openstack team mailing list archive

[Mark Nottingham <mnot@xxxxxxxx>]

On 12/06/2012, at 6:24 PM, Gabriel Hurley wrote:

> Mark,
> 
> Apparently you must have missed my lightning talk at the Essex summit... ;-) (http://gabrielhurley.github.com/slides/openstack/apis_like_orms/index.html)

Until I clone myself, I think that's going to happen pretty regularly; missing interesting sessions at the summit seems to happen too much :(

> Filtering, pagination, and many other API features are *critical* for a rich dashboard experience. If you want to talk specifics, the entire Horizon team would be happy to have a long chat with you.

Love to. Like I said, I'm still doing my research.

> That said, we have also considered the case you propose where you effectively "request everything and handle it on the client-side"... however, I see that as a tremendously lazy solution.

That's not the word I'd choose, but OK. What's concerning me here is that if someone has a new requirement for how the data is presented in Horizon (or any other control panel / app taking this approach), it's going to need to be reflected in the APIs, which is going to make them balloon out pretty quickly.

It'd be interesting to get a sense of how other API consumers are planning to use these features (or not); like I said, I didn't see it in the log data.

> On the service-provider end you have access to powerful database methods that can do these operations in fractions of the time the client-side can (especially with good indexes, etc.).

Oh, I don't know. CPU on the server side is often scarce, especially as you pile on features. Lots of clients means lots of power on the client side, in aggregate, and it also puts the responsibility for scaling expensive things in the hands of the party who's asking for them.

> And if you've ever worked in mobile applications you'll know that minimizing data across the wire is crucial.

Absolutely. But, I'd question if these features are actually going to help. Many mobile clients are going to want to show all of a user's VMs in a list, for example, and paginating them won't help unless the sort order is *exactly* aligned with how the client wants to present them (and from what I saw, the proposed API didn't offer sorting at all).

> The only argument I've heard in favor of that is basically "it's easier for us not to add API features".

Well, I'd put it slightly differently. The question is how much "footprint" our API should have; if we try to make everybody happy, it'll be absolutely huge, and have corresponding issues -- everything from security to documentation to server-side load, and so forth. 

However, this being HTTP, somebody can (and will!) slap an intermediary or client library in front of it to do interesting things, even if we don't provide them "out of the box."

So I think the question really is where we, as a community, want to concentrate our efforts.

> To speak on the specific feature of pagination, the problem of 'corruption' by simultaneous writers is no excuse for not implementing it. You think Google, Facebook, Flickr, etc. etc. etc. don't have this problem? If you consume their feeds you'll notice you can fetch offset-based pagination with ease. You'd never expect to see a navigation element at the bottom of Google search results that said "take me to results starting with the letter m".

Strangely, I have; I wrote an RFC about it.

> None of this is a case of "someone might use it". The Horizon team has been loudly asking for these features for 8+ months now.

OK, that's good to know -- I didn't know that, but now I do. Look forward to discussing it more.

> And not just from Keystone but from all the projects. I have a list a mile long of API features we need to really deliver a compelling experience. I was just adding some items to it today, in fact.
> 
> The rest of your points I have no strong feelings on and generally agree, but when it comes to API features... I feel *very* strongly.
> 
> All the best,
> 
>    - Gabriel
> 
> 
>> -----Original Message-----
>> From: openstack-bounces+gabriel.hurley=nebula.com@xxxxxxxxxxxxxxxxxxx
>> [mailto:openstack-
>> bounces+gabriel.hurley=nebula.com@xxxxxxxxxxxxxxxxxxx] On Behalf Of
>> Mark Nottingham
>> Sent: Monday, June 11, 2012 10:27 PM
>> To: Joseph Heck
>> Cc: openstack@xxxxxxxxxxxxxxxxxxx (openstack@xxxxxxxxxxxxxxxxxxx)
>> Subject: Re: [Openstack] [keystone] v3 API draft (update and questions to
>> the community)
>> 
>> On 11/06/2012, at 6:58 AM, Joseph Heck wrote:
>> 
>>> First - what's the current thought of support for PATCH vs PUT in updating
>> REST resources? Are there any issues with clients being able to use a PATCH
>> verb? It's not something I'm super familiar with, so I'm looking for feedback
>> from the community here. Ideally, I'd like to support the semantics of the
>> PATCH HTTP verb, and possibly just assert no support for the PUT verb to be
>> clear about intended functionality. Is that going to throw anyone for a loop?
>> 
>> I answered a question about PATCH before; don't want to repeat myself, but
>> it should be workable. Happy to chat more about it if you have specific
>> questions.
>> 
>>> Second - filtering/searching for resources. The draft includes a section
>> labelled "Query By Name", which is probably mis-labelled, as it's intended to
>> cover the general idea of passing in query parameters to general listing
>> resource endpoints to filter the result set. The API endpoints across all the
>> resources are defined as plurals, with the idea that specificity comes later in
>> the URI (for referencing a single resource), or that we could add on these
>> query parameters to restrict/filter by resource type.
>> 
>> I'm in the middle of doing some log analysis and other research about how
>> the APIs are used at Rackspace. It's too early to share results (although I do
>> intend to, in some form, because the idea is to inform future API design), but
>> one of the things that's very noticeable is how (extremely!) little pagination
>> and filtering seem to be used in anger.
>> 
>> In fact, if you take a look at the libraries, you'll find that they often don't use
>> or even support filtering or pagination; e.g., libcloud doesn't, AFAICT.
>> 
>> So, it's worth having a think about what the use cases actually are; both
>> filtering and pagination are usually ways to save one or more of:
>>  a) client-side work
>>  b) server-side work
>>  c) bandwidth / latency
>> 
>> One interesting exercise would be to estimate the largest number of users
>> (or whatever else you'd be listing) that a reasonable deployment would put
>> in a single response, triple it, do a dummy serialisation in JSON, and then gzip
>> it, so that you can estimate the size, see how long it takes to parse on the
>> client, etc.
>> 
>>> From what I've seen (in OpenStack as well as in other APIs that have
>> nothing to do with Cloud), API designers tend to overestimate the utility of
>> pagination and especially filtering ("somebody might use it"), but users just
>> ignore them, doing all of the work on the client side, except in extreme
>> circumstances (e.g., VERY large responses / very high latency).
>> 
>> Unless you have strong use cases for them, I'd be inclined to drop them; they
>> increase implementation, QA, and documentation complexity, as well as
>> making the API harder to understand. YMMV, of course :)
>> 
>> The other issue with pagination is that a relative paged approach (like you're
>> taking) means that readers' views of the complete set of items can be
>> corrupted by simultaneous writers. While in some instances this is just an
>> annoying UI bug (missing or duplicated entries on different pages, lower
>> cache hit rates), in some circumstances it can be more serious (clients not
>> understanding the true state of the system, and making bad decisions as a
>> result).
>> 
>> Again, a source of bugs and complexity (we came up with one approach to
>> this with archived feeds in RFC5005, but it's pretty heavyweight, especially
>> for use cases like this).
>> 
>> Hope this helps,
>> 
>> P.S. the X-Subject-Token stuff is breaking HTTP; you need to either put the
>> token (or a facsimile for it) in the URL, or put Vary: Subject-Token in EVERY
>> response those resources generate. The former is preferred; this is over TLS,
>> right? Sorry I didn't see that earlier.
>> 
>> P.P.S If it's not too late, drop the X- from that header!
>> <http://tools.ietf.org/html/draft-ietf-appsawg-xdash-05>
>> 
>> 
>> --
>> Mark Nottingham   http://www.mnot.net/
>> 
>> 
>> 
>> 
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~openstack
>> More help   : https://help.launchpad.net/ListHelp
> 
> 
> 
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp

--
Mark Nottingham   http://www.mnot.net/