What triage is holding right now, and whether it is keeping up. The top half is live —
each running replica reports its own in-memory queue and they are added together. The
bottom half is measured from finished work, so it survives a restart but cannot see
anything still waiting.
Replicas
The queue lives in each container's memory, so no single process knows the whole
picture. Every replica publishes its own depth and this table sums them. Normally
there is one; the app is allowed to scale to three under load, and each one gets a row.
Flow — the last few hours
Derived from tickets that finished, which is all storage can offer: nothing is written
until a run completes, so pending work leaves no trace here. Falling behind shows up as
lag climbing while throughput flattens.