Closed mapellidario closed 1 week ago
Jenkins results:
Dario, I can't possibly review the details here. But it looks that it is orthogonal to other things, so you can merge fearlessly. But I still do not understand why we need a different container image. What's wrong with the TW one ? Does pip install pandas
conflict with other things ?
What's wrong with the TW one
I had a chat with wa, and for the time being we prefer to keep the image for monitoring separate from the TW one
Jenkins results:
Jenkins results:
@novicecpp I implemented your suggestions, thanks for the review :)
Jenkins results:
Fixes #8504
status
tested:
details
I removed the old docker images, based on pure python and rpm taskworker.
I migrated the same logic to a new docker image inside
monit_pypi
directory.I added a stage in the pipeline that builds the new monit docker image every time we build taskworker and crabserver images. it takes only 2min, but if this annoys people than we can add some logic to skip this build for most cases.
Every time we build a new monit image, it is tagged with
registry.cern.ch/cmscrab/crabtaskworker:${IMAGE_TAG}.monit
and pushed.I added a new tag retention policy in harbor [4] so that we keep the latest 5 images that match
v3.*.monit
.when the image tag matches
v3.*.*
(or better,, when the pipeline satisfies the rule.default_rules["release"]
), the new monit image is also tagged withregistry.cern.ch/cmscrab/crabtaskworker:v3.latest.monit
and pushed.open questions
I propose to change the crontabs to always use
v3.latest.monit
, so that we can forget about deploying the new docker images for the monitoring scripts, and embrace "continuous delivery" at its fullest :)If we do not like to do this for production monitoring, I sugggest to at least do it with test monitoring, the one that currently runs in
crab-dev-tw04
. This can be a nice exercise for Vijay.[1]
[2]
[3]
https://monit-opensearch.cern.ch/dashboards/goto/8ec71c78057c6ea3150334053ae136ee
[4]