Digging up this old thread as any offer to help really deserves a response.  I 
had travelled to help dear old mom these two weeks and didn't get back to you 
after I had a bit more bandwidth.

> On Sep 18, 2021, at 9:31 AM, Jenkins, Rodney J (Rod) 
> <[email protected]> wrote:
> 
> David,
> 
> Thank you for the insights and explanation!
> 
> I completely understand the technical debt and the challenge of making this 
> better during a release.  I would like to jump in and see where I can help.  
> My problem is I am not a java developer.  What I am good it is automating 
> tasks, if I can be taught to execute them.
> 
> The big ask is:  Would anyone want to take the time with me to educate me on 
> what has to happen for a release (not during a release)?  I am thinking that 
> we could set up a dummy repo that has some simple small java code in it to be 
> a dummy TomEE release candidate.  Create some dummy destinations that mimic 
> where the artifacts must be placed.  Once I understand the process, I can see 
> about making it repeatable.
> 
> Personally, I would like to see it done in a way that someone with lesser 
> skills (like, but not necessarily,  me) does releases.  The way I see it now 
> is the heavy hitters do the releases.  I think their time would be better 
> spent on the technical debt, bugs, etc.  Maybe we could find a small few that 
> would be wiling the be release specialists.  I know some where I work that 
> MAY be interested.  If someone would teach me, I would teach them.

You have the right spirit.  In the early days of Geronimo I did the majority of 
releases and while I did a good job, wasn't the job I wanted to be doing and 
could see how me doing all the releases created a knowledge vacuum that 
actually hurts the community.

What I did was suggest we start a system where each release had a pilot and a 
copilot.  The pilot would be an experienced person who know what they were 
doing, the copilot would be someone learning.  Next release the copilot would 
become the pilot and do the release and a new person would become the copilot.  
Each release the documentation got a bit better, the technical debt paid down a 
bit.  Things improved, releases got more frequent and quality went up.

A significant blocker, however, is that a large number of the steps can't be 
done unless you're an Apache committer.  Not because of policy, just because 
they require access to systems only committers can reach.  You won't be able to 
run a lot of the commands.

That's not a show stopper, it just means it our creativity in how to leverage 
you will be heavily challenged.  I'm up for it if you are.

I have been trying to revitalize our release tools in java for automating as 
many release tasks as possible.  I do think that's the right way to go as the 
majority of people who do releases are java developers.  In the past I've 
written elaborate scripts in bash and inevitably they decay as I'd be the only 
one who understood them.  That doesn't necessarily mean there's no role for 
scripting.

One of the hardest parts of doing releases is remembering all the steps that 
have to be done and double checking they were done correctly.  This is 
absolutely something you could script up and run.  I will say in all my 
attempts to document release processes, I've noticed even in the best 
circumstances they're always a little out of date.  Turn your back even a 
little and they become just out of date enough no one even looks at the doc 
anymore, than it's game over.

If we had a release auditing script that could be pretty amazing.  It could be 
the checklist people use to "see" the release process as a whole.  Again I'm 
imagining something like the System V startup output where there's one line per 
task.  Each line is a step that has to be done a green/red colored status on 
the right.

Here are some things I frequently see done wrong and not noticed:

 - Forgetting to update the keys file.  Topic of the original thread.  Because 
it's something you only need to do once in a while, it's easy to forget.  
There's a release tools command to help people do it, but you have to remember 
you rotated your key a few months ago and need to run it.  Perfect thing to 
audit.

 - Missing signatures once published.  We've had infrastructure ask us to add 
signatures because we've forgotten a few times.  There's already a 
script/command to do it, but nothing to check if people use them.  As 
mentioned, people stop using scripts they find confusing, so they decide to 
wing it instead.  A script to audit this part would be great.  Apache infra has 
such a script and does run it every few months, but it's not ideal when they're 
the ones point it out.

 - Leaving SNAPSHOT or old TomEE dependencies in the examples or build.  I 
spent three days cleaning out references in the examples to older TomEE 7 and 8 
snapshots missed by past releases.  The way you check is delete your 
~/.m2/repository directory and build the release tag from git.  If there is 
anything with a SNAPSHOT dependency in there, that's a fail.  If there are any 
jars from previous releases in there, that's a fail.  We should not pass these 
release votes, we should fix the source and roll again.  Again, there is a tool 
to do this (maven) but it doesn't always catch everything.  It only renames 
references to the current version to the next version.  If there's a reference 
to a version say two releases ago, it will not see it.  It will also not rename 
references for any modules that are commented out, which happens with the 
examples from time to time.  It really needs auditing.

 - Updating the generated download page HTML.  Frequently people will directly 
update the download.html page, just hacking in the new release.  The problem of 
course is all the html pages are generated and it will inevitably get 
overridden and set back to the previous release the next time the website is 
generated.  It doesn't seem to matter that repo containing generated HTML files 
has "do make updates here, this is all generated" written on it.  We still get 
this mistake from committers and contributors alike.  We in fact just had a PR 
opened against the generated html repo.  On release, people need to know and 
remember to update the markdown/asciidoc source which is in a different repo.  
There's a specific file we could audit via a simple grep and flag up 
"tomee-site-generator download.adoc page updated [FAIL]"

 - Forgetting to publish the maven artifacts.  Only happened once, but there 
was an occasion where we forgot to click the publish button on the nexus 
repository holding the maven distributions of all the release jars.  They 
expired and were deleted.  That meant we had a release that was on the website, 
in the download area for people that wanted to just grab a zip file, but was 
nowhere maven user could access.  During the release process a special 
temporary release area is created on repository.apache.org, we need to keep an 
eye on that and audit that it was in fact released and everything there has 
reached Maven Central.


I could probably think of a half dozen more checks, but the above is hopefully 
enough to paint the picture that it would be useful.  There are systems to do 
everything above, just you know ... darn people being all human :)  None of the 
above is enough detail to actually implement most of this, so we can have 
better discussions if we actually want to do any of it.

Another aspect to this idea, Richard Zowalla has been doing a lot of fun 
experimentations with Github Actions.  There is definitely scripting potential 
there and some of the above could be triggered when release tags show up in 
git.  Unlike setting up Jenkins jobs, it's way easier to contribute to Github 
Actions as it's all git-based so PRs work perfectly.  You could test on a 
forked repo and then submit a PR.  Some real potential to make outside help 
possible with this capability.


Some ideas and thoughts.  Hope not an overwhelming amount.  :)

Any thoughts extremely welcome!


-David

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to