revi-archive/llvm-premerge-checks
1
0
Fork 0
Code Activity

Update monitoring_main_branch.md

Browse source 
fixed formatting, spelled out the background/problem statement
This commit is contained in:
ChristianKuehnel 2022-02-07 09:57:51 +01:00 committed by GitHub
parent 2739e3041b
commit 1444797003
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
docs/monitoring_main_branch.md
View file

@ -5,25 +5,24 @@ for use in pre-merge testing.
## Background
TODO explain background:
* Phab only has diffs, we need to apply these on a git base revision.
* Many of the base revisions are broken.
* pre-merge fails due to unrelated issues.
* goal: reduce % of broken revisions by getting them fixed faster
The stability of the pre-merge testing largely depends on the stability of the
LLVM main branch: Whenever something is broken on the main branch, also
pre-merge testing will fail: The patches from Phabricator are applied to some
revision on the main branch before they can be built. So the fewer revisions of
the main branch being broken, the more stable pre-merge testing will be.
## High-level Design
We propose to run a Buildbot worker on the main branch with the same
Docker image we're using for pre-merge testing. That worker shall check all
commits to main for build and test failures with regards to the configuration
we're using for pre-merge testing. Whenever these builds fails, Buildbot
notifies the commiters and gives them the opportunity to fix or revert their
patch.
We propose to run a Buildbot worker on the main branch with the same
Docker image we're using for pre-merge testing. That worker shall check all
commits to main for build and test failures with regards to the configuration
we're using for pre-merge testing. Whenever these builds fails, Buildbot
notifies the commiters and gives them the opportunity to fix or revert their
patch.
This is much faster than a having a human investigate the issue and notify the
committers. By having faster feedback the main branch is broken for fewer
revisions and this the probability of a false-positive pre-merge test is lower.
This is much faster than a having a human investigate the issue and notify the
committers. By having faster feedback the main branch is broken for fewer
revisions and this the probability of a false-positive pre-merge test is lower.
## Machine setup