const ADMIT_WAIT: Duration;Expand description
How long an export waits for room in the queue before it is shed.
Under the OTLP exporter timeout, which is 10s in every SDK that follows the
spec’s default, so a waiter is answered by this node rather than abandoned
by its client — an abandoned request is the one case where the work is paid
twice and nobody is told. Over max_block_age, which is 2s, so a queue that
is full only because a flush is in flight drains within the wait instead of
shedding around it. Five seconds sits in the middle of that range.
This is the tail bound, not a target: at the operating point nothing waits at all. It matters when a sender is faster than the disk, and there the choice is between a slow ack and a 503 that costs the sender a retry and this node the decode it already did.