Add a key to any concurrency limit, like event.accountId, and runs are served in turn across keys instead of first in, first out.
The dashboard shows queue depth and wait time per key, so you can spot a noisy tenant before customers do.
Existing limits without a key behave exactly as before.