osdir.com


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [BEAM-5442] Store duplicate unknown (runner) options in a list argument


+1 to Option 3

I'd rather have each SDK have a single point of well defined complexity to do something general, than have to make tiny but simple changes. Less toil and maintenance in the long run per SDK.

Similarly I don't have time to make it happen right now.

On Tue, Nov 13, 2018, 9:22 AM Thomas Weise <thw@xxxxxxxxxx> wrote:
Discovering options from the job server would be the only way to perform full validation (and provide upfront help to the user).

The runner cannot perform full validation, since it is not aware of the user and SDK options (that it has to forward to the SDK worker).

Special runner options flag to forward unknown options also wouldn't fully solve the problem (besides being subject to change in the future). Let's say runner understands --fancy-int-option and the user repeats that option multiple times. Not knowing the type of option, the SDK will pass it as a list and the runner will fail. 

Replicating SDK options is a workaround for known runners but it really goes against the idea of portability (making assumptions outside the API contract). We already have runners implemented outside of Beam and hope for the ecosystem to grow. What we do for options should also work for those runners.

I'm with Luke here that options discovery provides the best user experience and can address the other issues. Even the scenario of multiple intermediate runners could be addressed by forwarding the unparsed options with the discovery call. I don't see SDK implementation complexity as a significant drawback so far.

Thomas


On Mon, Nov 12, 2018 at 2:30 PM Lukasz Cwik <lcwik@xxxxxxxxxx> wrote:

On Mon, Nov 12, 2018 at 9:38 AM Maximilian Michels <mxm@xxxxxxxxxx> wrote:
Thank you Robert and Lukasz for your points.

> Note that I believe that we will want to have multiple URLs to support cross language pipelines since we will want to be able to ask other SDK languages/versions for their "list" of supported PipelineOptions.

Why is that? The Runner itself is the source of truth for its options.

Because other languages (or even different versions of the same language) may have their own options. For example, the Go SDK talks to a Java service which applies a SQL transform and returns the expanded form (this may require knowledge of some options like credentials for the filesystem, ...) and then talks to a Python service that performs another transform expansion. Finally the pipeline containing Go, Java and Python transforms is submitted to a runner and it performs its own internal replacements/expansions related to executing the pipeline.
 
Everything else is SDK-related and should be validated there.

I imagined the process to go like this:

   a) Parse options to find JobServer URL
   a) Retrieve options from JobServer
   c) Parse all options
   ...continue as always...

An option is just represented by a name and a type. There is nothing
more to it, at least as of now. So it should be possible to parse them
in the SDK without much further work.

Nevertheless, I agree with your sentiment, Robert. The "runner_option"
flag would prevent additional complexity. I still don't prefer it
because it's not nice from an end user perspective. If we were to
implement it, I would definitely go for the "option promotion" which you
mentioned.

I hadn't thought about delegating runners, although the PortableRunner
is basically a delegating Runner. If that was an important feature, I
suppose the "runner_option" would be the preferred way.

All in all, since there doesn't seem to be an excitement to implement
JobServer option retrieval and we will need the help of all SDK
developers, "runner_option" seems to be the more likely path.

I would say its a lack of time for people to improve this story over others but it can be revisited at some point in the future and I agree that using --runner_option as an interim provides value.
 

-Max

On 08.11.18 21:50, Lukasz Cwik wrote:
> The purpose of the spec would be to provide the names, type and
> descriptions of the options. We don't need anything beyond the JSON
> types (string, number, bool, object, list) because the only ambiguity we
> get is how do we parse command line string into the JSON type (and that
> ambiguity is actually only between string and non-string since all the
> other JSON types are unambiguous).
>
> Also, I believe the flow would be
> 1) Parse options
>    a) Find the URL from args specified and/or additional methods on
> PipelineOptions that exposes a programmatic way to set the URL during
> parsing.
>    b) Query URL for option specs
>    c) Parse the remainder of the options
> 2) Construct pipeline
> 3) Choose runner
> 4) Submit job to runner
>
> Note that I believe that we will want to have multiple URLs to support
> cross language pipelines since we will want to be able to ask other SDK
> languages/versions for their "list" of supported PipelineOptions.
>
> On Thu, Nov 8, 2018 at 11:51 AM Robert Bradshaw <robertwb@xxxxxxxxxx
> <mailto:robertwb@xxxxxxxxxx>> wrote:
>
>     There's two questions here:
>
>     (A) What do we do in the short term?
>
>     I think adding every runner option to every SDK is not sustainable
>     (n*m work, assuming every SDK knows about every runner), and having a
>     patchwork of options that were added as one-offs to SDKs is not
>     desirable either. Furthermore, it seems difficult to parse unknown
>     options as if they were valid options, so my preference here would be
>     to just use a special runner_option flag. (One could also pass a set
>     of unparsed/unvalidated runner options to the runner, even if they're
>     not distinguished for the user, and runners (or any intermediates)
>     could run a "promote" operation that promotes any of these unknowns
>     that they recognize to real options before further processing. The
>     parsing would be done as repeated-string, and not be intermingled with
>     the actually validated options. This is essential a variant of
>     option 1.)
>
>     (B) What do do in the long term? While the JobServer approach sounds
>     nice, I think it introduces a lot of complexity (we have too much of
>     that already) and still doesn't completely solve the problem. In
>     particular, it changes the flow from
>
>     1. Parse options
>     2. Construct pipeline
>     3. Choose runner
>     4. Submit job to runner
>
>     to
>
>     1. Parse options
>     2. Construct pipeline
>     3. Choose runner
>     4a. Query runner for option specs
>     4b. Re-parse options
>     4c. Submit job to runner
>
>     In particular, doing 4b in the SDK rather than just let the runner
>     itself do the validation as part of (4) doesn't save much and forces
>     us to come up with a (probably incomplete) spec as to how to define
>     options, their types, and their validations. It also means that a
>     delegating runner must choose and interact with its downstream
>     runner(s) synchronously, else we haven't actually solved the issue.
>
>     For these reasons, I don't think we even want to go with the JobServer
>     approach in the long term, which has bearing on (A).
>
>     - Robert
>
>
>     On Wed, Nov 7, 2018 at 8:50 PM Maximilian Michels <mxm@xxxxxxxxxx
>     <mailto:mxm@xxxxxxxxxx>> wrote:
>      >
>      > +1
>      >
>      > If the preferred approach is to eventually have the JobServer
>     serve the
>      > options, then the best intermediate solution is to replicate common
>      > options in the SDKs.
>      >
>      > If we went down the "--runner_option" path, we would end up with
>      > multiple ways of specifying the same options. We would eventually
>     have
>      > to deprecate "runner options" once we have the JobServer
>     approach. I'd
>      > like to avoid that.
>      >
>      > For the upcoming release we can revert the changes again and add the
>      > most common missing options to the SDKs. Then hopefully we should
>     have
>      > fetching implemented for the release after.
>      >
>      > Do you think that is feasible?
>      >
>      > Thanks,
>      > Max
>      >
>      > On 30.10.18 23:00, Lukasz Cwik wrote:
>      > > I still like #3 the most, just can't devote the time to get it
>     done.
>      > >
>      > > Instead of going with a fully implemented #3, we could hardcode
>     the a
>      > > subset of options and types within each SDK until the job server is
>      > > ready to provide this information and then migrate to the
>     "full" list.
>      > > This would be an easy path for SDKs to take on. They could
>     "know" of a
>      > > few well known options, and if they want to support all
>     options, they
>      > > implement the integration with the job server.
>      > >
>      > > On Fri, Oct 26, 2018 at 9:19 AM Maximilian Michels
>     <mxm@xxxxxxxxxx <mailto:mxm@xxxxxxxxxx>
>      > > <mailto:mxm@xxxxxxxxxx <mailto:mxm@xxxxxxxxxx>>> wrote:
>      > >
>      > >      > I would prefer we don't introduce a (quirky) way of passing
>      > >     unknown options that forces users to type JSON into the
>     command line
>      > >     (or similar acrobatics)
>      > >     Same here, the JSON approach seems technically nice but too
>     bulky
>      > >     for users.
>      > >
>      > >      > To someone wanting to run a pipeline, all options are
>     equally
>      > >     important, whether they are application specific, SDK
>     specific or
>      > >     runner specific.
>      > >
>      > >     I'm also reluctant to force users to use `--runner_option=`
>     because the
>      > >     division into "Runner" options and other options seems
>     rather arbitrary
>      > >     to users. Most built-in options are also Runner-related.
>      > >
>      > >      > It should be possible to *optionally* qualify/scope (to
>     cover
>      > >     cases where there is ambiguity), but otherwise I prefer the
>     format
>      > >     we currently have.
>      > >
>      > >     Yes, namespacing is a problem. What happens if the user
>     defines a
>      > >     custom
>      > >     PipelineOption which clashes with one of the builtin ones?
>     If both are
>      > >
>      > >     set, which one is actually applied?
>      > >
>      > >
>      > > Note that PipelineOptions so far has been treating name
>     equality to mean
>      > > option equality and the Java implementation has a bunch of
>     strict checks
>      > > to make sure that default values aren't used for duplicate
>     definitions,
>      > > they have the same type, etc...
>      > > With 1), you fail the job if the runner can't understand your
>     option
>      > > because its not represented the same way. User then needs to fix-up
>      > > their declaration of the option name.
>      > > With 2), there are no name conflicts, the SDK will need to
>     validate that
>      > > the option isn't set in both formats and error out if it is before
>      > > pipeline submission time.
>      > > With 3), you can prefetch all the options and error out to the user
>      > > during argument parsing time.
>      > >
>      > >
>      > >
>      > >     Here is a summary of the possible paths going forward:
>      > >
>      > >
>      > >     1) Validate PipelineOptions at Runner side
>      > >     ==========================================
>      > >
>      > >     The main issue raised here was that we want to move away
>     from parsing
>      > >     arguments which look like options without validating them.
>     An easy fix
>      > >     would be to actually validate them on the Runner side. This
>     could be
>      > >     done by changing the deserialization code of
>     PipelineOptions which so
>      > >     far ignores unknown JSON options.
>      > >
>      > >     See: PipelineOptionsTranslation.fromProto(Struct protoOptions)
>      > >
>      > >     Actually, this wouldn't work for user-defined
>     PipelineOptions because
>      > >     they might not be known to the Runner (if they are defined
>     in Python).
>      > >
>      > >
>      > >     2) Introduce a Runner-Option Flag
>      > >     =================================
>      > >
>      > >     In this approach we would try to add as many pipeline
>     options for a
>      > >     Runner to the SDK, but allow additional Runner options to
>     be passed
>      > >     using the `--runner-option=key=val` flag. The Runner, like
>     in 1), would
>      > >     have to ensure validation. I think this has been the most
>     favored
>      > >     way so
>      > >     far. Going forward, that means that `--parallelism=4` and
>      > >     `--runner-option=parallelism=4` will have the same effect
>     for the Flink
>      > >     Runner.
>      > >
>      > >
>      > >     3) Implement Fetching of Options from JobServer
>      > >     ===============================================
>      > >
>      > >     The options are retrieved from the JobServer before
>     submitting the
>      > >     pipeline. I think this would be ideal but, as mentioned
>     before, it
>      > >     increases the complexity for implementing new SDKs and
>     might overall
>      > >     just not be worth the effort.
>      > >
>      > >
>      > >     What do you think? I'd implement 2) for the next release,
>     unless there
>      > >     are advocates for a different approach.
>      > >
>      > >     Cheers,
>      > >     Max
>