Classic pub/sub broadcasts to every node on every publish, which does not scale in a cluster and is the reason sharded channels exist. Choosing a channel name that…
Queue depth without throughput says nothing — a hundred jobs is fine at fifty a second and an incident at one a minute. Oldest-waiting-age measures the promise directly…
Recording that a message was handled and doing the work are two writes, and a crash between them is either a duplicate or a silent loss. Claiming outside…
A retry limit lives in the consumer and resets when the consumer restarts, so a message that crashes the process is retried forever. A message that kills the…
A Redis stream consumer group tracks messages delivered and not acknowledged, and that list is where a crashed consumer’s work sits. A pending count that grows without bound…
SKIP LOCKED lets several workers claim different rows from one table without blocking each other, which is what makes a database usable as a queue. This is genuinely…
A published message has consumers you do not control, and its shape is a contract whether or not anybody wrote one down. Publishing both versions during a transition…