launchpad-dev team mailing list archive

[Gary Poster <gary.poster@xxxxxxxxxxxxx>]

= Meta =
 * Changed name of this meeting/email from "postmortem" to "retrospective".
 * Had meeting Thursday because gary_poster is out tomorrow (also gmb & 
frankban).

= Attendance =
 * bac, gmb, gary_poster
 * benji sick
 * frankban holiday

= Project status =

 * Missed our biweekly call yesterday because gary_poster was out. 
gary_poster will send document to flacoste and lifeless for review; they 
can determine next steps (reschedule meeting? send out notes as is?)
 * Only one previously-seen test failure remaining: intermittent 
memcache ConnectionError also seen on non-parallel lpbuildbot.
 * Other new test failures to be filed.
 * Still waiting on IS for testing machine.  Many more IS tasks for 
future before project is done.
 * Keeping our tools working in the face of underlying changes in lxc, 
ec2 images, etc.
 * Polishing & improving experience (e.g., buildbot reporting)
 * Looking forward to possibly working on related features (lpsetup, 
parallel ec2, etc.) while we wait on IS tasks

= Learning and problem solving =

== Summary ==

 * We improved zope.testing fork's maintenance
 * gmb and bac both had cards that took longer in coding than they 
liked.  In discussion, it sounded like pairing, or at least asking for 
help sooner, would have helped.  We proposed an experiment that we aim 
for cards to take no more than one day within a single lane, and any 
card that stays longer requires us to identify it in our daily calls and 
treat it as a problem to be solved.  gary_poster opened 
http://support.leankitkanban.com/requests/5369 as a possible way to help 
with this.  We will evaluate on retrospective call in two weeks, April 
20.  Success is defined as a higher speed of cards through board, as 
well as a higher feeling of accomplishment on team.

== In detail ==

=== Tricks ===

No new tricks

=== Successes to learn from ===

 * More of a "good news" item, but in regards to last week's 
zope.testing problems, a few aspects of the situation are improved.
   * gary_poster had made the p4 branch from lifeless' p4 egg changes. 
 He had missed a change that lifeless made, and so it was not in p5. 
This caused us problems.  We updated the p4 branch so it now accurately 
reflects the changes in p4.
   * We made a new branch lp:~launchpad/zope.testing/3.9.4-fork to be 
the main branch for our fork.  This should be the destination for all of 
our changes in the future for this fork (rather than making -pN branches).
   * We changed versions.cfg to have all the information requested by 
https://dev.launchpad.net/PolicyForDocumentingCustomDistributions 
(though not in the requested format).

=== Problems ===

gmb: Working on the test automation script has taken too long.  The task 
is to make a script that fires up a juju environment and then runs a 
test, so we can verify regularly that everything, from the configuration 
to the test, works.  Why has it taken too long?
 * One day was consumed with Precise issues causing problems with his 
development environment
 * Trying to do TDD on a script that calls Juju was surprisingly 
complex, and the best automation he could come up with meant hacking his 
~/.juju
 * Precise ec2 images had a problem with apt-get update (now resolved) 
that caused automation to break.
What could have made this go better?
 * pairing/sharing would have been better.

bac: Similar story to gmb.  juju on ec2 testing: took awhile to get back 
into testing.  fragility in lxc, ec2 images and apt caused slowdowns.

gary_poster: duplicated work (needs new bullet point in type of problem 
for meeting): benji and I both did zope.testing integration in LP; 
gary_poster mentioned it in an email but not clearly enough.  Gary 
should have clarified what the branch was doing.  Surprised both Brad 
and Benji.  What should have made this go better?
 * Be more explicit in communications

gary_poster: A bit of this week and last was rework: parallel testing 
machinery was working two weeks ago, and we've been trying to fix it: 
ec2, zope.testing, and apt on ec2 all caused issues.  What should have 
been better?
 * We should have identified the automated test harness task (gmb's 
current card) sooner.  We initially worked on automated tests, but not 
automated tests of the automated tests--that sounds silly, but the 
deliverable is the automated tests.

gary_poster: two items discussed were about needing pairing or sharing 
sooner. gmb: what rate of work indicates that I need to get help? 
gary_poster: 1 idea: if card doesn’t move for one day, ask about it. 
Maybe a problem, maybe not.