[
https://issues.apache.org/jira/browse/BEAM-25?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15408886#comment-15408886
]
ASF GitHub Bot commented on BEAM-25:
------------------------------------
GitHub user kennknowles opened a pull request:
https://github.com/apache/incubator-beam/pull/793
[BEAM-25] WIP: Tweeze StateSpec out of StateTag
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.
- [ ] If this contribution is large, please file an Apache
[Individual Contributor License
Agreement](https://www.apache.org/licenses/icla.txt).
---
R: @bjchambers @tgroh
This was a hack sprint. The tests pass for the SDK and direct runner; I've
introduced a minor issue to the Flink runner that I will fix tomorrow.
My thoughts when doing this:
- I can prep `StateSpec` now so it is ready for incorporation into `DoFn`
whenever that work takes off.
- Since the `StateSpec` carries the disjoint union now, the binder should
visit that.
- `StateTag` = `String id` + `StateSpec`. First step would be to express
it that way, second step might be to delete it.
It actually turned out to be a nice cleanup, but there are wrinkles:
- Since the `State` has to be able to know where to write (in the general
case) it still needs an `id`, so the visitor needs an id. But `StateSpec`
doesn't have one so it is just passed along. So the visitor just becomes a
curried version of the prior.
- That's all fine, but then the `StateTable#get` also needs the spec
because it lazily inits based on it. This is the only time it is used, since
beyond then it is contained in the state cell.
- And then to hack up the `CopyOnAccessInMemoryStateInternals` I even had
to re-build the tag `id`. So eliminating `StateTag` entirely would just mean
more parameters in a bunch of places.
So I have this feeling that actually there may be a simpler visitor
pattern, or no visitor pattern, that becomes more natural.
Anyhow, I'm not happy with the change, and certainly haven't polished
`StateSpecs` to where it needs to be for user consumption, but I wanted to put
this out for early feedback.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/kennknowles/incubator-beam StateSpec
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/incubator-beam/pull/793.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 #793
----
commit 488e8955000ee905ab26635e0efbd0834a3d0dcf
Author: Kenneth Knowles <[email protected]>
Date: 2016-08-05T03:50:28Z
Create StateSpec parallel to StateTag
commit e6294682daba9835a030c146389bc633e8f280a5
Author: Kenneth Knowles <[email protected]>
Date: 2016-08-05T04:48:48Z
Make StateTag carry a StateSpec separately from its id
----
> Add user-ready API for interacting with state
> ---------------------------------------------
>
> Key: BEAM-25
> URL: https://issues.apache.org/jira/browse/BEAM-25
> Project: Beam
> Issue Type: Sub-task
> Components: sdk-java-core
> Reporter: Kenneth Knowles
> Assignee: Kenneth Knowles
> Labels: State
>
> Our current state API is targeted at runner implementers, not pipeline
> authors. As such it has many capabilities that are not necessary nor
> desirable for simple use cases of stateful ParDo (such as dynamic state tag
> creation). Implement a simple state intended for user access.
> (Details of our current thoughts in forthcoming design doc)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
