A Redis stream consumer group tracks messages delivered and not acknowledged, and that list is where a crashed consumer’s work sits.
> XPENDING orders workers
1) (integer) 412 # pending count
2) "1623661262-0" # oldest
4) 1) 1) "worker-3"
2) "412" # all of them, on one worker
> XAUTOCLAIM orders workers worker-1 60000 0 COUNT 100
# reassign anything idle for over a minute
A pending count that grows without bound means messages are being delivered and never acknowledged, which is a crashed or wedged consumer rather than a throughput problem. XAUTOCLAIM in 6.2 replaces the awkward scan-and-claim dance that XPENDING plus XCLAIM required. A claimed message that keeps failing needs a delivery-count check and a dead letter path, or it is reclaimed forever.