I would prefer if you'd start a vote with a new cleaned up proposal.
On 18.02.2019 15:23, Robert Metzger wrote:
I added "Runtime / Configuration" to the proposal:
https://cwiki.apache.org/confluence/display/FLINK/Proposal+for+new+JIRA+Components
Since this discussion has been open for 10 days, I assume we have reached
consensus here. I will soon start renaming components.
On Wed, Feb 13, 2019 at 10:51 AM Chesnay Schepler <ches...@apache.org>
wrote:
The only parent I can think of is "Infrastructure", but I don't quite
like it :/
+1 for "Runtime / Configuration"; this is too general to be placed in
coordination imo.
On 12.02.2019 18:25, Robert Metzger wrote:
Thanks a lot for your feedback Chesnay!
re build/travis/release: Do you have a good idea for a common parent for
"Build System", "Travis" and "Release System"?
re legacy: Okay, I see your point. I will keep the Legacy Components
prefix.
re library: I think I don't have a argument here. My proposal is based on
what I felt as being right :) I added the "Library / " prefix to the
proposal.
re core/config: From the proposed components, I see the best match with
"Runtime / Coordination", but I agree that this example is difficult to
place into my proposed scheme. Do you think we should introduce "Runtime
/
Configuration" as a component?
I updated the proposal accordingly!
On Tue, Feb 12, 2019 at 12:19 PM Chesnay Schepler <ches...@apache.org>
wrote:
re build/travis/release: No, I'm against merging build system, travis
and release system.
re legacy: So going forward you're proposing to move dropped features
into the legacy bucket and make it impossible to search for specific
issues for that component? There's 0 overhead to having these
components, so I really don't get the benefit here, but see the
overhead.
I don't buy the argument of "people will not open issues if the
component doesn't exist", they will just leave the component field blank
or add a random one (that would be wrong). In fact, if you had a
storm/tez component (that users would adhere to) then it would be
_easier_ to figure out whether an issue can be rejected right away.
re library: If you are against a library category, what's your argument
for a connector category?
re tests: I don't mind "tests" being removed from tickets about test
instabilities, but you specified the migration as "rename E2E tests"
which is not equivalent.
Under what category would you file modifications to
flink-test-utils-junit?
I would propose to not differentiate between e2e and other tests; I
would go along with "Test infrastructure", and remove the major "Tests"
category.
re core/config: As an example, where (under Runtime) would you place the
introduction of the ConfigOption class?
On 11.02.2019 11:31, Robert Metzger wrote:
Thanks a lot for your feedback!
@Timo:
I've followed your suggestions and updated the proposed names in the
wiki.
Regarding a new "SQL/Connectors" component: I (with admittedly not much
knowledge) would not add this component at the moment, and put the SQL
stuff into the respective connector component.
It is probably pretty difficult for a user to decide whether a but
belongs
to "SQL/Connector" to "Connectors/Kafka" when Kafka in SQL does not
work.
@Chesnay:
- You are suggesting to rename "Build System" to "Maven" and still
merge
it
with "Travis", "Release System" etc. as in the proposal?
- "Runtime / Control Plan" vs "Runtime / Coordination" -- I changed the
proposal
- Re. "Documentation": Yes, I think that would be better in the long
run.
We are already in a situation where there are groups within the
community
focusing on certain areas of the code (such as SQL, the runtime,
connectors). Those groups will monitor their components, but it will
be a
lot of overhead for them to monitor the "Documentation" component.
We can also try to assign documentation components to both
"Documentation"
and the affected component, such as "Runtime / Metrics".
- Removed "Misc / " prefix.
- "Legacy Components": Usually legacy components usually have very few
tickets. "Flink on Tez" has 13, "Storm Compat" ~30, and JIRA has a bulk
edit feature :)
The benefit of having it generalized is that people will probably not
add
tickets to it.
- "Libraries /" prefix: I don't think that it is necessary. Some
libraries
might grow in the future (like the Table API), then we need to rename.
the "flink-libraries" module does contain stuff like the sql client or
the
python api, which are already covered by other components in my
proposal
--
so going with the maven module structure is not an argument here.
- "End to end infrastructure" and "Tests: The same argument as with the
"Documentation" applies here. The maintainers of Kafka, Metrics, ..
should
get visibility into "their" test instabilities through "their"
components.
Not many people will feel responsible for the "Tests" component.
For "Core" and "Configuration", I will move the tickets to the
appropriate
components in "Runtime /".
For "API / Scala": Good point. I will add that component.
How to do it? I will just go through the pain and do it.
Best,
Robert
On Fri, Feb 8, 2019 at 2:40 PM Chesnay Schepler <ches...@apache.org>
wrote:
Some concerns:
Travis and build system / release system are entirely different. I
would
even keep the release system away from the build-system, as it is more
about the release scripts and documentation, while the latter is about
maven. Actually I'd just rename build-system to maven.
Control Plane is a term I've never heard before in this context; I'd
replace it with Coordination.
The "Documentation" descriptions refers to it as a "Fallback
component".
In other words, if I make a change to the metrics documentation I
shouldn't use this component any more?
I don't see the benefit of a `Misc` major category. I'd attribute
everything that doesn't have a major category implicitly to "Misc".
Not a fan of a generalized "Legacy components" category; this seems
unnecessary. It's also a bit weird going forward as we'd have to touch
every JIRA for a component if we drop it.
How come gelly/CEP don't have a Major category (libraries?)
"End to end infrastructure" is not equivalent to "E2E tests".
Infrastructure is not about fixing failing tests, which is what we
partially used this component for so far.
I don't believe you can get rid of the generic "Tests" component;
consider any changes to the `flink-test-utils-junit` module.
You propose deleting "Core" and "Configuration" but haven't listed any
migration paths.
If there's a API / Python category there should also be a API / Scala
category. This could also include the shala-shell. Note that the
existing Scala API category is not mentioned anywhere in the document.
How do you actually want to do the migration?
On 08.02.2019 13:13, Timo Walther wrote:
Hi Robert,
thanks for starting this discussion. I was also about to suggest
splitting the `Table API & SQL` component because it contains already
more than 1000 issues.
My comments:
- Rename "SQL/Shell" to "SQL/Client" because the long-term goal might
not only be a CLI interface. I would keep the generic name "SQL
Client" for now. This is also what is written in FLIPs,
presentations,
and documentation.
- Rename "SQL/Query Planner" to "SQL/Planner" a query is read-only
operation but we support things like INSERT INTO etc.. Planner is
more
generic.
- Rename "Gelly" to "Graph Processing". New users don't know what
Gelly means. This is the only component that has a "feature name". I
don't know if we want to stick with that in the future.
- Not sure about this: Introduce a "SQL/Connectors"? Because SQL
connectors are tightly bound to SQL internals but also to the
connector itself.
- Rename "Connectors/HCatalog" to "Connectors/Hive". This name is
more
generic and reflects the efforts about Hive Metastore and catalog
integration that is currenlty taking place.
Thanks,
Timo
Am 08.02.19 um 12:39 schrieb Robert Metzger:
Hi all,
I am currently trying to improve how the Flink community is handling
incoming pull requests and JIRA tickets.
I've looked at how other big communities are handling such a high
number of
contributions, and I found that many are using GitHub labels
extensively.
An integral part of the label use is to tag PRs with the component /
area
they belong to. I think the most obvious and logical way of tagging
the PRs
is by using the JIRA components. This will force us to keep the JIRA
tickets well-organized, if we want the PRs to be organized :)
I will soon start a separate discussion for the GitHub labels.
Let's first discuss the JIRA components.
I've created the following Wiki page with my proposal of the new
component,
and how to migrate from the existing components:
https://cwiki.apache.org/confluence/display/FLINK/Proposal+for+new+JIRA+Components
Please comment here or directly in the Wiki to let me know what you
think.
Best,
Robert
