blob: 5212f701a977e89dcb4199f1c3f9c41f65c5f8ba [file] [log] [blame]
= Continuous Improvement for Continuous Development
:author: Drew Foulks and Gavin McDonald, Infrafolk
[.notes]
--
dfoulks:
Hi I'm Drew Foulks, You might have seen me before at a convention or on
a Jira Ticket, I'm a member of the Apache Infrastructure Team and have
been since 2019. I've been a member of the Foundation for about 3 years.
TODO: gmcdonald:
--
== Before we get started
Thank you builds@ community.
[.notes]
--
dfoulks:
make a shout out to our builds@ community. for those of you who
don't know this is the community that has collected around the ASF's CI
systems. These are regular committer / members with build engineer / power user
experience that have taken up the mantle of build super hero to come to the
aid of ASF Infra time and time again with useful advice and helping hands.
these folks bring a project independent, solutions based perspective to the
foundations and I want to ensure that we include more than subtle nod to
all of the effort they've put in and perspective that they've provided to us.
thank you builds@ crew
TODO: gmcdonald:
--
== A year of changes, 2023
[%step]
* more listening
* better arm support
* ephemeral nodes for jenkins
* build caching
* better github actions support
[.notes]
--
dfoulks:
2023 was a big year for the Infrastructure team at the ASF. apachecon 2022
in New Orleans really showed us that in order to strengthen the Foundation
as a whole from an infra point of view, we needed to put more emphasis on
communicating with projects and committers not just so that we can address
pain points but also so that we can be more proactive in our endeavors
instead of becoming a reactive entity like so many corporate infra teams.
One of the common themes in the work we've been doing this last year is
listening. I'm sure that some of you have heard of or maybe even been to
an Infra roundtable. for those of you who haven't -- we've been hosting
these topical panel discussions for about a year now, and have encouraged
projects to bring us their input on a variety of topics. The CI improvement
initiative is no exception -- We've been taking stock of the CI concerns
brought to us both historically and in the past year during these meetings
We took up the threads and started to work to address those concerns by
making changes to the CI systems to make them more flexible and capable
of addressing the use cases that you, the committers and contributors,
have presented to us.
--
== 3rd party builds systems
[%step]
* Travis Exodus
* GitHub Actions
* Self hosted runners in Azure
[.notes]
--
dfoulks:
Early 2023 saw the wrapping up of our exit from TravisCI. At the time
GitHub actions presented an easy, well documented path for exiting travis.
which at the time was our primary goal, and it really appealed to many of
our users / Infra.
It's easy to pick up / migrate to, the vendor was trustworthy
and many people already undertstood the processes under the hood.
Since then adoption has only increased which has given rise to new processes.
The uptick of GitHub Actions usage brought with it the concerns of security and
scalability. The current situation is that actions are managed by whitelist and
require an infra ticket for new actions to be approved for use.
We're currently looking at streamlining this process.
gmcdonald:
* Can you discuss how the self hosted runners in Azure extend scalability?
* is there anything else that you'd like to say about this?
--
== Buildbot
[%step]
* project websites enabled blogs retirement
* .asf.yaml changes
[.notes]
--
dfoulks:
Infrastructure spent some time this year working on how it generates content.
this was a wide reaching initiative that reached right to the heart of what
are considered the Foundation critical core services that infrastructure has
always supported.
The way that project websites get built and served has changed a lot in the
last few years. As our means have changed, so have our methods, and while
ASF Infra structure internal website builders are nothing new, we have been
putting in the work to streamline the process for new projects at the foundation.
This year saw the retirement of the Blogs system in favor of a more friendly
and project-centric method of self-hosted blogs. Infra had been laying the
groundwork for this for some time with the pelican and jekyll builder work
changes to .asf.yaml have further streamlined the blogs changes into a
relatively simple self-serve offering. for more information on how this
gets done exactly, see us or our blogs expert after the presentation.
gmcdonald:
* is there anything you wanted to say here?
--
== Jenkins
[%step]
* Gradle Integrations
* Ephemeral Nodes
* ARM Support
* GPU Support
[.notes]
--
infrastructure spent a lot of time this year trying to get Gradle Enterprise
integrated with our jenkins systems.
now, ASF Infrastructure and Gradle bring you ge.apache.org. Additionally,
we are working to make build caching available to any project that wants it.
We've been working with beam and a few other projects as a pilot program
for the service and we've found:
### FACT:
Gradle reports an average time savings of 16 minutes per build over the
last 10K builds due to local and remote caching alone.
16 minutes of savings doesn't sound like a lot on its face, but beam
is running well over 10K builds a week. 10K builds @ 16 minutes is
111 days of time saved.
that's a lot of build cycle time!
Additionally, Gradle supports Maven and we look forward to rolling out
gradle integration to projects using maven so that they also can leverage
the time savings.
### Ephemeral nodes in Jenkins
We've further expanded Jenkins capbility by configuring ephemeral build
nodes for jenkins.
currently this has been deployed to ci-builds so if you have a job there
this applies especially to you:
we've got auto deploying EC2 instances now, you may not have noticed but
we've started breaking into the cloud by using a few nodes as `ubuntu` tagged
nodes to help bring down the queue time.
we _also_ have other nodes though
###### FACT
we added GPU (G4dn.16xlarge) nodes for GPU support with the `gpu` tag
we added ARM (T4g.Small) nodes for ARM support with the `arm` tag
and we custom built the image so that all of the tools that your builds
are already configured to use are in the exact same place.
we're also working on getting these images integrated with the new build
systems so that it's a one and done process.
--
== Questions and Comments