> ## Documentation Index
> Fetch the complete documentation index at: https://infercrane.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Autoscaling

> Scale replicas with bounded decisions, durable evidence, and generation-safe drain.

# Autoscaling

Elastic deployments use bounded queue-aware scaling between explicit minimum and maximum replicas. Persisted vLLM running/waiting signals, consecutive-interval thresholds, and cooldowns produce an auditable scaling decision. Scale-up creates durable replica intents. Scale-down withdraws the worker from the matching router generation, drains it, and then terminates it.

`infercrane explain scaling DEPLOYMENT` returns the latest action, old/new capacity, reason, signal snapshot, and timestamp. If evidence is insufficient or cooldown prevents a change, the persisted no-op decision explains why.

Serverless deployments delegate zero-to-N worker scheduling to the registered provider-native
backend and retain one logical endpoint. InferCrane does not implement a GPU serverless scheduler.
RunPod supplies the first registered native Serverless implementation.