remove CRAB Server external_config

[belforte]

CRABServer pulls from gitlab the external config https://gitlab.cern.ch/crab3/CRAB3ServerConfig/-/blob/master/cmsweb-rest-config.json

In a way I'd like to fully get rid of this "external config" things and those obscure "modes".

The only customization that we have now via 'mode is to use a different S3 bucket, and it would be much more clear if we put that in CRABServer's config.py in CMSKubernetes.

Currently we

• disguise the S3 bucket name as an URL
• put in a JSON file which the REST fetches every 30 minutes
• so that the Client can retrieved it at crab submit time and pass it to the REST again via the DB
• so that the REST knows which bucket to use

Sounds pretty damn stupid.

And I fear that if I am not here, it would take you a long time to figure this out.

Plan

• change server config.py to support new code by adding following two lines:

  • data.s3_bucket = 'crabcache_dev' or crabcache_preprod or crabcache_prod as appropriate for each cluster
  • data.compatibleClientVersions = ["v3", "development"]
  • in test and preprod update CRABServer config.py for test and preprod CMSKubernetes#1657
  • in prod

• change REST ( Stop using external config part1 #9180 ) to:

  • pick bucket from config (alredy picks delegate-dn from there), save it in DB and ignore cacheurl HTTP param in RESTUserWorkflow.py so that it does not matter whether client sends it or not.
  • remove banned-out-destinations which we never used
  • remove from TW as well (unused code anyhow) remove bannedOutDestination from TW #9176
  • move compatible version to CRABServer/.../config.py

• deploy new REST

  • in preprod (testbed)
  • in prod

• change config.py and cmsweb-rest-config.json to eliminate now-useles things (see description in Stop using external config part1 #9180)

• change client not to send useless cacheurl and once in production remove unused cacheSSL and filecacheurl variables/args CRABClient#5411

  • remove cacheurl submit argument for submit from RESTUserWorkflow.py (where it is now optional)

• move handling of "htcondorPool" and htcondorScheddsLink (i.e. list of enabled schedulers) to TW and manage them via puppet

• deploy new TW, then

  • remove them for REST code "htcondorPool" and htcondorScheddsLink (
  • remove all references to "external config" everywhere

And finally we can:

• get rid of @conn_handler() everywhere in CRABServer repo

Some of those actions can be pursued independently. I will open ad-hoc issues and possibly hand to DevOps