Honor forwarder_stop_wait_for_inflight during shutdown. #2560

Background

The core agent's forwarder_stop_wait_for_inflight setting selects how forwarder workers stop. Its default value, false, cancels in-flight HTTP requests; true waits for them to complete. The wait remains bounded by forwarder_stop_timeout.

ADP does not currently read this setting. Closing its forwarder handle stops new dispatch, waits for active request tasks to finish, and then flushes pending transactions to the retry queue. The topology-wide shutdown timeout bounds this drain.

• Core agent shutdown selection
• ADP forwarder shutdown
• ADP in-flight request drain

Problem

ADP always behaves like forwarder_stop_wait_for_inflight: true. An operator cannot select the core agent's immediate-cancellation behavior by setting the key to false. This can make ADP wait for active requests until they finish or until the topology-wide shutdown timeout expires.

Proposal

Honor forwarder_stop_wait_for_inflight in the Datadog forwarder shutdown path:

• Preserve the current bounded drain when the setting is true.
• When it is false, cancel active HTTP requests without waiting for normal completion.
• Preserve or requeue interrupted transactions where possible, and test both shutdown modes under active requests.

The implementation should continue to respect ADP's topology-wide shutdown timeout.

Alternatives and Discussion

ADP could retain its unconditional bounded-drain policy and document the difference. Supporting the setting instead gives operators the same shutdown-policy choice as the core agent.