Some ideas from our previous meeting:

• Measure the cache misses, https://phabricator.wikimedia.org/T427752#12037841
• The size of the queue
• One thing we could consider adding metrics for is the rsync timeout case: https://phabricator.wikimedia.org/T427752#12037841 "rsync --contimeout 3 || :" - “--contimeout 3” means rsync gives up if it can't establish the connection to the castor daemon within 3 seconds. the "|| :" is a no-op fallback (":" is bash's true builtin), so a failed/timed-out rsync is silently masked - the job exits 0, /cache is left empty, and the build proceeds straight to a cold npm ci. Here’s the line: https://gerrit.wikimedia.org/g/integration/config/+/refs/heads/master/jjb/castor-load-sync.bash#18

Next steps:

• Create an Epic: https://phabricator.wikimedia.org/T432684
• Create a dashboard to track the metric