Trace ingestion stopped for every tenant at once

How to confirm it

• How deep is the backlog?

  redis-cli -h HOST LLEN bull:ingestion-queue:wait && redis-cli -h HOST LLEN bull:ingestion-queue:active

  A growing wait list with a static active count means workers have stopped consuming, not that they are slow.

• What is the store actually doing?

  redis-cli -h HOST --stat

  Watch blocked_clients and ops/sec together. Blocked clients climbing while ops flatline is contention, not load.

• Find the noisy producer

  redis-cli -h HOST CLIENT LIST | awk '{print $2, $NF}' | sort | uniq -c | sort -rn | head

  Identifies which client is issuing the most commands — usually one tenant is responsible for the burst.

• Drain the poisoned queue

  redis-cli -h HOST --scan --pattern 'bull:ingestion-queue*' | xargs -n 100 redis-cli -h HOST DEL

  Destructive: this discards queued work. Only after you have accepted the data loss and captured what was in flight.