Layer proxy

Use the layer proxy, Mode A, only when target handler code cannot change. It is experimental, runtime-specific, and more fragile than adapters.

Add the layer

Deploy the bootstrap stack and pass the exported layer ARN into your function:

KhoneLayerArm64Arn="$(aws cloudformation list-exports \
  --query 'Exports[?Name==`KhoneLayerArm64Arn`].Value | [0]' \
  --output text)"

Configure the exec wrapper

Set these variables in the function's Environment.Variables block, not in a shell:

Environment:
  Variables:
    AWS_LAMBDA_EXEC_WRAPPER: /opt/khone/exec-wrapper.sh
    KHONE_MAX_CONCURRENCY: 4

Set KHONE_MAX_CONCURRENCY to the route maxBatchSize as a starting point.

Runtime notes

• Node is the most tested layer proxy runtime.
• The exec wrapper sets AWS_LAMBDA_NODEJS_USE_ALTERNATIVE_CLIENT_1=true when multi-concurrency is enabled, unless you override it.
• Python 3.14 concurrency remains experimental and uses a telemetry file descriptor workaround.
• User-code streaming is not supported in the layer proxy.
• Duplicate request ids in a batch are rejected.

Prefer Adapters when target code can change.