Hi,
On 5/11/23 15:36, Seyram Komla Sapaty wrote:
So far, the selected users I've reached out to are
maintainers who have
their tools sorted in the "Needs Custom Image" column(excluding those
needing multistack support)on the grid migration project board[0].
My understanding is that most of the tools in the "needs custom image"
column need either multistack support or custom Apt packages installed
in the image, neither of which are currently supported. So starting from
that column seems like a very strange decision to me.
In this next phase, we are slowly expanding the
selected users from the
hundreds of tools that are still in the 'Backlog' column on the
migration board, with emphasis on the ones maintaining Python tools.
Your previous message indicates that you were planning to send a large
amount of invites next Monday (May 15th). What does "slowly" mean here?
I still believe that sending any non-small amount of invites should not
happen at this stage. There is a large amount of known issues that make
debugging very difficult[0], cause confusing and unexplainable
failures[1] or completely break support for the most popular[2] runtime
on Toolforge[3].
[0]:
https://phabricator.wikimedia.org/T336225
[1]:
https://phabricator.wikimedia.org/T336360
[2]:
https://k8s-status.toolforge.org/images/
[3]:
https://phabricator.wikimedia.org/T335865
The only thing that's going to happen if we unnecessarily rush this
release is that people are going to get a mental image where Kubernetes,
our tooling around it and the Build Service are difficult to use and are
full of bugs and annoying limitations, which is going to make it even
harder and slower to get everyone to migrate off the Grid. We've made
this mistake before and I would really like to avoid it this time.
Taavi