[Labs-l] [Labs-announce] [Tools] Kubernetes picked to provide alternative to GridEngine

[Yuvi Panda]

We have picked kubernetes.io to provide an alternative to GridEngine
on Tool Labs! See https://phabricator.wikimedia.org/T106475 for more
details about the evaluation itself,
https://docs.google.com/spreadsheets/d/1YkVsd8Y5wBn9fvwVQmp9Sf8K9DZCqmyJ-ew-PAOb4R4/edit?pli=1#gid=0
for the spreadsheet with the evaluation matrix and
https://lists.wikimedia.org/pipermail/labs-l/2015-August/003955.html
for the previous email listing pros and cons of the others solutions
that we have considered.

== Rough Timeline ==

* October: Start beta testing by opening up Kubernetes to whitelisted
tools. Allows people to run arbitrary Docker images in Kubernetes,
both for continuous jobs and web services. If you are interested in
this,please add a comment to https://phabricator.wikimedia.org/T112824
* December: Work on migrational tooling to assist in switching from
GridEngine to Kubernetes should be in a good place. This will begin
with a '--provider=kubernetes' parameter to the webservice command
that will allow people to easily switch to Kubernetes for webservice.
We will have something similar implemented for jsub / jstart.
* January: Kubernetes cluster is opened up for all tools users. Fairly
complete switching between GridEngine and Kubernetes for at least
continuous jobs and webservices.

== What does this give developers? ==

# A software stack that is made up of widely-used and
actively-developed software components. We are looking to dramatically
reduce the surface of code that we write and maintain in-house.
# (When out of beta) More stability and reliability.
# It allows us to get rid of a lot of our customizations (the entire
webservice setup, for example) which has proven to be a lot of work
and flaky at times.
# We can migrate tools that don't require NFS off of it, and since it
has historically been one of our flakiest servies this allows tools to
opt-out of it and get a lot more reliability.
# Using Docker allows more user freedom in a structured way that is
easier to support.
# Has an active upstream / is used elsewhere as well (Google Container
Engine, Microsoft Azure, RedHat OpenShift, AWS, etc.), so much more
help / community available when users run into issues than we have now
# Probably more :) It opens up a lot of possibilities and I'm sure
developers will use it createively :)

== Do I need to change anything? ==

No, especially if you do not want to. We think Docker and Kubernetes
are exciting technologies, so we would like volunteers who are
interested in exploring these platforms to have the option of getting
their hands dirty. At the same time, it is important for us to avoid
complicating life for people for whom the current setup works well. If
we are successful, the migration to Docker and Kubernetes will be
seamless, and users of Tool Labs will not have to know what either
Docker or Kubernetes are, let alone how to operate them.

We will begin by offering arbitrary Docker image execution only.
Eventually we will make it super-easy to switch between the current
setup and Kubernetes, and allow people to be able to use this without
having to know what Docker is or what Kubernetes is. That will slowly
be built over the next few months.

Absolutely nothing changes for developers or users right now.

== Will GridEngine be going away? ==

Not anytime soon!

Thanks!

-- 
Yuvi Panda T
http://yuvi.in/blog

_______________________________________________
Labs-announce mailing list
Labs-announce at lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/labs-announce