metron-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Miklavcic <michael.miklav...@gmail.com>
Subject Re: [DISCUSS] Parser Aggregation in Management UI
Date Tue, 07 May 2019 14:33:52 GMT
This was my expectation as well.

Shane, Tibor, Tamas - how did you go about breaking this down into chunks
and/or microchunks when you collaborated offline? As Nick mentioned, you
obviously split up work and shared it amongst yourselves. Some explanation
around this process would be helpful for reviewers as well. We might be
able to provide better guidance and examples to future contributors as well.

I talked a little bit with Shane about this offline last week. It looks
like you guys effectively ran a local feature branch. Replicating that
process in a feature branch in Apache is probably what you guys should
be doing for a change this size. We don't have hard limits on line change
size, but in the past it's been somewhere around 2k-3k lines and above
being the tipping point for discussing a feature branch. Strictly speaking,
line quantity alone is not the only metric, but it's relevant here. If you
want to make smaller incremental changes locally, there's nothing to keep
you from doing that - I would only advise that you consider squashing those
commits (just ask if you're unclear about how to handle that) into a single
larger commit/chunk when you're ready to publish them as a chunk to the
public feature branch. So it would look something like this:

Commits by person locally
Shane: 1,2,3 -> squash as A
Tibor: 4,5,6 -> squash as B
Tamas: 7,8,9 -> squash as C

Commits by person in Apache
Shane:  A
Tibor: B
Tamas: C

We need to maintain a record of attribution. Your real workflow may not be
that cleanly delineated, but you can choose how you want to squash in those
cases. Even in public collaboration, there are plenty of cases where folks
submit PRs against PRs, abstain from accepting attribution, and it all gets
squashed down into one person's final PR commit. There are many options.

Hope this helps.

Best,
Mike

On Mon, May 6, 2019 at 8:19 AM Nick Allen <nick@nickallen.org> wrote:

> Have you considered creating a feature branch for the effort? This would
> allow you to break the effort into chunks, where the result of each PR may
> not be a fully working "master-ready" result.
>
> I am sure you guys tackled the work in chunks when developing it, so
> consider just replaying those chunks onto the feature branch as separate
> PRs.
>
>
>
> On Mon, May 6, 2019 at 5:24 AM Tibor Meller <tibor.meller@gmail.com>
> wrote:
>
> > I wondered on the weekend how we could split that PR to smaller chunks.
> > That PR is a result of almost 2 months of development and I don't see how
> > to split that to multiple WORKING parts. It is as it is a whole working
> > feature. If we split it by packages or files we could provide smaller
> > non-functional PR's, but can end up having a broken Management UI after
> > having the 1st PR part merged into master. I don't think that would be
> > acceptable by the community (or even by me) so I would like to suggest
> two
> > other option to help review PR#1360.
> >
> > #1 We could extend that PR with our own author comments in Github. That
> > would help following which code part belongs to where and why it was
> > necessary.
> > #2 We can schedule an interactive code walkthrough call with the ones who
> > interested in reviewing or the particular changeset.
> >
> > Please share your thoughts on this! Which version would support you the
> > best? Or if you have any other idea let us know.
> >
> > PS: I think the size of our PR's depends on how small independently
> > deliverable changesets we can identify before we starting to implement a
> > relatively big new feature. Unfortunately, we missed to do that with this
> > feature.
> >
> > On Fri, May 3, 2019 at 1:49 PM Shane Ardell <shane.m.ardell@gmail.com>
> > wrote:
> >
> > > NgRx was only used for the aggregation feature and doesn't go beyond
> > that.
> > > I think the way I worded that sentence may have caused confusion. I
> just
> > > meant we use it to manage more pieces of state within the aggregation
> > > feature than just previous and current state of grouped parsers.
> > >
> > > On Fri, May 3, 2019 at 1:32 AM Michael Miklavcic <
> > > michael.miklavcic@gmail.com> wrote:
> > >
> > > > Shane, thanks for putting this together. The updates on the Jira are
> > > useful
> > > > as well.
> > > >
> > > > > (we used it for more than just that in this feature, but that was
> the
> > > > initial reasoning)
> > > > What are you using NgRx for in the submitted work that goes beyond
> the
> > > > aggregation feature?
> > > >
> > > >
> > > >
> > > > On Thu, May 2, 2019 at 12:22 PM Shane Ardell <
> shane.m.ardell@gmail.com
> > >
> > > > wrote:
> > > >
> > > > > Hello everyone,
> > > > >
> > > > > In response to discussions in the 0.7.1 release thread, I wanted
to
> > > > start a
> > > > > thread regarding the parser aggregation work for the Management UI.
> > For
> > > > > anyone who has not already read and tested the PR locally, I've
> > added a
> > > > > detailed description of what we did and why to the JIRA ticket
> here:
> > > > > https://issues.apache.org/jira/browse/METRON-1856
> > > > >
> > > > > I'm wondering what the community thinks about what we've built thus
> > > far.
> > > > Do
> > > > > you see anything missing that must be part of this new feature in
> the
> > > UI?
> > > > > Are there any strong objections to how we implemented it?
> > > > >
> > > > > I’m also looking to see if anyone has any thoughts on how we can
> > > possibly
> > > > > simplify this PR. Right now it's pretty big, and there are a lot
of
> > > > commits
> > > > > to parse through, but I'm not sure how we could break this work out
> > > into
> > > > > separate, smaller PRs opened against master. We could try to
> > > cherry-pick
> > > > > the commits into smaller PRs and then merge them into a feature
> > branch,
> > > > but
> > > > > I'm not sure if that's worth the effort since that will only reduce
> > the
> > > > > number commits to review, not the lines changed.
> > > > >
> > > > > As an aside, I also want to give a little background into the
> > > > introduction
> > > > > of NgRx in this PR. To give a little background on why we chose to
> do
> > > > this,
> > > > > you can refer to the discussion thread here:
> > > > >
> > > > >
> > > >
> > >
> >
> https://lists.apache.org/thread.html/06a59ea42e8d9a9dea5f90aab4011e44434555f8b7f3cf21297c7c87@%3Cdev.metron.apache.org%3E
> > > > >
> > > > > We previously discussed introducing a better way to manage
> > application
> > > > > state in both UIs in that thread. It was decided that NgRx was a
> > great
> > > > tool
> > > > > for many reasons, one of them being that we can piecemeal it into
> the
> > > > > application rather than doing a huge rewrite of all the application
> > > state
> > > > > at once. The contributors in this PR (myself included) decided this
> > > would
> > > > > be a perfect opportunity to introduce NgRx into the Management UI
> > since
> > > > we
> > > > > need to manage the previous and current state with the grouping
> > feature
> > > > so
> > > > > that users can undo the changes they've made (we used it for more
> > than
> > > > just
> > > > > that in this feature, but that was the initial reasoning). In
> > addition,
> > > > we
> > > > > greatly benefited from this when it came time to debug our work in
> > the
> > > UI
> > > > > (the discussion in the above thread link goes a little more into
> the
> > > > > advantages of debugging with NgRx and DevTools). Removing NgRx from
> > > this
> > > > > work would reduce the numbers of lines changed slightly, but it
> would
> > > > still
> > > > > be a big PR and a lot of that code would just move to the component
> > or
> > > > > service level in the Angular application.
> > > > >
> > > > > Shane
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message