ResourcePressurePolicyProtocol
Defined in the protocols module.AI Agent Protocol for memory/resource-aware admission decisions. Pure, import-free decision contract consumed by the wrapper’s admission gate. The wrapper samples its own resource usage on a lightweight cadence and hands the policy a :class:
ResourceSample; the policy returns an
:class:AdmissionDecision — ADMIT below the soft threshold, QUEUE
to apply backpressure above it, and REJECT above the hard threshold so
the process sheds load before the OOM killer fires. Sampling and
enforcement (the asyncio.Semaphore ceiling and bounded wait queue)
live in the wrapper; this keeps the decision provable in isolation,
symmetric with :class:GatewayConcurrencyPolicyProtocol.
A config-driven default (:class:MemoryPressurePolicy) is provided for
the common “soft/hard RSS threshold” case.
Methods
evaluate()
Return an :class:
AdmissionDecision for the supplied sample.Source
View on GitHub
praisonaiagents/gateway/protocols.py at line 3358