST --json --data {action: delete, data
Re: [openstack-dev] [horizon] REST and Django
Tihomir, Your comments in the patch were the actually the clearest to me about ease of customizing without requiring upstream changes and really made me think more about your points. Here are a couple of bullet points for consideration. * Will we take on auto-discovery of API extensions in two spots (python for legacy and JS for new)? * As teams move towards deprecating / modifying APIs who will be responsible for ensuring the JS libraries stay up to date and keep tabs on every single project? Right now in Glance, for example, they are working on some fixes to v2 API (soon to become v2.3) that will allow them to deprecate v1 so that Nova can migrate from v1. Part of this includes making simultaneous improvements to the client library so that the switch can happen more transparently to client users. This testing and maintenance of the service team already takes on. * The service API documentation almost always lags (helped by specs now) and the service team takes on the burden of exposing a programmatic way to access which is tested and easily consumable via the python clients which removes some guesswork from using the service. * This is going to be an incremental approach with legacy support requirements anyway, I think. So, incorporating python side changes won’t just go away. * A tangent that needs to be considered IMO since I’m working on some elastic search things right now. Which of these would be better if we introduce a server side caching mechanism or a new source of data such as elastic search to improve performance? * Would the client just be able to handle changing whether or not it used cache with a header and in either case the server side appropriately uses the cache? (e.g. Cache-Control: no-cache) I’m not sure I fully understood your example about Cinder. Was it the cinderclient that held up delivery of that horizon support or there cinder API or both? If the API isn’t in, then it would hold up delivery of the feature in any case. If it is just about delivering new functionality, all that would be required in Richard’s approach is to drop in a new file of decorated classes / functions from his utility with the API’s you want? None of the API calls have anything to do with how your view actually replaces the upstream view. These are all just about accessing the data. Finally, I mentioned the following in the patch related to your example below about the client making two calls to do an update, but wanted to mention here to see if it is an approach that was purposeful (I don’t know the history): Do we really need the lines: project = api.keystone.tenant_get(request, id) kwargs = _tenant_kwargs_from_DATA(request.DATA, enabled=None) I agree that if you already have all the data it is really bad to have to do another call. I do think there is room for discussing the reasoning, though. As far as I can tell, they do this so that if you are updating an entity, you have to be very specific about the fields you are changing. I actually see this as potentially a protectionary measure against data loss and a sometimes very nice to have feature. It perhaps was intended to *help* guard against race conditions *sometimes*. Here's an example: Admin user Joe has an Domain open and stares at it for 15 minutes while he updates the description. Admin user Bob is asked to go ahead and enable it. He opens the record, edits it, and then saves it. Joe finished perfecting the description and saves it. Doing this action would mean that the Domain is enabled and the description gets updated. Last man in still wins if he updates the same fields, but if they update different fields then both of their changes will take affect without them stomping on each other. Whether that is good or bad may depend on the situation… From: Tihomir Trifonov t.trifo...@gmail.commailto:t.trifo...@gmail.com Reply-To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Date: Thursday, December 11, 2014 at 7:53 AM To: OpenStack List openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Subject: Re: [openstack-dev] [horizon] REST and Django Client just needs to know which URL to hit in order to invoke a certain API, and does not need to know the procedure name or parameters ordering. That's where the difference is. I think the client has to know the procedure name and parameters. Otherwise we have a translation factory pattern, that converts one naming convention to another. And you won't be able to call any service API if there is no code in the middleware to translate it to the service API procedure name and parameters. To avoid this - we can use a transparent proxy model - direct mapping of a client call to service API naming, which can be done if the client invokes the methods with the names in the service API, so that the middleware will just pass parameters
Re: [openstack-dev] [horizon] REST and Django
On Fri Dec 12 2014 at 1:06:08 PM Tripp, Travis S travis.tr...@hp.com wrote: Do we really need the lines: project = api.keystone.tenant_get(request, id) kwargs = _tenant_kwargs_from_DATA(request.DATA, enabled=None) I agree that if you already have all the data it is really bad to have to do another call. I do think there is room for discussing the reasoning, though. As far as I can tell, they do this so that if you are updating an entity, you have to be very specific about the fields you are changing. I actually see this as potentially a protectionary measure against data loss and a sometimes very nice to have feature. It perhaps was intended to *help* guard against race conditions *sometimes*. Yep, it looks like I broke this API by implementing it the way I did, and I'll alter the API so that you pass both the current object (according to the client) and the parameters to alter. Thanks everyone for the great reviewing! Richard ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [horizon] REST and Django
in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django -- On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S *travis.tr...@hp.com* travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. I would certainly like to see us take some measurements when performance issues pop up. Optimising without solid metrics is bad idea :) Richard [1] *https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py* https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2] *https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign* https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis *From: *Richard Jones *r1chardj0...@gmail.com* r1chardj0...@gmail.com * Date: *Wednesday, November 26, 2014 at 11:55 PM * To: *Travis Tripp *travis.tr...@hp.com* travis.tr...@hp.com, Thai Q Tran/Silicon Valley/IBM *tqt...@us.ibm.com* tqt...@us.ibm.com, David Lyle *dkly...@gmail.com* dkly...@gmail.com, Maxime Vidori *maxime.vid...@enovance.com* maxime.vid...@enovance.com, Wroblewski, Szymon *szymon.wroblew...@intel.com* szymon.wroblew...@intel.com, Wood, Matthew David (HP Cloud - Horizon) *matt.w...@hp.com* matt.w...@hp.com, Chen, Shaoquan *sean.ch...@hp.com* sean.ch...@hp.com, Farina, Matt (HP Cloud) *matthew.far...@hp.com* matthew.far...@hp.com, Cindy Lu/Silicon Valley/IBM *c...@us.ibm.com* c...@us.ibm.com, Justin Pomeroy/Rochester/IBM *jpom...@us.ibm.com* jpom...@us.ibm.com, Neill Cox *neill@ingenious.com.au* neill@ingenious.com.au * Subject: *Re: REST and Django I'm not sure whether this is the appropriate place to discuss this, or whether I should be posting to the list under [Horizon] but I think we need to have a clear idea of what goes in the REST API and what goes in the client (angular) code. In my mind, the thinner the REST API
Re: [openstack-dev] [horizon] REST and Django
: {name: item 2 } , { action : delete , payload: {name: item 3 } ] } The idea is that the middleware should not know the actual data. It should ideally just unpack the data: responses = [] for cmd in request.POST['batch']: responses .append( getattr(controller, cmd['action'] )(** cmd['payload'] )) return responses and the frontend(JS) will just send batches of simple commands, and will receive a list of responses for each command in the batch. The error handling will be done in the frontend(JS) as well. For the more complex example of 'put()' where we have dependent objects: project = api.keystone.tenant_get(request, id) kwargs = self._tenant_kwargs_from_DATA(request.DATA, enabled=None) api.keystone.tenant_update(request, project, **kwargs) In practice the project data should be already present in the frontend(assuming that we already loaded it to render the project form/view), so POST --json --data { batch : [ { action : tenant_update , payload: {project: js_project_object.id, name: some name, prop1: some prop, prop2: other prop, etc.} ] } So in general we don't need to recreate the full state on each REST call, if we make the Frontent full-featured application. This way - the frontend will construct the object, will hold the cached value, and will just send the needed requests as single ones or in batches, will receive the response from the API backend, and will render the results. The whole processing logic will be held in the Frontend(JS), while the middleware will just act as proxy(un/packer). This way we will maintain just the logic in the frontend, and will not need to duplicate some logic in the middleware. On Tue, Dec 2, 2014 at 4:45 PM, Adam Young ayo...@redhat.com wrote: On 12/02/2014 12:39 AM, Richard Jones wrote: On Mon Dec 01 2014 at 4:18:42 PM Thai Q Tran tqt...@us.ibm.com wrote: I agree that keeping the API layer thin would be ideal. I should add that having discrete API calls would allow dynamic population of table. However, I will make a case where it *might* be necessary to add additional APIs. Consider that you want to delete 3 items in a given table. If you do this on the client side, you would need to perform: n * (1 API request + 1 AJAX request) If you have some logic on the server side that batch delete actions: n * (1 API request) + 1 AJAX request Consider the following: n = 1, client = 2 trips, server = 2 trips n = 3, client = 6 trips, server = 4 trips n = 10, client = 20 trips, server = 11 trips n = 100, client = 200 trips, server 101 trips As you can see, this does not scale very well something to consider... This is not something Horizon can fix. Horizon can make matters worse, but cannot make things better. If you want to delete 3 users, Horizon still needs to make 3 distinct calls to Keystone. To fix this, we need either batch calls or a standard way to do multiples of the same operation. The unified API effort it the right place to drive this. Yep, though in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django -- On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S *travis.tr...@hp.com* travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some
Re: [openstack-dev] [horizon] REST and Django
Yep, though in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com To: "Tripp, Travis S" travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (Ive listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the APIs fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. I would certainly like to see us take some measurements when performance issues pop up. Optimising without solid metrics is bad idea :) Richard [1]https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2]https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis
Re: [openstack-dev] [horizon] REST and Django
Sorry for the late reply, just few thoughts on the matter. IMO the REST middleware should be as thin as possible. And I mean thin in terms of processing - it should not do pre/post processing of the requests, but just unpack/pack. So here is an example: instead of making AJAX calls that contain instructions: POST --json --data {action: delete, data: [ {name: item1}, {name: item2}, {name: item3 ]} I think a better approach is just to pack/unpack batch commands, and leave execution to the frontend/backend and not middleware: POST --json --data { batch : [ { action : delete , payload: {name: item1} , { action : delete , payload: {name: item 2 } , { action : delete , payload: {name: item 3 } ] } The idea is that the middleware should not know the actual data. It should ideally just unpack the data: responses = [] for cmd in request.POST['batch']: responses .append( getattr(controller, cmd['action'] )(** cmd['payload'] )) return responses and the frontend(JS) will just send batches of simple commands, and will receive a list of responses for each command in the batch. The error handling will be done in the frontend(JS) as well. For the more complex example of 'put()' where we have dependent objects: project = api.keystone.tenant_get(request, id) kwargs = self._tenant_kwargs_from_DATA(request.DATA, enabled=None) api.keystone.tenant_update(request, project, **kwargs) In practice the project data should be already present in the frontend(assuming that we already loaded it to render the project form/view), so POST --json --data { batch : [ { action : tenant_update , payload: {project: js_project_object.id, name: some name, prop1: some prop, prop2: other prop, etc.} ] } So in general we don't need to recreate the full state on each REST call, if we make the Frontent full-featured application. This way - the frontend will construct the object, will hold the cached value, and will just send the needed requests as single ones or in batches, will receive the response from the API backend, and will render the results. The whole processing logic will be held in the Frontend(JS), while the middleware will just act as proxy(un/packer). This way we will maintain just the logic in the frontend, and will not need to duplicate some logic in the middleware. On Tue, Dec 2, 2014 at 4:45 PM, Adam Young ayo...@redhat.com wrote: On 12/02/2014 12:39 AM, Richard Jones wrote: On Mon Dec 01 2014 at 4:18:42 PM Thai Q Tran tqt...@us.ibm.com wrote: I agree that keeping the API layer thin would be ideal. I should add that having discrete API calls would allow dynamic population of table. However, I will make a case where it *might* be necessary to add additional APIs. Consider that you want to delete 3 items in a given table. If you do this on the client side, you would need to perform: n * (1 API request + 1 AJAX request) If you have some logic on the server side that batch delete actions: n * (1 API request) + 1 AJAX request Consider the following: n = 1, client = 2 trips, server = 2 trips n = 3, client = 6 trips, server = 4 trips n = 10, client = 20 trips, server = 11 trips n = 100, client = 200 trips, server 101 trips As you can see, this does not scale very well something to consider... This is not something Horizon can fix. Horizon can make matters worse, but cannot make things better. If you want to delete 3 users, Horizon still needs to make 3 distinct calls to Keystone. To fix this, we need either batch calls or a standard way to do multiples of the same operation. The unified API effort it the right place to drive this. Yep, though in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard [image: Inactive hide details for Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr]Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django -- On Fri Nov 28 2014 at 5:58:00 AM Tripp
Re: [openstack-dev] [horizon] REST and Django
On 12/02/2014 12:39 AM, Richard Jones wrote: On Mon Dec 01 2014 at 4:18:42 PM Thai Q Tran tqt...@us.ibm.com mailto:tqt...@us.ibm.com wrote: I agree that keeping the API layer thin would be ideal. I should add that having discrete API calls would allow dynamic population of table. However, I will make a case where it */might/* be necessary to add additional APIs. Consider that you want to delete 3 items in a given table. If you do this on the client side, you would need to perform: n * (1 API request + 1 AJAX request) If you have some logic on the server side that batch delete actions: n * (1 API request) + 1 AJAX request Consider the following: n = 1, client = 2 trips, server = 2 trips n = 3, client = 6 trips, server = 4 trips n = 10, client = 20 trips, server = 11 trips n = 100, client = 200 trips, server 101 trips As you can see, this does not scale very well something to consider... This is not something Horizon can fix. Horizon can make matters worse, but cannot make things better. If you want to delete 3 users, Horizon still needs to make 3 distinct calls to Keystone. To fix this, we need either batch calls or a standard way to do multiples of the same operation. The unified API effort it the right place to drive this. Yep, though in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard Inactive hide details for Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.trRichard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com mailto:travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com mailto:r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com mailto:travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org mailto:openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S _travis.tripp@hp.com_ mailto:travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some
Re: [openstack-dev] [horizon] REST and Django
On Mon Dec 01 2014 at 4:18:42 PM Thai Q Tran tqt...@us.ibm.com wrote: I agree that keeping the API layer thin would be ideal. I should add that having discrete API calls would allow dynamic population of table. However, I will make a case where it *might* be necessary to add additional APIs. Consider that you want to delete 3 items in a given table. If you do this on the client side, you would need to perform: n * (1 API request + 1 AJAX request) If you have some logic on the server side that batch delete actions: n * (1 API request) + 1 AJAX request Consider the following: n = 1, client = 2 trips, server = 2 trips n = 3, client = 6 trips, server = 4 trips n = 10, client = 20 trips, server = 11 trips n = 100, client = 200 trips, server 101 trips As you can see, this does not scale very well something to consider... Yep, though in the above cases the client is still going to be hanging, waiting for those server-backend calls, with no feedback until it's all done. I would hope that the client-server call overhead is minimal, but I guess that's probably wishful thinking when in the land of random Internet users hitting some provider's Horizon :) So yeah, having mulled it over myself I agree that it's useful to have batch operations implemented in the POST handler, the most common operation being DELETE. Maybe one day we could transition to a batch call with user feedback using a websocket connection. Richard [image: Inactive hide details for Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr]Richard Jones ---11/27/2014 05:38:53 PM---On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: From: Richard Jones r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject: Re: [openstack-dev] [horizon] REST and Django -- On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S *travis.tr...@hp.com* travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. I would certainly like to see us take some measurements when performance issues pop up. Optimising without solid metrics is bad idea :) Richard [1] *https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py* https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2] *https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign* https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis *From: *Richard Jones *r1chardj0...@gmail.com* r1chardj0...@gmail.com * Date: *Wednesday, November 26, 2014 at 11:55 PM * To: *Travis Tripp *travis.tr...@hp.com* travis.tr...@hp.com, Thai Q Tran/Silicon Valley/IBM *tqt...@us.ibm.com* tqt...@us.ibm.com
Re: [openstack-dev] [horizon] REST and Django
I agree that keeping the API layer thin would be ideal. I should add that having discrete API calls would allow dynamic population of table. However, I will make a case where it might be necessary to add additional APIs. Consider that you want to delete 3 items in a given table. If you do this on the client side, you would need to perform: n * (1 API request + 1 AJAX request) If you have some logic on the server side that batch delete actions: n * (1 API request) + 1 AJAX request Consider the following: n = 1, client = 2 trips, server = 2 trips n = 3, client = 6 trips, server = 4 trips n = 10, client = 20 trips, server = 11 trips n = 100, client = 200 trips, server 101 trips As you can see, this does not scale very well something to consider... From: Richard Jones r1chardj0...@gmail.com To: Tripp, Travis S travis.tr...@hp.com, OpenStack List openstack-dev@lists.openstack.org Date: 11/27/2014 05:38 PM Subject:Re: [openstack-dev] [horizon] REST and Django On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. I would certainly like to see us take some measurements when performance issues pop up. Optimising without solid metrics is bad idea :) Richard [1] https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2] https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis From: Richard Jones r1chardj0...@gmail.com Date: Wednesday, November 26, 2014 at 11:55 PM To: Travis Tripp travis.tr...@hp.com, Thai Q Tran/Silicon Valley/IBM tqt...@us.ibm.com, David Lyle dkly...@gmail.com, Maxime Vidori maxime.vid...@enovance.com, Wroblewski, Szymon szymon.wroblew...@intel.com, Wood, Matthew David (HP Cloud - Horizon) matt.w...@hp.com, Chen, Shaoquan sean.ch...@hp.com, Farina, Matt (HP Cloud) matthew.far...@hp.com, Cindy Lu/Silicon Valley/IBM c...@us.ibm.com, Justin Pomeroy/Rochester/IBM jpom...@us.ibm.com, Neill Cox neill@ingenious.com.au Subject: Re: REST and Django I'm not sure whether this is the appropriate place to discuss this, or whether I should be posting to the list under [Horizon] but I think we need to have a clear idea of what goes in the REST API and what goes in the client (angular) code. In my mind, the thinner the REST API the better. Indeed if we can get away with proxying requests through without touching any *client code, that would be great. Coding additional logic into the REST API means that a developer would need to look in two places, instead of one, to determine what was happening for a particular call. If we keep it thin then the API presented to the client developer is very, very similar to the API presented by the services. Minimum surprise. Your thoughts? Richard On Wed Nov 26 2014 at 2:40:52 PM Richard Jones r1chardj0...@gmail.com wrote: Thanks
[openstack-dev] [horizon] REST and Django
Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. [1] https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2] https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis From: Richard Jones r1chardj0...@gmail.commailto:r1chardj0...@gmail.com Date: Wednesday, November 26, 2014 at 11:55 PM To: Travis Tripp travis.tr...@hp.commailto:travis.tr...@hp.com, Thai Q Tran/Silicon Valley/IBM tqt...@us.ibm.commailto:tqt...@us.ibm.com, David Lyle dkly...@gmail.commailto:dkly...@gmail.com, Maxime Vidori maxime.vid...@enovance.commailto:maxime.vid...@enovance.com, Wroblewski, Szymon szymon.wroblew...@intel.commailto:szymon.wroblew...@intel.com, Wood, Matthew David (HP Cloud - Horizon) matt.w...@hp.commailto:matt.w...@hp.com, Chen, Shaoquan sean.ch...@hp.commailto:sean.ch...@hp.com, Farina, Matt (HP Cloud) matthew.far...@hp.commailto:matthew.far...@hp.com, Cindy Lu/Silicon Valley/IBM c...@us.ibm.commailto:c...@us.ibm.com, Justin Pomeroy/Rochester/IBM jpom...@us.ibm.commailto:jpom...@us.ibm.com, Neill Cox neill@ingenious.com.aumailto:neill@ingenious.com.au Subject: Re: REST and Django I'm not sure whether this is the appropriate place to discuss this, or whether I should be posting to the list under [Horizon] but I think we need to have a clear idea of what goes in the REST API and what goes in the client (angular) code. In my mind, the thinner the REST API the better. Indeed if we can get away with proxying requests through without touching any *client code, that would be great. Coding additional logic into the REST API means that a developer would need to look in two places, instead of one, to determine what was happening for a particular call. If we keep it thin then the API presented to the client developer is very, very similar to the API presented by the services. Minimum surprise. Your thoughts? Richard On Wed Nov 26 2014 at 2:40:52 PM Richard Jones r1chardj0...@gmail.commailto:r1chardj0...@gmail.com wrote: Thanks for the great summary, Travis. I've completed the work I pledged this morning, so now the REST API change set has: - no rest framework dependency - AJAX scaffolding in openstack_dashboard.api.rest.utils - code in openstack_dashboard/api/rest/ - renamed the API from identity to keystone to be consistent - added a sample of testing, mostly for my own sanity to check things were working https://review.openstack.org/#/c/136676 Richard On Wed Nov 26 2014 at 12:18:25 PM Tripp, Travis S travis.tr...@hp.commailto:travis.tr...@hp.com wrote: Hello all, Great discussion on the REST urls today! I think that we are on track to come to a common REST API usage pattern. To provide quick summary: We all agreed that going to a straight REST pattern rather than through tables was a good idea. We discussed using direct get / post in Django views like what Max originally used[1][2] and Thai also started[3] with the identity table rework or to go with djangorestframework [5] like what Richard was prototyping with[4]. The main things we would use from Django Rest Framework were built in JSON serialization (avoid boilerplate), better exception handling, and some request wrapping. However, we all weren’t sure about the need for a full new framework just for that. At the end of the
Re: [openstack-dev] [horizon] REST and Django
On Fri Nov 28 2014 at 5:58:00 AM Tripp, Travis S travis.tr...@hp.com wrote: Hi Richard, You are right, we should put this out on the main ML, so copying thread out to there. ML: FYI that this started after some impromptu IRC discussions about a specific patch led into an impromptu google hangout discussion with all the people on the thread below. Thanks Travis! As I mentioned in the review[1], Thai and I were mainly discussing the possible performance implications of network hops from client to horizon server and whether or not any aggregation should occur server side. In other words, some views require several APIs to be queried before any data can displayed and it would eliminate some extra network requests from client to server if some of the data was first collected on the server side across service APIs. For example, the launch instance wizard will need to collect data from quite a few APIs before even the first step is displayed (I’ve listed those out in the blueprint [2]). The flip side to that (as you also pointed out) is that if we keep the API’s fine grained then the wizard will be able to optimize in one place the calls for data as it is needed. For example, the first step may only need half of the API calls. It also could lead to perceived performance increases just due to the wizard making a call for different data independently and displaying it as soon as it can. Indeed, looking at the current launch wizard code it seems like you wouldn't need to load all that data for the wizard to be displayed, since only some subset of it would be necessary to display any given panel of the wizard. I tend to lean towards your POV and starting with discrete API calls and letting the client optimize calls. If there are performance problems or other reasons then doing data aggregation on the server side could be considered at that point. I'm glad to hear it. I'm a fan of optimising when necessary, and not beforehand :) Of course if anybody is able to do some performance testing between the two approaches then that could affect the direction taken. I would certainly like to see us take some measurements when performance issues pop up. Optimising without solid metrics is bad idea :) Richard [1] https://review.openstack.org/#/c/136676/8/openstack_dashboard/api/rest/urls.py [2] https://blueprints.launchpad.net/horizon/+spec/launch-instance-redesign -Travis From: Richard Jones r1chardj0...@gmail.com Date: Wednesday, November 26, 2014 at 11:55 PM To: Travis Tripp travis.tr...@hp.com, Thai Q Tran/Silicon Valley/IBM tqt...@us.ibm.com, David Lyle dkly...@gmail.com, Maxime Vidori maxime.vid...@enovance.com, Wroblewski, Szymon szymon.wroblew...@intel.com, Wood, Matthew David (HP Cloud - Horizon) matt.w...@hp.com, Chen, Shaoquan sean.ch...@hp.com, Farina, Matt (HP Cloud) matthew.far...@hp.com, Cindy Lu/Silicon Valley/IBM c...@us.ibm.com, Justin Pomeroy/Rochester/IBM jpom...@us.ibm.com, Neill Cox neill@ingenious.com.au Subject: Re: REST and Django I'm not sure whether this is the appropriate place to discuss this, or whether I should be posting to the list under [Horizon] but I think we need to have a clear idea of what goes in the REST API and what goes in the client (angular) code. In my mind, the thinner the REST API the better. Indeed if we can get away with proxying requests through without touching any *client code, that would be great. Coding additional logic into the REST API means that a developer would need to look in two places, instead of one, to determine what was happening for a particular call. If we keep it thin then the API presented to the client developer is very, very similar to the API presented by the services. Minimum surprise. Your thoughts? Richard On Wed Nov 26 2014 at 2:40:52 PM Richard Jones r1chardj0...@gmail.com wrote: Thanks for the great summary, Travis. I've completed the work I pledged this morning, so now the REST API change set has: - no rest framework dependency - AJAX scaffolding in openstack_dashboard.api.rest.utils - code in openstack_dashboard/api/rest/ - renamed the API from identity to keystone to be consistent - added a sample of testing, mostly for my own sanity to check things were working https://review.openstack.org/#/c/136676 Richard On Wed Nov 26 2014 at 12:18:25 PM Tripp, Travis S travis.tr...@hp.com wrote: Hello all, Great discussion on the REST urls today! I think that we are on track to come to a common REST API usage pattern. To provide quick summary: We all agreed that going to a straight REST pattern rather than through tables was a good idea. We discussed using direct get / post in Django views like what Max originally used[1][2] and Thai also started[3] with the identity table rework or to go with djangorestframework [5] like what Richard was prototyping with[4]. The main things we would