Dogwood Retrospective

Dogwood Retrospective

Date

Feb 16, 2016

Participants

@Ned Batchelder (Deactivated) @Joel Barciauskas (Deactivated) @Feanil Patel (Deactivated) @Alison Hodges (Unlicensed)

 

Retrospective

What did we do well?

  • Django 1.8 upgrade automated

  • Building nightly boxes now

  • Interrupt-driven devops interaction was ok

    • One devops person for whole process was good

  • "Venn diagram of success"

  • Got good with vagrant and virtualbox snapshots

What should we have done better?

  • The Dogwood wiki page had details that Ned overlooked

  • Named release process page didn't have important details (like docs details)

  • Some projects didn't know what "being in Dogwood" meant.

    • Analytics

    • Otto

  • Pulling together release notes was difficult

    • Weekly release notes are aimed at a different audience

  • Some repos didn't release properly, so the Dogwood tag is on an old commit.

  • Didn't test all the configurations

    • Sandbox

    • 12.04

  • Not clear which of the many installation methods need to be supported in a named release

  • Unclear about milestones getting to final (are we waiting on docs, devops, or eng at the moment?)

  • Needed to announce Dogwood rc cut!

    • It was announced in scrum-of-scrums

    • We never sent an email about it

  • readthedocs was very difficult

  • Tagging tool was broken at the last minute

  • Still working to make docs completely Cypress-clean

  • Lack of transparency about dogwood status

  • Doc process page was old and out of date

  • Test cycle is slow, upgrading to dogwood could take 90 min

 

Things We Know Now?

  • Being in Dogwood means "these repos work together", maybe not install, maybe not supported.

Actions

@Alison Hodges (Unlicensed) update the doc process page
@Ned Batchelder (Deactivated) update the Process to Make a Named Release page
@Feanil Patel (Deactivated) Somehow get people to tag PRs for needing release notes in the next named release?
@Ned Batchelder (Deactivated) work out how to reduce the number of install options, with Feanil: fullstack is like sandbox.sh, but with a base box to avoid numpy/scipy compile time. This is now https://openedx.atlassian.net/browse/OPEN-1241.
@Feanil Patel (Deactivated) try to get numpy or scipy wheels to reduce installation time, to reduce test cycles
@Ned Batchelder (Deactivated) Make an explicit list of configurations to test.  This is https://openedx.atlassian.net/browse/OPEN-1302
@Ned Batchelder (Deactivated) Send an email to eng pointing to the Eucalyptus page, asking for updates