+1 to clear documentation of what the various Payloads are allowed to reference. Outside of those, the various proto fields are pretty clear by the proto structure, but the payloads are an escape hatch to anything.
On Tue, Jul 7, 2020 at 2:52 PM Luke Cwik <lc...@google.com> wrote: > Kenn, all the runner based PTransform replacement/fusion/... works on the > proto version of the pipeline. Only at the final stages where we need to > reify certain objects for execution do we reuse the non-proto based objects > (well known coders, window fns, ...). The conversion back to Java > PCollection/Coder/PTransform/... objects and then back to proto again had > similar issues to what Python is experiencing. Unfortunately, Beam Java is > not well prepared to be the caller of the XLang expansion service as it > currently does not remember the components map as part of the pipeline so > that the ids can be propagated correctly. I think there would be a good > amount of code that would need to change related to pipeline construction > to fix this. Also, the alpha conversion would have been a lot simpler if we > had forced ids to be globally unique otherwise each local -> id mapping > needs to be scoped to what it is representing. > > Cham, I was looking to see what people were thinking in this space and to > share it with the wider community and I think making sure that an object -> > id context map is part of the pipeline construction makes a lot of sense as > a solution to preserve these ids. The other part of this is that we need to > document what these opaque blobs representing windows fns, do fns, ... are > allowed to reference since runners and SDKs would have the ability to > deduplicate common protos. We do have one level of redirection for > inputs/outputs on PTransforms since it was recognized early on that runners > will need to have the ability to rewire PTransforms. We also have it on > user state, timers and side inputs on ParDoPayload since runners may need > to rewrite the coder associated with them to add length prefixing when > necessary. This all ties into documenting what can be saved inside these > opaque blobs and what SDKs and runners can assume about what is legal to do > to the pipeline proto representation. > > On Thu, Jul 2, 2020 at 10:19 AM Robert Bradshaw <rober...@google.com> > wrote: > >> I just finished reading BEAM-10143 and pr/12067, and I agree this is >> exactly the issue. The inability to do alpha conversions is exactly why the >> namespace prefix was introduced for external transforms, but the fact that >> the caller uses an ephemeral context to create ids for the input >> PCollections (and their coders, windowing strategies, etc.) when calling >> external transforms, which is again distinct from that used to convert the >> whole pipeline, causes issues here. >> >> Keeping this mapping around for the lifetime of the pipeline, as is done >> in pr/12067, seems a reasonable solution. Another possibility is to create >> these ephemeral contexts with the prescribed namespace in the caller, and >> leverage the fact that we can alpha convert the PCollections, to avoid >> collisions for stuff like this. (The coders and windowing strategies would >> be duplicated in the graph, so that's less than ideal.) >> >> >> On Thu, Jul 2, 2020 at 10:03 AM Kenneth Knowles <k...@apache.org> wrote: >> >>> I had a very long chat with Brian about this the other day over >>> https://github.com/apache/beam/pull/12067 >>> >>> The most important piece of background: ** the ids are just pointers [1] >>> and every other use of them is a misuse of the structure **. Equivalently, >>> they are bound variables and the binding is in the Components. Using the >>> name for any purpose other than referring to the component is a misuse [2] >>> of the structure. >>> >>> For anyone that has not done language programming, this is the same as >>> managing bindings. It is sometimes tricky to get right, but it is also >>> solved [3]. The term to search for is "alpha conversion" and there are many >>> resources. I think talking about unique ids muddles the issue. These are >>> locally unique for exactly the same mundane reason that variables have >>> different names. >>> >>> However, it is more complicated for Beam, because payloads by design can >>> be opaque, so they cannot be alpha converted. This is true even without >>> xlang. Once a payload exists within a proto with components, it cannot be >>> separated from those components since you cannot (in general) know which >>> are referenced. This design could be revised to allow alpha conversion of >>> opaque payloads by exposing ids via a local layer of indirection. >>> >>> Suppose you have components with "id_a", "id_b", ... >>> Instead of { urn: "xyz", payload: <<opaque payload referencing bound ids >>> "id_a", "id_b", ...>> } >>> We could encode as { urn: "xyz", references: { "local_a": "id_a", >>> "local_b": "id_b", ... }, payload: <<opaque payload referencing local >>> bindings "local_a", "local_b", ...>> } >>> >>> With this change we can alpha convert opaque payloads by altering the >>> values in the "references" map. I am not advocating for such a major change >>> now, but this might help to think about things. >>> >>> I think the simple guideline of "only translate once, and keep the >>> Components around" will be enough. >>> >>> The issue as I came to understand it is as simple as this: if you pass a >>> parameter to a PTransform by way of xlang/runner API, then the resulting >>> payloads are able to have ids embedded from both the Components you sent >>> and the Components you got back. You have to keep all of them or you may >>> produce unbound variables. You always initialize a new pipeline or pipeline >>> fragment with the Components gathered so far and their associations. >>> >>> In the Python SDK this is complicated by a lot of this stuff being >>> controlled and passed in to to_runner_api which could be too late. >>> >>> In the Java SDK this is probably complicated by the separation of >>> runners-core-construction, though I don't know if that has bitten yet. >>> >>> Kenn >>> >>> [1] The reason for using this approach at all is to represent a graph in >>> proto and also for saving space by reusing things like coders and windowing >>> strategies (Dataflow's v1b3 has major size issues). I don't know that there >>> is another way to do this. >>> [2] Since Dataflow's v1b3 API design has major space problems, the v1b3 >>> message was redefined to be inside the scope of the Components bindings. >>> Now the message has no meaning outside of it. >>> [3] I use "solved" in the same sense as "breadth-first search is >>> solved". We know how to do it, there is not much to discuss, and if you try >>> to do it some other way you are probably making a mistake. >>> >>> On Thu, Jul 2, 2020 at 9:04 AM Robert Bradshaw <rober...@google.com> >>> wrote: >>> >>>> I agree this has been a major source of pain. The primary cause of >>>> issues is the conversion from Beam protos back to the SDK objects (which >>>> doesn't always have a good representation, especially for foreign-language >>>> components). In my experience, the SDK objects -> Beam Proto conversions >>>> aren't generally a problem (except sometimes for objects that were formerly >>>> converted from protos). >>>> >>>> In my opinion, the solution is to convert to Beam protos and never >>>> convert back. (Well, not until we get to the workers, but at that point we >>>> can confidently say everything we need to decode does actually belong to >>>> the ambient environment.) As mentioned, Dataflow is being fixed, and the >>>> only other runner (in Python) that doesn't consume the Beam protos directly >>>> is the old direct runner (which Pablo is working on making obsolete, and >>>> doesn't support cross language anyway). So finally fixing dataflow should >>>> be all we need to do. (Have we seen these issues on other runners?) >>>> >>>> On the Java side, I think all the optimization stuff works on its SDK >>>> representation so care needs to be done to make that conversion faithful or >>>> convert that code to act on the protos directly. >>>> >>>> As for why we went with the current approach, it's simply the fact that >>>> SDK representation -> Dataflow v1beta3 predated any of the beam protos >>>> stuff, and re-using that code seemed easier than updating the Dataflow >>>> service to accept Beam protos (or re-writing it) as we had the SDK >>>> representation for everything in hand (until cross-langauge came along that >>>> is). >>>> >>>> >>>> On Thu, Jul 2, 2020 at 3:11 AM Heejong Lee <heej...@google.com> wrote: >>>> >>>>> >>>>> >>>>> On Wed, Jul 1, 2020 at 7:18 PM Robert Burke <rob...@frantil.com> >>>>> wrote: >>>>> >>>>>> From the Go SDK side, it was built that way nearly from the start. >>>>>> Historically there was a direct SDK rep -> Dataflow rep conversion, but >>>>>> that's been replaced with a SDK rep -> Beam Proto -> Dataflow rep >>>>>> conversion. >>>>>> >>>>>> In particular, this approach had a few benefits: easier to access >>>>>> local context for pipeline validation at construction time, to permit as >>>>>> early a failure as possible, which might be easier with native language >>>>>> constructs vs beam representations of them.(Eg. DoFns not matching ParDo >>>>>> & >>>>>> Collection types, and similar) >>>>>> Protos are convenient, but impose certain structure on how the >>>>>> pipeline graph is handled. (This isn't to say an earlier conversion isn't >>>>>> possible, one can do almost anything in code, but it lets the structure >>>>>> be >>>>>> optimised for this case.) >>>>>> >>>>>> The big advantage of translating from Beam proto -> to Dataflow Rep >>>>>> is that the Dataflow Rep can get the various unique IDs that are mandated >>>>>> for the Beam proto process. >>>>>> >>>>>> However, the same can't really be said for the other way around. A >>>>>> good question is "when should the unique IDs be assigned?" >>>>>> >>>>> >>>>> This is very true and I would like to elaborate more on the source of >>>>> friction when using external transforms. As Robert mentioned, pipeline >>>>> proto refers to each component by unique IDs and the unique ID is only >>>>> assigned when we convert SDK pipeline object to pipeline proto. Before >>>>> XLang, pipeline object to pipeline proto conversion happened one time >>>>> during the job submission phase. However, after XLang transform was >>>>> introduced, it also happens when we request expansion of external >>>>> transforms to the expansion service. Unique ID generated for the expansion >>>>> request can be embedded in the returning external proto and conflicted >>>>> later with other unique IDs generated for the job submission. >>>>> >>>>> >>>>>> >>>>>> While I'm not working on adding XLang to the Go SDK directly (that >>>>>> would be our wonderful intern, Kevin), I've kind of pictured that the >>>>>> process was to provide the Expansion service with unique placeholders if >>>>>> unable to provide the right IDs, and substitute them in returned pipeline >>>>>> graph segment afterwards, once that is known. That is, we can be >>>>>> relatively >>>>>> certain that the expansion service will be self consistent, but it's the >>>>>> SDK requesting the expansion's responsibility to ensure they aren't >>>>>> colliding with the primary SDKs pipeline ids. >>>>>> >>>>> >>>>> AFAIK, we're already doing this in Java and Python SDKs. Not providing >>>>> a "placeholder" but remembering which pipeline object maps to which unique >>>>> ID used in the expanded component proto. >>>>> >>>>> >>>>>> >>>>>> Otherwise, we could probably recommend a translation protocol (if one >>>>>> doesn't exist already, it probably does) and when XLang expansions are to >>>>>> happen in the SDK -> beam proto process. So something like Pass 1, intern >>>>>> all coders and Pcollections, Pass 2 intern all DoFns and environments, >>>>>> Pass >>>>>> 3 expand Xlang, ... Etc. >>>>>> >>>>> >>>>> Not sure I understand correctly but a following transform who consumes >>>>> the output of an external transform needs some information like the output >>>>> pcollection information from the expanded external transform during the >>>>> pipeline construction phase. >>>>> >>>>> >>>>>> The other half of this is when happens when Going from Beam proto a >>>>>> -> SDK? This happens during pipeline execution, but at least in the >>>>>> Go SDK partly happens when creating the Dataflow rep. In particular, >>>>>> Coder >>>>>> reference values only have a populated ID when they've been "rehydrated" >>>>>> from the Beam proto, since the Beam Proto is the first place where such >>>>>> IDs >>>>>> are correctly assigned. >>>>>> >>>>>> Tl;dr; i think the right question to sort out is when should IDs be >>>>>> expected to be assigned and available during pipeline construction. >>>>>> >>>>>> On Wed, Jul 1, 2020, 6:34 PM Luke Cwik <lc...@google.com> wrote: >>>>>> >>>>>>> It seems like we keep running into translation issues with XLang due >>>>>>> to how it is represented in the SDK. (e.g. Brian's work on context map >>>>>>> due >>>>>>> to loss of coder ids, Heejong's work related to missing environment ids >>>>>>> on >>>>>>> windowing strategies). >>>>>>> >>>>>>> I understand that there is an effort that is Dataflow specific where >>>>>>> the conversion of the Beam proto -> Dataflow API (v1b3) will help with >>>>>>> some >>>>>>> issues but it still requires the SDK pipeline representation -> Beam >>>>>>> proto >>>>>>> to occur correctly which won't be fixed by the Dataflow specific effort. >>>>>>> >>>>>>> Why did we go with the current approach? >>>>>>> >>>>>>> What other ways could we do this? >>>>>>> >>>>>>
