[action] Update workflow to parse & monitor pending automation PRs. (#16446)

Why I did it
There are many automation PRs pending for PR checker failure issue.
As PR number grows, github api to list prs comes to its limit.
We need to monitor and send alert for these PRs.

Work item tracking
Microsoft ADO (number only): 25064441
How I did it
For auto-cherry pick PRs:
- more than 3 days, comment @author to check
- more than 10 days, stop comment.
- more than 28 days, comment @author PR will be closed
- more than 30 days, close PR

For submodule update HEAD PRs:
- more than 3 days, send alert(submodule PR)

How to verify it
Which release bra
This commit is contained in:
Liu Shilong 2023-09-07 13:34:34 +08:00 committed by GitHub
parent 7d2e3cb011
commit 52568ceab0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -16,58 +16,74 @@ jobs:
set -e
echo ${TOKEN} | gh auth login --with-token
gh pr list -R sonic-net/sonic-buildimage -A mssonicbld --json additions,assignees,author,baseRefName,body,changedFiles,closed,closedAt,comments,commits,createdAt,deletions,files,headRefName,headRepository,headRepositoryOwner,id,isCrossRepository,isDraft,labels,latestReviews,maintainerCanModify,mergeCommit,mergeStateStatus,mergeable,mergedAt,mergedBy,milestone,number,potentialMergeCommit,projectCards,reactionGroups,reviewDecision,reviewRequests,reviews,state,statusCheckRollup,title,updatedAt,url > prs.log
gh pr list -R sonic-net/sonic-buildimage -A mssonicbld -L 100 -S "label:automerge" --json url,body,title,createdAt,labels,statusCheckRollup > prs.log
cat prs.log | jq
- name: Main
run: |
set -e
# PR merge run per 2 hours
# Other operation run per day.
# Cherry pick PR:
# more than 3 days, comment @author to check
# more than 10 days, stop comment.
# more than 28 days, comment @author PR will be closed
# more than 30 days, close PR
date_3d_ago=$(date --date "3 day ago" -u +"%FT%TZ")
date_10d_ago=$(date --date "10 day ago" -u +"%FT%TZ")
date_28d_ago=$(date --date "28 day ago" -u +"%FT%TZ")
date_30d_ago=$(date --date "30 day ago" -u +"%FT%TZ")
date_now=$(date -u +"%T")
operate=false
[[ "$date_now" < "02:00:00" ]] && operate=true
count=$(cat prs.log | jq 'length')
for ((i=0;i<$count;i++))
do
url=$(cat prs.log | jq -r ".[$i].url")
body=$(cat prs.log | jq -r ".[$i].body")
title=$(cat prs.log | jq -r ".[$i].title")
origin_pr_id=$(echo $title | grep -Eo "\[action\] \[PR:[0-9]*\]" | grep -Eo [0-9]* || true)
created_at=$(cat prs.log | jq -r ".[$i].createdAt")
echo PR: $(($i+1))/$count, URL: $url, createdAt: $created_at, now: $(date -u +"%FT%TZ")
echo PR: $(($i+1))/$count, URL: $url, origin PR: $origin_pr_id, createdAt: $created_at, operate: $operate
[[ "$url" == "" ]] && continue
[[ $created_at > $(date --date "1 hour ago" -u +"%FT%TZ") ]] && continue
# only check automerge PR.
cat prs.log | jq -r ".[$i].labels[].name" | grep automerge || continue
checks=$(cat prs.log | jq ".[$i].statusCheckRollup")
checks_count=$(echo $checks | jq 'length')
echo Checks count: $checks_count
pr_success=true
for ((j=0;j<$checks_count;j++))
do
check=$(echo $checks | jq ".[$j]")
state=$(echo $check | jq -r '.state')
status=$(echo $check | jq -r '.status')
conclusion=$(echo $check | jq -r '.conclusion')
name=$(echo $check | jq -r '.name')
# EasyCLA success flag: state=SUCCESS
# Others success flag: conclusion in SUCCESS,NEUTRAL
# Ignore Azure.sonic-buildimage stage check result. It may be set continueOnError
echo "$name" | grep "Azure.sonic-buildimage (" && continue
# rerun Azure.sonic-buildimage per day
if [[ "$name" == "Azure.sonic-buildimage" ]] && [[ "$conclusion" == "FAILURE" ]];then
completedAt=$(echo $check | jq -r '.completedAt')
[[ "$completedAt" < $(date --date "2 hour ago" -u +"%FT%TZ") ]] && [[ $(date -u +"%T") < "02:00:00" ]] && gh pr comment $url --body "/azp run Azure.sonic-buildimage"
fi
# Ignore Semgrep, it has issues.
[[ "$name" == "Semgrep" ]] && continue
if [[ "$state" == "SUCCESS" ]];then
# check pass
continue
elif [[ "$conclusion" == "SUCCESS" ]] || [[ "$conclusion" == "NEUTRAL" ]];then
# check pass
continue
else
echo "$url Check failed!!!"
echo $check | jq
continue 2
fi
# only check Azure.sonic-buildimage currently
echo "$name" | grep -v "Azure.sonic-buildimage" > /dev/null && continue
[[ "$status" != "COMPLETED" ]] && echo "$name: $status" && continue 2
success=true
( [[ "$conclusion" == "FAILURE" ]] || [[ "$conclusion" == "CANCELLED" ]] ) && success=false && pr_success=false
! $success && echo "FAIL: $name"
done
# rerun Azure.sonic-buildimage per day
! $pr_success && $operate && gh pr comment $url --body "/azp run Azure.sonic-buildimage"
# If auto cherry pick PRs failed, comment in original PR and close cherry pick PR
if [ -n "$origin_pr_id" ] && [[ $created_at < $date_3d_ago ]] && ! $pr_success;then
origin_pr_url=https://github.com/sonic-net/sonic-buildimage/pull/$origin_pr_id
author=$(gh pr view $origin_pr_url --json author | jq .author.login -r)
echo "Original author will check."
$operate && [[ $created_at > $date_10d_ago ]] && gh pr comment $origin_pr_url --body "@$author cherry pick PR didn't pass PR checker. Please check!!!<br>$url"
$operate && [[ $created_at < $date_28d_ago ]] && gh pr comment $origin_pr_url --body "@$author cherry pick PR didn't pass PR checker. Please check!!! Auto cherry pick PR will be closed in 2 days.<br>$url"
$operate && [[ $created_at < $date_30d_ago ]] && echo "$url Closed" && gh pr close $url
fi
! $pr_success && continue
# merge the PR
echo ========Merging PR========
if echo $title | grep "^\[submodule\]";then
@ -77,3 +93,4 @@ jobs:
fi
echo ========Finished PR========
done