Try find & improve some bottleneck in Debian project Hideki Yamane <henrich @ debian.org>
● There are a lot of issues among Debian project, so we need MORE contributors to fix it. But - how do we get more contributors? ● As process management (Theory of Constraints), it says "most of issues are caused by "bottleneck"". If we would fix it "bottleneck", then many issues wil be gone (wow! :) ● In this session, I want to share my view/idea to find and fix bottleneck in Debian and get feedback.
$ find /debianproject -name bottleneck -print ● NEW queue ● all NEW packages should be reviewed by ftpmasters to get into repository ● ftpmasters should check whole package, then accept or reject it ● it takes many days (2,3 or 4 months, now)
wrong assumptions ● all NEW packages should be reviewed by "(only)" ftpmasters to get into repository ● ftpmasters should check "whole" package, then accept or reject it – if some reliable person would review package and check its review, isn't it enough?
Do NOT Burden with Yourself, please ● We are all project members, probably we can help you a bit :-)
add "preprocessor" for NEW queue ● review by contributors -> can reject obvious fault earlier ● Improve "upload -> check -> reject -> fix -> reupload" cycle – don't need to wait for 2 or 3 months to be rejected ;)
from serial to parallel ● "serial" process by ftpmasters (several) -> "parallel" process by contributors (hundreds) ● faster (we're multi-core monster machine! :) ● less load average (pressure) for each worker
contributors = buffer cache ● stable output – review process speed would NOT depend on ftpmasters workload by adding buffer cache
what ftpmasters should do? ● collect reviews from contributors and say "Go/Nogo(need-more-review)" to uploaders and contributors who reviewed it – maybe mailing list is better to share "howto" knowledge and make teach some patterns for review
reason why it's restricted to ftpmasters? ● Guess: probably because of "distribute" risk – copyright violation, patent, etc... – then use GPG encryption ● first, for only @debian.org address is in keyring is ok – it's already "trusted" person. then expand to others...
If we do it, what is expected to happen? ● Make unstable more attractive for developers – Al your software are belong to us :)
review contributor ● if they would be well trained, maybe would be future ftpmasters/assistants (more improvements :)
how to check it's success (or not) ● Statics: daily package numbers and average days for stay in "NEW" queue