HI,

-1 from me.

You're right that 3 IPMC +1s over 72 hours satisfies the release policy on 
paper, so my objection isn't about the number of votes. The 72 hours is there 
so everyone on the PMC has a chance to take part in the decision to release. 
For a podling, the PMC is the IPMC, and the IPMC takes part on general@, not on 
a podling dev list that most of the IPMC doesn't read. So 3 IPMC members voting 
on dev@ doesn't mean the IPMC has approved the release.

A release is an act of the Foundation, and that comes from the IPMC vote. It's 
what the legal protection for release managers and PMC members is based on. The 
release policy says that deviating from it may weaken that protection and 
affect the Foundation's insurance, and that it requires board approval first. 
Any change to release policy would also need VP Legal approval.

Reviewing after the release also assumes you can pull it if something is wrong, 
and you can't. Maven Central doesn't remove artefacts, crates.io only yanks, 
PyPI and npm burn the version. Whatever is wrong is out there under the Apache 
name for good. The general@ vote is where people other than the mentors look at 
the release, and that's where most issues get found.

I also don't think the data supports it. Of the last nine podling releases on 
general@, only three had 3 or more IPMC votes on dev@. The two with none, Fesod 
and Seata, were the two where issues were found on general@. Where podlings do 
wait, it's usually for a second or third IPMC vote because mentors haven’t 
voted, and that's a mentoring problem, not a policy one.

Running the two votes in parallel is a different thing, and that may be worth 
considering, as long as both run for 72 hours and the PPMC vote is closed with 
a result before anything is published. The Incubator release policy would need 
to be revised to allow it.

If a release really does need to be quick, say for a known security issue, the 
expedited release process already applies to podlings. Both votes can be 
shorter and the vote email needs to say why, and the board needs to be 
notified, usually in the next Incubator report or earlier if it's urgent. 
That's for exceptional cases only.

Something similar came up in 2019 [1], where the idea was for mentors to vote 
on dev@ and close on general@ 72 hours later if nobody objected. What we do 
now, carrying IPMC votes from dev@ and still running the general@ vote for 72 
hours, is more or less where that ended up.

Kind regards,

Justin

[1] https://lists.apache.org/thread/g8k7xbgpqtxm5cfl919r72s2xm43drmq

Reply via email to