PostgreSQL's commitfest clog

Posted Aug 13, 2021 9:18 UTC (Fri) by k3ninho (subscriber, #50375)
In reply to: PostgreSQL's commitfest clog by iabervon
Parent article: PostgreSQL's commitfest clog

Contemporary team practices talk about 'definition of done' where the community agree on the checklist of things to mark a task complete so it can be made generally available to users.

I like the 'attempted review' idea, in any case this needs extra effort in release management. I'd aim for making visual the amount of stuff at the review stage -- so that a contemporary approach like Kanban can stop people working on new items until the tasks In Review are down to some acceptable number (Toyota car manufacturing, where this came from, set this to zero for producing no more parts until stock is consumed).

You could visualise the 'attempted review' stage, too, for stuff that's failing to garner adequate technical expertise to move to a state where they're 'signed off'. I have an addiction to sticky notes on whiteboards, but online visual dashboards also exist.

K3n.

PostgreSQL's commitfest clog

Posted Aug 13, 2021 14:05 UTC (Fri) by mathstuf (subscriber, #69389) [Link] (1 responses)

> (Toyota car manufacturing, where this came from, set this to zero for producing no more parts until stock is consumed)

My understanding is that Kanban started at Toyota, but was misunderstood and simplified by other manufacturers into being just-in-time manufacturing (where stocks are kept lean as an end rather than as a side-effect of the underlying Kanban principles). Toyota themselves understood that some buffers are necessary and is why they fared better than other car manufacturers during shortages. This video[1] explains some of the backstory here.

[1] https://www.youtube.com/watch?v=b1JlYZQG3lI

PostgreSQL's commitfest clog

Posted Aug 19, 2021 9:12 UTC (Thu) by k3ninho (subscriber, #50375) [Link]

>stocks are kept lean as an end rather than as a side-effect of the underlying Kanban principles

I don't disagree -- I think that buffers are essential to flow (the real goal) but they need a whole-system view to find the not-too-little, not-too-much Goldilocks zone. An example discussed here is Bufferbloat with some WiFi devices having too large a set of buffers which caused problems with the dumb flow control of TCP sessions.

K3n.