In an attempt to keep this from becoming unbearably long, I will try
to keep my responses short, but I would be happy to elaborate.  That's a
fairly good timeline and summary, but here are some clarifications in
corresponding order:

- The plugin history is quite short and you can probably get a good bit of
context just by looking at the commits
<https://github.com/apache/metron-bro-plugin-kafka/commits/master>.
- The plugin is only useful to the bro community, but it is rather popular.
- The Bro team created the idea of bro packages, which can include bro
plugins, bro scripts, or BroControl plugins.  So, instead of having a
'plugins' repo, they moved to have a 'packages' repo which is by default
referenced by a bro-pkg tool they wrote for package management.
- I believe I kicked this off (or at least I did in my head) when I started
complaining <https://issues.apache.org/jira/browse/METRON-348?filter=-2>
about the plugin divergence that occurred due to the move to bro/plugins
(the right move at the time), but Metron's use of a local directory that
hadn't been kept up to date.  My current efforts are an attempt to make
sure this doesn't happen again, and to take advantage of the bro-pkg
benefits.
- The gap between ~3/31 and actual progress on 11/12 is completely on me -
I had intentions of doing this work sooner but failed to do so.
- You can most definitely still install/use the bro plugin without using
bro-pkg.  In fact, the README in my PR
<https://github.com/JonZeolla/metron-bro-plugin-kafka/blob/7402a48cab8be7802fa0c395648597ea6a1cbd71/README.md>
still has the instructions on how to do so.

Q1:  The simple explanation is that the only thing that makes a plugin a
bro package is the inclusion of a bro-pkg.meta file
<https://github.com/apache/metron-bro-plugin-kafka/blob/master/bro-pkg.meta>,
and it includes a build_command which could easily be manually performed to
install by hand (assuming dependencies are met).

I've worked with other projects that use submodules and while I'm fine
discussing it, I suggest that we don't implement it.  I put together a
quick example of why here[1], using the bro project as an example since
it's top of mind.

Q2:   I think the answer to Q1 answers this.  There is absolutely nothing
stopping a git clone && cd $dir && configure && make && make install, but
using bro-pkg to install/load takes into account dependencies
<https://github.com/apache/metron-bro-plugin-kafka/blob/master/bro-pkg.meta#L10-L14>
and
unit tests
<https://github.com/apache/metron-bro-plugin-kafka/blob/master/bro-pkg.meta#L6>
when it is loaded (and thus fails early and more intuitively).  It only
must be a separate repo (or, more technically correct, a git branch that
includes only the package) because of how bro-pkg works.  If you'd like to
get an idea of how this would work in application for Bro users, you can
see my test instructions here
<https://github.com/apache/metron-bro-plugin-kafka/pull/4> (specifically
step #3).  If a 0.1 tag gets pushed to apache/metron-bro-plugin-kafka, the
command could be `bro-pkg install metron-bro-plugin-kafka --version 0.1` or
`bro-pkg install apache/metron-bro-plugin-kafka --version 0.1` due to this
<https://github.com/bro/packages/blob/master/apache/bro-pkg.index> (the
--force is just to remove user interaction, for an ansible spin-up).


1:  To clone the Bro git repo, you must run `git clone --recursive
https://github.com/bro/bro` (note the --recursive).  Not too big of a deal,
but requires that you remember it and existing instructions/blog posts may
give users inaccurate steps.  Let's make this worse and try to checkout
their latest release, v2.5.2, and automatically update the submodules
appropriately via `git checkout v2.5.2 --recurse-submodules`.  This fails
because aux/plugins (https://github.com/bro/plugins) was removed since
their latest release.  Okay, we can work around this using `git checkout
v2.5.2` and then remember to `git submodule update` every time you checkout
a release or branch.  But because they have nested submodules, we actually
need to run `git submodule update --recursive`.  I can't imagine opting
*into* a workflow anything like this.  There are other options as well,
such as git subtrees, but those I am less familiar with.

Jon

On Mon, Nov 27, 2017 at 8:59 PM Otto Fowler <ottobackwa...@gmail.com> wrote:

> I am not sure that our use of the plugin necessarily equates to it being
> implicitly coupled to Metron.  It seems like the Right Thing To Do[image:
> ™], esp.
>  for an Apache project would be to make this available for use by the
> greater bro community.
> Unless we expect to do extensive iterative work on the plugin, which would
> then make the decision to spin it out now premature.
>
> Then again, I might be wrong ;)
>
>
> On November 27, 2017 at 19:58:11, Matt Foley (ma...@apache.org) wrote:
>
> [Please pardon me that the below is a little labored. I’m trying to
> understand the implications for both release and use, which requires some
> explanation as well as the two questions needed. Q1 and Q2 below are
> probably the same question, asked in slightly different contexts. Please
> consider them together.]
>
> So this made me go back and look at the history that caused us to put the
> bro plugin in a separate repo. As best I can see, this was in
> https://issues.apache.org/jira/browse/METRON-813 , which cites an email
> discussion thread. Also please see
> https://issues.apache.org/jira/browse/METRON-883 for background on the
> plugin itself.
>
> As best I can assemble the many bits brought up in the threads, the reasons
> to put it in a separate repo was:
> - The plugin was thought to be useful to multiple clients of bro and kafka,
> including Storm and Spark, as well as Metron.
> - Originally the bro project was maintaining bro plugins and it was thought
> they might adopt this one.
> - Bro then formalized their plugin framework BUT dumped all plugins out of
> their sphere of maintenance.
> - As of 3/31/2017, Nick said that “the [bro] package mechanism requires
> that a package live within its own repo”. Jon said “the bro packages model
> doesn't allow colocation with anything else.”
> - So on 3/31 Jon opened METRON-813, and the metron-bro-plugin-kafka repo
> was created a few days later. But Metron wasn’t actually modified to remove
> the metron-sensors/bro-plugin-kafka/ subdirectory and start using the
> plugin from the metron-bro-plugin-kafka repo until Nov 12 – two weeks ago!
> – with https://issues.apache.org/jira/browse/METRON-1309 .
> - Presumably the need to have metron-bro-plugin-kafka in a separate repo
> remain valid, if the bro plugin mechanism is used. But obviously there are
> (non-conforming) ways to build the plugin as part of metron, and install it
> in a way that works.
>
> Q1. I think that last statement needs some explanation. Nick or Jon, can
> you please expand on it, especially wrt how the end user installs the
> plugin once the plugin is built the two different ways? And whether it’s
> still valuable to have a separate repo for the plugin?
>
> Nick suggests using a submodule approach to managing the bro plugin, for
> Metron versioning purposes. As I understand it, this would continue the
> existence of the metron-bro-plugin-kafka repo, but copy it into the metron
> code tree for building, versioning, and release purposes. Git submodules
> are documented here: https://git-scm.com/book/en/v2/Git-Tools-Submodules .
> We would use the submodule capability to clone the metron-bro-plugin-kafka
> source code into a subdirectory of Metron at the time one clones the metron
> repo. It would then be released with Metron as part of the source code
> release for a given version of Metron. Part of the way submodules are
> managed, is that git stores the SHA1 hash of the submodule into a file
> named .gitmodules, which in turn gets saved when you do a git push. So
> indeed submodules would ensure that everyone cloning a given version of
> metron would get the expected “version” (sha, actually) of
> metron-bro-plugin-kafka.
>
> This sounds like a good idea, although it isn’t without cost. Submodules
> impose the need for additional commands to actually get a copy of the
> submodule source, and if the plugin repo advanced beyond the version in a
> metron repo, it causes some ‘git status’ artifacts that could be confusing
> to folks who aren’t familiar with submodules. But these can be documented.
>
> Q2. Nick, what I’m not clear about is the process by which the
> metron-bro-plugin-kafka would be built and “plugged in” by (a) metron
> developers, and (b) end users. If it “must” be in a separate repo to be
> successfully built and managed by the bro plugin mechanism, does that mean
> it can’t be built from the copy in the Metron source tree? Yet until
> November, that’s exactly what we were doing. Do we go back to doing that?
> What does that mean wrt users installing the plugin?
>
> Thanks for your patience in reading this far.
> --Matt
>
>
> On 11/27/17, 2:58 PM, "James Sirota" <jsir...@apache.org> wrote:
>
> I agree with Nick. Since the plugin is tightly coupled with Metron why not
> just pull it into the main repo and version it with the rest of the code?
> Do we really need the second repo for the plug-in?
>
> Thanks,
> James
>
>
>
> 16.11.2017, 08:06, "Nick Allen" <n...@nickallen.org>:
> >> I would suggest that we institute a release procedure for the package
> >> itself, but I don't think it necessarily has to line up with metron
> >> releases (happy to be persuaded otherwise). Then we can just link metron
> >> to metron-bro-plugin-kafka by pointing to specific
> >> metron-bro-plugin-kafka releases (git tags
> >> <
> http://bro-package-manager.readthedocs.io/en/stable/package.html#package-
> >> versioning>
> >> ).
> >> Right now, full-dev spins up against the
> >> apache/metron-bro-plugin-kafka master branch, which is not a good idea
> to
> >> have in place for an upcoming release. That is the crux of why I think
> we
> >> need to finalize the move to bro 2.5.2 and the plugin packaging before
> our
> >> next release (working on it as we speak).
> >> Jon
> >
> > ​I replayed Jon's comments from the other thread above.​
> >
> > My initial thought, is that I would not want to manage two separate
> release
> > processes. I don't want to have a roll call, cut release candidates and
> > test both.
> >
> > I was thinking we would just need to change some of the behind-the-scenes
> > processes handled by the release manager. This is one area where I had
> > thought using a submodule in Git would help.
> >
> > On Thu, Nov 16, 2017 at 9:58 AM, Nick Allen <n...@nickallen.org> wrote:
> >
> >> + Restarting the thread to include mentors.
> >>
> >> The code of the 'Kafka Plugin for Bro' is now maintained in the external
> >> repository that we set up a while back.
> >>
> >> - Metron Core: git://git.apache.org/metron.git
> >> - Kafka Plugin for Bro: git://git.apache.org/
> >> metron-bro-plugin-kafka.git
> >>
> >> (Q) Do we need to change anything in the release procedure to account
> for
> >> this?
>
> -------------------
> Thank you,
>
> James Sirota
> PMC- Apache Metron
> jsirota AT apache DOT org
>
-- 

Jon

Reply via email to