[ https://issues.apache.org/jira/browse/BEAM-1198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15768387#comment-15768387 ]
ASF GitHub Bot commented on BEAM-1198: -------------------------------------- GitHub user kennknowles opened a pull request: https://github.com/apache/incubator-beam/pull/1678 [BEAM-1198, BEAM-846, BEAM-260] Refactor Dataflow translator to decouple input and output graphs more easily Be sure to do all of the following to help us incorporate your contribution quickly and easily: - [x] Make sure the PR title is formatted like: `[BEAM-<Jira issue #>] Description of pull request` - [ ] Make sure tests pass via `mvn clean verify`. (Even better, enable Travis-CI on your fork and ensure the whole test matrix passes). - [x] Replace `<Jira issue #>` in the title with the actual Jira issue number, if there is one. - [x] If this contribution is large, please file an Apache [Individual Contributor License Agreement](https://www.apache.org/licenses/icla.txt). --- This is preparatory work to make it possible for the translator to have a more loosely coupled relationship between its input and output. You can merge this pull request into a Git repository by running: $ git pull https://github.com/kennknowles/incubator-beam Dataflow-Translator Alternatively you can review and apply these changes as the patch at: https://github.com/apache/incubator-beam/pull/1678.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #1678 ---- commit 8ed4bb68660c537e4a12c1077ecfa104f9a82eaa Author: Kenneth Knowles <k...@google.com> Date: 2016-12-21T22:21:50Z Inline needless interface DataflowTranslator.TranslationContext The only implementation was DataflowTranslator.Translator. This class needs some updating and the extra layer of the interface simply obscures that work. commit 272d06d7507ad7162616dd1b613efa7c8f5f4069 Author: Kenneth Knowles <k...@google.com> Date: 2016-12-21T22:34:27Z Explicitly pass Step to mutate in Dataflow translator Previously, there was always a "current" step that was the most recent step created. This makes it cumbersome or impossible to do things like translate one primitive transform into a small subgraph of steps. Thus we added hacks like CreatePCollectionView which are not actually part of the model at all - in fact, we should be able to add the needed CollectionToSingleton steps simply by looking at the side inputs of a ParDo node. ---- > ViewFn: explicitly decouple runner materialization of side inputs from > SDK-specific mapping > ------------------------------------------------------------------------------------------- > > Key: BEAM-1198 > URL: https://issues.apache.org/jira/browse/BEAM-1198 > Project: Beam > Issue Type: New Feature > Components: beam-model-fn-api, beam-model-runner-api > Reporter: Kenneth Knowles > Assignee: Kenneth Knowles > > For side inputs, the field {{PCollectionView.fromIterableInternal}} implies > an "iterable" materialization of the contents of a PCollection, which is > adapted to the desired user-facing type within a UDF (today the > PCollectionView "is" the UDF) > In practice, runners get adequate performance by special casing just a couple > of types of PCollectionView in a rather cumbersome manner. > The design to improve this is https://s.apache.org/beam-side-inputs-1-pager > and this makes the de facto standard approach the actual model. -- This message was sent by Atlassian JIRA (v6.3.4#6332)