I do think minimal window duration is a meaningful concept for
WindowFns, but from the pragmatic perspective I would ask is it useful
enough to require all implementers of WindowFn to specify it (given
that a default value of 0 would not be very useful).
On Mon, Apr 26, 2021 at 10:05 AM Jan
Confirming that the cut date is 4/28/2021 (in two days), right?
On Wed, Apr 21, 2021 at 4:41 PM Tomo Suzuki wrote:
>
> Thank you for the preparation!
>
> > a few responses that some high priority changes
>
> Would you be willing to share the items for visibility?
There are several PRs in flight
d squash by default
> in the upcoming squash bot even for single commit PRs?
>
Does squash-and-merge in that case preserve the commit as-is if there's
only one? In that case, there'd be no issues of history. (I opted to not
comment on 1-commit PRs to be less chatty.)
>
> On Thu, Apr 22, 2
even there there is high variance. I think the issue is
"oh, I didn't think to squash vs. merge" rather than "who cares, I always
press merge anyway" in which case a timely reminder will go a long way.
Kenn
>
> [1]
> https://lists.apache.org/thread.html/4a65fb0b66935c
er paid more attention (if not
> yet) on these “non code” things before reviewing/merging a PR.
>
> [1] https://github.com/apache/spark/blob/master/dev/merge_spark_pr.py
>
>
> On 22 Apr 2021, at 01:28, Robert Bradshaw wrote:
>
> I am also in the camp that it often makes sense
d and verified PRs.
>> We could solve the unwanted commit issue if we have a policy to always
>> "Squash and Merge" PRs with rare exceptions.
>>
>> I agree jira/PR titles could be better, I'm not sure what we can do about
>> it aside from reminding comm
+1 to better descriptions for JIRA (and PRs). Thanks for bringing this up.
For merging unwanted commits, can we automate a simple check (e.g. with
github actions)?
On Wed, Apr 21, 2021 at 8:00 AM Tomo Suzuki wrote:
> BEAM-12173 is on me. I'm sorry about that. Re-reading committer guide
> [1],
herrypicked
>>>> to the release branch.
>>>>
>>>> Kenn
>>>>
>>>> On Mon, Apr 19, 2021 at 8:34 PM Kenneth Knowles
>>>> wrote:
>>>>
>>>>> OK it sounds like I need to re-roll the artifacts in question. I don'
I commented on the doc, but I'm also in favor of dropping "harness."
On Mon, Apr 19, 2021 at 3:10 PM Tyson Hamilton wrote:
> I'm in favor of dropping "harness" and going with "sdk_container_image". I
> don't feel like the word "harness" adds value or clarity.
>
> On Mon, Apr 19, 2021 at 11:34
It looks like the wheels are also versioned "2.29.0.dev".
Not sure if it's important, but the source tarball also seems to contain
some release script changes that are not reflected in the github branch.
On Mon, Apr 19, 2021 at 8:41 AM Kenneth Knowles wrote:
> Thanks for the details, Valentyn
While working on BEAM-6597, I ran into the issue that not all runners
support the non-deprecated metrics protocol. Additionally, there is no way
for the runner to inform the worker whether it supports this protocol.
I was thinking about adding a "runner_capabilities" field to ProvisionInfo
[1]
I'd be happy with yapf + docformatter + isort, but I'd like to understand
why yapf lets breakable lines go longer than 80 chars.
On Fri, Apr 9, 2021 at 4:19 PM Brian Hulette wrote:
> Currently we have two different format checks for the Python SDK. Most
> format checks are handled by yapf,
Python (again a language) has a slower release cycle, fairly strict
backwards compatibility stance (with the ability to opt-in before changes
become the default) and clear ownership for maintenance of each minor
version until end-of-life (so each could be considered an LTS release).
; with some "coder-version-registry". I suppose there
> might have been a discussion about this in the past, does anyone know of
> any conclusion?
>
> Jan
> On 4/6/21 10:54 PM, Robert Bradshaw wrote:
>
> I do think there's value in having an LTS release, if there's suffici
I do think there's value in having an LTS release, if there's sufficient
interest to fund it (specifically, figuring out who would be backporting
fixes and cutting the new releases).
On Mon, Apr 5, 2021 at 1:14 PM Elliotte Rusty Harold
wrote:
> Hi,
>
> I'd like to return to the discussion
Fn, etc. I'm not sure how realistic that is though. It
>> would be nice if this argument worked the other way
>>
>> [1] https://theneuralbit.github.io/beam-site/pydoc/inherited-members
>> [2]
>> https://theneuralbit.github.io/beam-site/pydoc/inherited-members/apache_beam.transforms.
Congratulations!
On Fri, Apr 2, 2021 at 10:19 AM Chamikara Jayalath
wrote:
> Congrats Tomo!
>
> On Fri, Apr 2, 2021 at 9:54 AM Brian Hulette wrote:
>
>> Congratulations Tomo! Well deserved :)
>>
>> On Fri, Apr 2, 2021 at 9:51 AM Yichi Zhang wrote:
>>
>>> Congratulations!
>>>
>>> On Fri, Apr
I agree with the idea of adding a label to this, which can be attached to
real bugs, etc. that are in specific components. I don't think it makes
sense to try to offer user support though jira though--if it's a bug let's
rephrase the question as such, otherwise we could post the question (and
+1 to an example. In particular, are these inherited members grouped in
such a way that it makes it easy to ignore them once they get to "low" in
the stack? If it can't be per-module, is there a "nice" set of ancestors to
avoid (as it seems this option takes such an argument).
On Wed, Mar 31,
1929, I
>> removed it from the release blockers since there is a workaround (use a
>> NamedTuple type), but it's probably worth cherrypicking the fix.
>>
>> On Thu, Mar 25, 2021 at 4:44 PM Robert Bradshaw
>> wrote:
>>
>>> This could be https://issues.apac
This could be https://issues.apache.org/jira/browse/BEAM-11929
On Thu, Mar 25, 2021 at 4:26 PM Robert Bradshaw wrote:
> This is definitely wrong. Looking into what's going on here, but this
> seems severe enough to be a blocker for the next release.
>
> On Thu, Mar 25, 2021 at 3:39
This is definitely wrong. Looking into what's going on here, but this seems
severe enough to be a blocker for the next release.
On Thu, Mar 25, 2021 at 3:39 PM Xinyu Liu wrote:
> Hi, folks,
>
> I am playing around with the Python Dataframe API, and seemly got an
> schema issue when converting
lectionTuple is extensible
too, as one could add more (or better) outputs in the future without
changing the signature.
>
>
>
>> Kenn
>>
>> On Wed, Mar 24, 2021 at 5:36 PM Robert Bradshaw
>> wrote:
>>
>>> Yeah, the entire input is not alw
gt;
>> Am I reading this wrong?
>>
>> Kenn
>>
>> On Wed, Mar 24, 2021 at 4:35 PM Alex Amato wrote:
>>
>>> How about a PCollection containing every element which was successfully
>>> written?
>>> Basically the same things which were passed into it.
>
nd just add new
> funtionality. Though, we need to follow the same pattern for user API and
> maybe even naming for this feature across different IOs (like we have for
> "readAll()” methods).
> >
> > I agree that we have to avoid returning PDone for such cases.
> >
I would encourage flex templates over further proliferation of
ValueProviders.
On Tue, Mar 23, 2021 at 12:42 PM Alexey Romanenko
wrote:
> I think you are right - now with SDF support in KafkaIO it should be
> possible to determine the number of splits in the runtime and support
> ValueProviders
Thanks, that looks quite useful.
A not on POJOs, Serializable has several disadvantages (e.g.
non-deterministic coding for key grouping, less efficient serialization).
You could look into making them compatible with Beam schemas.
On Mon, Mar 22, 2021 at 1:11 PM Cristian Constantinescu
wrote:
>
+1
On Wed, Mar 17, 2021 at 5:48 PM Kenneth Knowles wrote:
> Hi all,
>
> Please vote on the proposal for Tomo Suzuki to become a committer in the
> Apache Beam project, as follows:
>
> [ ] +1, Approve the proposal for the candidate to become a committer
> [ ] -1, Disapprove the proposal for the
nt hourly output, I cannot trigger
>> source with lower frequency. If I trigger source with hourly, but do not
>> propagate this as fast as possible, I'm inevitably introducing additional
>> latency (that's the definition of "not as fast as possible") in downstream
>> p
+1
On Tue, Mar 16, 2021 at 4:00 PM Kenneth Knowles wrote:
> Please review the release of the following artifacts that we vendor:
> * beam-vendor-grpc-1_36_0
>
> Hi everyone,
>
> Please review and vote on the release candidate #1 for the version 0.1, as
> follows:
> [ ] +1, Approve the release
Do we now support 1.8 through 1.12?
Unless there are specific objections, makes sense to me.
On Fri, Mar 12, 2021 at 8:29 AM Alexey Romanenko
wrote:
> +1 too but are there any potential objections for this?
>
> On 12 Mar 2021, at 11:21, David Morávek wrote:
>
> +1
>
> D.
>
> On Thu, Mar 11,
I am in the same boat: I'm subscribed to both but the dev@ ones are more
visible to me. That being said, I do think it's valuable to have segregated
discussion (otherwise why have two lists), and "lurkers" on users@ can
learn from questions answered there.
I would say that it is on us to redirect
I've noticed this sometimes for Python as well: Jenkins is happy with the
exact same tests that Github checks fails on.
On Thu, Mar 4, 2021 at 8:40 AM Alexey Romanenko
wrote:
> Hi,
>
> Does anyone know why some Java Tests, that run as Github checks, fail? For
> example for this PR [1], this [2]
Any reason not to simply pass these parameters into the DoFn constructor?
On Mon, Mar 1, 2021 at 3:42 PM Xinyu Liu wrote:
> Hi, all,
>
> Currently the @Setup method signature in DoFn does not support any
> arguments. This is a bit cumbersome to use for use cases such as creating a
> db
Thanks. I've filed https://issues.apache.org/jira/browse/BEAM-11882 .
If you want to take a stab at fixing it, you could try replacing the
arguemnt passed to merge_accumulators at
https://github.com/apache/beam/blob/release-2.28.0/sdks/python/apache_beam/transforms/combiners.py#L963
with a new
agate down is
> confusing (it's also a bit confusing for Windows, but with Windows the
> propagation at least makes sense). The fact that users rarely have access
> to the actual GBK operation means that allowing them to specify triggers on
> their sinks is the best approach.
>
> On Mon
E. We still
>>> remove ALREADY_MERGED. This would allow a later GBK to make no sense
>>> because there's not likely to be any merging for the same reason. But
>>> merging WindowFns don't have to work like sessions so they might merge
>>> based on some other interesting cr
ially if there are very few user-authored
> merging WindowFns out there (and I agree that this is probably true).
> Choice (2) also has the benefit that it matches Python and that it is
> trivial to implement.
>
> Kenn
>
> On Thu, Feb 18, 2021 at 3:18 PM Robert Bradshaw
> wr
d actually mean, that the
>> correct place, where to specify triggering is not Window PTransform, but
>> the GBK, i.e.
>>
>> input.apply(GroupByKey.create().triggering(...))
>>
>> That would imply we simply have default trigger for all GBKs, unless
>> explicitly cha
>> wrote:
>>
>>> Are you referring to the list that starts with "dev@ archives and
>>> statistics"?
>>>
>>> How about I add a "process" section at the end with that information?
>>>
>>> On Tue, Feb 16, 2021 at 5:46 PM
+1 (binding)
I validated the signatures and package contents, as well as running a small
Python pipeline in a fresh virtual environment.
On Fri, Feb 19, 2021 at 1:36 PM Chamikara Jayalath
wrote:
> +1
>
> I ran the release candidate verification script and updated the
> spreadsheet [1]
> Also,
I think you're right about Python. I think it's fine for the SDK to
prohibit (or require explicit user action) for ambiguous things like
stacked sessions. This illegal state wouldn't generally need to be
represented in proto (but maybe it'd be nice for quicker errors in cross
language).
On Thu,
On Wed, Feb 17, 2021 at 1:56 PM Kenneth Knowles wrote:
>
> On Wed, Feb 17, 2021 at 1:06 PM Robert Bradshaw
> wrote:
>
>> I would prefer to leave downstream triggering up to the runner (or,
>> better, leave upstream triggering up to the runner, a la sink trigg
vior on Java on non-portable runners. At
least until we can figure out what we really want and add it to the model.
>
> Kenn
>
> On Wed, Feb 17, 2021 at 11:16 AM Robert Bradshaw
> wrote:
>
>> OK, so to move forward, shall we update the default Sessions to not do
>> th
stages. However continuation triggers silently switching to
>> synchronized processing time has defeated that, and it wasn't clear to
>> users why.
>>
>> On Wed, Feb 17, 2021 at 11:12 AM Robert Bradshaw
>> wrote:
>>
>>> On Fri, Feb 12, 2021 at 9:0
less there is some special consideration why it doesn't matter. So
> I wonder what happens when a pipeline has a few different joins.
>
> Kenn
>
> On Fri, Feb 12, 2021 at 12:37 AM Robert Bradshaw
> wrote:
>
>> Yes, unless you manually set the timestamp combiner to e
On Fri, Feb 12, 2021 at 9:09 AM Kenneth Knowles wrote:
>
> On Thu, Feb 11, 2021 at 9:38 PM Robert Bradshaw
> wrote:
>
>> Of course the right answer is to just implement sink triggers and
>> sidestep the question altogether :).
>>
>&
e-pull-requests.storage.googleapis.com/13871/contribute/become-a-committer/index.html#what-are-the-traits-of-an-apache-beam-committer
> (Note I don't have control over the preview staged at test-beam.surge.sh,
> so that's now out of date for this page)
>
> +Kenneth Knowles +Robert Brads
to fix?
>
> On Fri, Feb 12, 2021 at 12:25 AM Robert Bradshaw
> wrote:
>
>> The default timestamp combiner used to be earliest as well.
>>
>> On Fri, Feb 12, 2021 at 12:10 AM Reuven Lax wrote:
>>
>>> IIRC, this was introduced because at the time users complain
e allowed customizing the timestamp combiner, so
> maybe this is less of a problem now?
>
> On Thu, Feb 11, 2021 at 10:53 PM Robert Bradshaw
> wrote:
>
>> On Wed, Feb 10, 2021 at 8:03 PM Kenneth Knowles wrote:
>>
>>>
>>>
>>> On Wed, Feb 10, 2021
On Wed, Feb 10, 2021 at 8:03 PM Kenneth Knowles wrote:
>
>
> On Wed, Feb 10, 2021 at 2:24 PM Alex Amato wrote:
>
>>
>>
>> On Wed, Feb 10, 2021 at 12:14 PM Kenneth Knowles wrote:
>>
>>> On a PR (https://github.com/apache/beam/pull/13927) we got into a
>>> discussion of a very old and strange
Of course the right answer is to just implement sink triggers and sidestep
the question altogether :).
In the meantime, I think leaving AfterSynchronizedProcessingTime in the
model makes the most sense, and runners can choose an implementation
between firing eagerly and waiting some amount of
Are there any substantive changes from what we did last year?
On Thu, Feb 11, 2021 at 1:40 PM Brittany Hermann
wrote:
> Dear Project Management Committee,
>
> The Beam Summit is a community event funded by a group of sponsors and
> organized by a steering committee formed by members of the Beam
oving the details. Our bullets highlight specific aspects of the
> code of conduct. Specifically, we split up one of the items into separate
> bullet points, and we did not emphasize "be concise" or "step down
> considerately". I am OK going ahead with this change
+1 based on some basic browsing around.
It would have been nice if the old and new site could have been generated
from the same content, to keep the old one around for a bit (if needed, in
that case we might have more confidence nothing was lost), but probably not
worth making and staging a copy.
When we were using maven rather than gradle, it was impossible to make the
direct runner a dependency for core tests without making the direct runner
a dependency for core itself (which was cyclic). I would be super happy if
this got fixed; having to jump through hoops to run the majority of core
ble to stage files to SDK
> containers, so it's something we should consider making into a general
> feature, perhaps based on the artifact API.
>
+1
>
> On Thu, Feb 4, 2021 at 3:52 PM Robert Bradshaw
> wrote:
>
>> On Thu, Feb 4, 2021 at 3:33 PM Kyle Weaver wrote:
>
should have led with this. Someone wanted to mount credentials into the
> SDK harness [1]. So in this particular case the user just wants to mount
> files into their SDK harness, which is a pretty common use case, so
> resource hints are probably a more appropriate solution.
>
> [1]
> https://lists.apache.org/thre
On Thu, Feb 4, 2021 at 12:38 PM Kyle Weaver wrote:
> So, an external transform is uniquely identified by its URN. An external
>> transform identified by a URN may refer to an arbitrary composite which may
>> have sub-transforms that refer to different environments. I think with the
>> above
You could write a DoFn that "consumes" the output of write as a side input
and touches the file manually. E.g.
write_result = ... | beam.io.WriteToText(...)
p | beam.Create([None]) | beam.Map(lambda unused_none, unused_side:
create_file(), unused_side=write_result)
where create_file()
n ARM64
>
>
> On Tue, Jan 26, 2021 at 8:48 PM Robert Burke wrote:
> >
> > I believe so.
> >
> > The Go SDK requires in most instances for a user to Register their DoFns
> at package init time, linked to the type/functions fully qualified path as
> detemined
> set to amd64 linux at present, but that's a separate issue.
>
> [1] https://golangcookbook.com/chapters/running/cross-compiling/
> [2] https://golang.org/cmd/cgo/
>
> On Tue, Jan 26, 2021, 10:25 AM Robert Bradshaw
> wrote:
>
>> +1
>>
>> I don't think it wou
+1
I don't think it would be that hard to build and release arm-based docker
images. (Perhaps just a matter of changing the docker file to depend on a
different base, and doing some cross-compile. That would suss out whether
we're inadvertently taking on any incompatible dependencies.)
Thanks, Piotr. Well deserved.
On Fri, Jan 22, 2021 at 9:22 AM Tyson Hamilton wrote:
> Congrats Piotr! Well deserved.
>
> On Fri, Jan 22, 2021 at 9:18 AM Ismaël Mejía wrote:
>
>> Congratulations Piotr ! Thanks for all your work !
>>
>> On Fri, Jan 22, 2021 at 5:33 PM Alexey Romanenko
>> wrote:
is is it provides the runner the
ability to say to these (potentially long-running usefns) "please stop
gracefully as soon as you can."
>
> On Thu, Jan 21, 2021 at 8:34 PM Robert Bradshaw
> wrote:
> >
> > I don't know that SDF vs. BoundedSources changes things
Congratulations, Cham!
On Thu, Jan 21, 2021 at 3:13 PM Brian Hulette wrote:
> Great news, congratulations Cham!
>
> On Thu, Jan 21, 2021 at 3:08 PM Robin Qiu wrote:
>
>> Congratulations, Cham!
>>
>> On Thu, Jan 21, 2021 at 3:05 PM Tyson Hamilton
>> wrote:
>>
>>> Woo! Congrats Cham!
>>>
>>> On
I don't know that SDF vs. BoundedSources changes things here--for both one
can implement take(n) by running until one has N elements and then
canceling the pipeline.
One could have a more sophisticated First(n) operator that has a
"back-edge" to checkpoint/splits the upstream operators once a
On Wed, Jan 20, 2021 at 1:48 PM Kenneth Knowles wrote:
> Yes, completely sound nullability checking has been added to the project
> via checkerframework, based on a large number of NPE bugs (1-5% depending
> on how you search, but many other bugs likely attributable to
> nullness-based design
bert Burke wrote:
> > > Hmmm. Fair. I'm mostly concerned about the pathological case where we
> end
> > > up with a distinct Environment per transform, but there are likely
> > > practical cases where that's reasonable (High mem to GPU to TPU, to
> ARM)
> &g
;
> On Wed, Jan 13, 2021 at 3:59 AM Brian Hulette wrote:
>
>>
>>
>> On Mon, Jan 11, 2021 at 11:00 AM Robert Bradshaw
>> wrote:
>>
>>> On Mon, Jan 11, 2021 at 10:38 AM Brian Hulette
>>> wrote:
>>>
>>>> I spoke with Gri
Yes, a PTansform can have no sub-transforms, as long as it only returns its
inputs. Updating the docs would be a good idea.
On Tue, Jan 12, 2021 at 1:04 PM Brian Hulette wrote:
> A recent bug with SqlTransform on Dataflow Runner V2 [1] revealed an
> interesting ambiguity in the Beam model: it's
cognizing and resolving conflicts on the owners of the website-revamp
branch.
It might be worth highlighting an example of a content change that makes
any of these workflows difficult.
> [1] https://github.com/apache/beam/tree/website-revamp
>
> On Mon, Jan 11, 2021 at 10:03 AM R
A site-wide freeze during which there was a huge, rushed code dump was not
the most effective way to manage or review the large website changes last
time, and I don't think it would be a good idea to attempt that again.
Instead, can we create a parallel directory/site in our repo, incrementally
Somewhat belated, but another +1 (binding); the release artifacts and
signatures all look good to me. Thanks for doing this release!
On Fri, Jan 8, 2021 at 11:01 AM Pablo Estrada wrote:
> Hi all,
> I'm happy to announce that we have unanimously approved this release.
>
> There are 8 approving
+1 to keeping the distinction between Runner and Engine as Kenn described,
and cleaning up the site with these in mind (I don't think the term engine
is widely used yet).
On Wed, Jan 6, 2021 at 11:15 AM Yichi Zhang wrote:
> I agree with what kenn said, in most cases I would refer to the term
>
If readers are expensive to create, this seems like an important (and not
too difficult) optimization.
On Mon, Dec 21, 2020 at 11:04 AM Jan Lukavský wrote:
> Hi Boyuan,
>
> I think your analysis is correct - with one exception. It should be
> possible to reuse the reader if and only if the
There are two ways to emit multiple outputs: either to multiple distinct
PCollections (e.g. withOutputTags) or multiple (including 0) outputs to a
single PCollection (the difference between Map and FlatMap). In full
generality, one can always have a CombineFn that outputs lists (say *) followed by
I agree. Borrowing the mutation detection from the direct runner as an
intermediate point sounds like a good idea.
On Mon, Dec 21, 2020 at 8:57 AM Kenneth Knowles wrote:
> I really think we should make a plan to make this the default. If you test
> with the DirectRunner it will do mutation
If all it takes is bumping these numbers up a bit, that seems like a
reasonable thing to do ASAP. (I would argue that perhaps they shouldn't be
static, e.g. it might be preferable to start emitting results right away,
but use larger batches for the steady state if there are performance
benefits.)
+1 (binding). I've verified the release artifacts and signatures, and
validated some simple pipelines with a freshly installed wheel.
On Thu, Dec 10, 2020 at 10:00 AM Brian Hulette wrote:
> +1 (non-binding). Ran a Python pipeline using DataframeTransform on
> DirectRunner and DataflowRunner
ew is that there aren't two types of timers--"static" timers are just
dynamic timers with a fixed (aka static) tag (that we provide for you as a
convenience).
On Thu, Dec 10, 2020 at 10:43 AM Robert Bradshaw
> wrote:
>
>> Yep.
>>
>> A slight variant on this is to
et(timestamp) -> timer.set(timestamp, dynamic_timer_tag=a_tag)
> timer.clear() -> timer.clear(dynamic_timer_tag=a_tag)
>
> and have the default value of dynamic_timer_tag to be empty (the special
> case)
>
> On Wed, Dec 9, 2020 at 5:12 PM Robert Bradshaw
> wrote:
>
>> O
On Wed, Dec 9, 2020 at 3:48 PM Kyle Weaver wrote:
> Possibly a dumb question, but: if "the static timer is just a special case
> of the dynamic timer," why do we need to use different classes at all?
>
I agree, I would argue that there is little if any value to the user
to distinguish between
Makes sense to me.
On Thu, Dec 3, 2020 at 12:29 PM Boyuan Zhang wrote:
>
> Hi folks,
>
> As we are reaching a stable state on Splittable DoFn APIs both in Java and
> Python SDK, I'm proposing to remove Experimental annotations from these APIs.
>
> I have opened one PR[1] to do so. Please feel
good enough. We have a proposal, we
> have clear boundaries for it. It's fine if the discussion continues, but I
> see no evidence of concerns that should prevent starting an implementation,
> because it seems we'll need both anyway.
>
> On Wed, Nov 25, 2020, 10:25 AM Robert Bra
I for one am ready to see a PR.
+1
> On Mon, Nov 23, 2020, 6:02 PM Kenneth Knowles wrote:
>>
>>
>>
>> On Mon, Nov 23, 2020 at 3:04 PM Robert Bradshaw wrote:
>>>
>>> On Fri, Nov 20, 2020 at 11:08 AM Mirac Vuslat Basaran
>>> wrote:
>&
ds to be detailed.
> > >>
> > >> I am curious about how the end user APIs of this will look maybe in
> > >> Java or Python, just an extra method to inject a Map or via Java
> > >> annotations/Python decorators?
> > >>
> > >> We
Guage certainly seems wrong for DistributionResult. Yes, using a
Histogram would be a welcome PR.
On Thu, Nov 19, 2020 at 12:58 PM Kyle Weaver wrote:
>
> What are the advantages of using a Histogram instead of a Gauge?
>
> Also, check out this design doc for adding histogram metrics to Beam if
cation via
>> Environments but it could also just delegate this to a resource
>> manager which is what we could do for example for GPU (or data
>> locality) cases on the Spark/Flink classic runners. If we tie this to
>> environments we will leave classic runners out of the
I agree things like GPU, high-mem, etc. belong to the environment. If
annotations are truly advisory, one can imagine merging environments
by taking the union of annotations and still producing a correct
pipeline. (This would mean that annotations would have to be a
multi-map...)
On the other
so thanks for the clarification
> because
> it makes more sense now, my initial understanding was that it was more to
> 'debug' SDK Harness processed elements (that's why I mentioned Instructions)
> but
> it is clearly beyond that.
Yeah, I think this will be useful.
> On Thu, Oct 29, 2
On Thu, Oct 29, 2020 at 3:18 AM Ismaël Mejía wrote:
>
> Thanks for sharing,
>
> I was initially confused by the title/terminology, I thought it was
> about an end-user transform but this is a 'protocol' for a runner to
> get the string representation of an element encoded by a SDK Harness
>
+1 (binding).
I verified the release artifacts and signatures, and tried a couple of
Python pipelines from an install of a wheel in a fresh virtual
environment. All looks good to me.
On Thu, Oct 22, 2020 at 4:54 PM Tyson Hamilton wrote:
>
> +1
>
> I went through the Nexmark queries and
On Fri, Oct 23, 2020 at 10:16 AM Luke Cwik wrote:
>
> An additional thing I forgot to mention was that if we only had portable
> runners our BOM story would be simplified since we wouldn't have the runner
> on the classpath and users would have a consistent experience across runners
> with
Welcome. Looking forward to the website improvements.
I would like to call out that it'd be a good idea to have a plan for how
this can be developed/reviewed incrementally. We were able to get the last
major website change in, but huge monolithic PRs that change the world are
difficult to review
Support for Flink 1.11 is
https://issues.apache.org/jira/browse/BEAM-10612 . It has been
implemented and will be included in the next release (Beam 2.25). In
the meantime, you could try building yourself from head.
On Fri, Oct 16, 2020 at 4:39 AM Kishor Joshi wrote:
>
> Hi Team,
>
> Since the
Thanks, Ning and Ahmet.
+1 (binding) Approve the release.
On Fri, Oct 16, 2020 at 1:34 PM Ning Kang wrote:
> Sorry, if you cannot see the missing thread history in the previous
> thread, here is another copy:
>
> On Fri, Oct 16, 2020 at 9:55 AM Robert Bradshaw
> wrote:
>
&
Personally, I think that it makes sense to not see an old timer after one
has cleared (or moved) it. This is consistent with the idea that the
behavior should be the same if every element was processed in its own
bundle. The fact that a user can (sometimes, but not always) see an old
timer after
should
> deprecate a v1 IO ONLY when we have full feature parity in the v2 version.
> I think we don't have a replacement for AWSv1 S3 IO so that one should not
> be
> deprecated.
>
> On Tue, Sep 15, 2020 at 6:07 PM Robert Bradshaw
> wrote:
> >
> > The 10x-100x
The 10x-100x ratio looks like an answer right there about (non-)suitability
for deprecation. The new question would be *why* people are using the v1
APIs. Is it because it was the original, or that it's been around longer,
or it has more features?
201 - 300 of 1301 matches
Mail list logo