process 100 koji builds with logjuicer and evaluate them as a data set

[TomasTomecek]

Just discussed this with @jpodivin

Let's write a simple script that will:

• iterate over failed production builds in koji
• for every failed build, it will find last successful one in the same arch/package/chroot setup
• run logjuicer on these 2 build logs
• save the output
• [bonus points] send the output to chatgpt/bard and see what they say about it

[FrostyX]

• iterate over failed production builds in koji
• for every failed build, it will find last successful one in the same arch/package/chroot setup

We have a script that does this in Copr. I was supposed to polish it, so it can be run via Cron but it is still to be done. If Copr would be good enough, somebody can work on the remaining steps and we can glue it together afterward.

[TomasTomecek]

@FrostyX wow, amazing! please share it with me, I can do that for Copr

[TomasTomecek]

I have made solid progress on this: https://github.com/TomasTomecek/log-detective-website/tree/logjuicer-and-llm

Logjuicer is producing amazing snippets and GPT4 can give us excellent descriptions for problems such as compilation failures, pytest failures, "autotools"-like checks