[Jonathan-Eid]

Sets priorityClassName: catchup-infra on the redis, preload and job monitor pods so the scheduler places them before the 1024 workers.

• build 1968 failed "job monitor not reachable": per the API audit log the scheduler didn't attempt those 3 pods until 15:36, 9 min after creation, while it cycled through workers waiting on Karpenter nodes. The untainted nodes had room. Redis bound at 15:36, the driver's 10-min /status timeout fired at 15:38:07
• the class (value 1000, preemptionPolicy: Never) comes from stellar/kube#5671. Merge that first and confirm it exists on ssc-eks and ssc-test, or every pod using it is rejected
• workers keep default priority

Testing:

• helm template: the 3 infra pods get catchup-infra, workers none
• ssc-test run (class applied there): redis/preload/monitor at priority 1000 Never, workers 0, all scheduled

🤖 Generated with Claude Code