Dynamic V8 Heap Resizing for Kubernetes In-Place Scaling (--max-old-space-size-percentage)

What is the problem this feature will solve?

Node.js supports percentage-based heap configuration via --max-old-space-size-percentage. However, this flag currently calculates V8's maximum old space size only once at process startup based on initial system or cgroup memory limits.

Kubernetes supports in-place resource resizing (InPlacePodVerticalScaling), which alters container memory limits dynamically at runtime without restarting the container. Because --max-old-space-size-percentage evaluates only at startup, it cannot react to runtime cgroup changes:

1. Memory Upscaling: If Kubernetes increases a running container's memory limit, Node.js remains restricted to the initial V8 heap limit calculated at startup, failing to utilize the newly available RAM.
2. Memory Downscaling: If Kubernetes decreases the container's memory limit below the initial V8 heap allocation, the container risks being terminated by the Linux Out-Of-Memory (OOM) killer before V8 triggers garbage collection or lowers its heap boundary.

To fully support zero-downtime Kubernetes in-place resizing, --max-old-space-size-percentage needs to dynamically recalculate and apply heap limits when cgroup memory parameters change at runtime.

What is the feature you are proposing to solve the problem?

Extend --max-old-space-size-percentage to actively monitor cgroup memory limit changes and adjust the V8 heap size dynamically during process execution.

Example Workflow

# Enable percentage-based allocation
node --max-old-space-size-percentage=80 app.js

1. Initialization: Container starts with a 2 GiB memory limit. Node.js evaluates --max-old-space-size-percentage=80 and sets V8 max old space to ~1.6 GiB.
2. Kubernetes Event: Kubernetes resizes the running pod in-place from 2 GiB to 4 GiB memory limit.
3. Event Detection: Node.js detects the cgroup change event.
4. Runtime Update: Node.js recalculates 80% of 4 GiB (~3.2 GiB) and adjusts V8's isolate limits dynamically while the application continues processing traffic.

What alternatives have you considered?

No response

Begin with the implementation of --max-old-space-size-percentage and the cgroup memory parameters it reads, then inspect how V8 isolate limits can be updated at runtime. Validate the design against Kubernetes InPlacePodVerticalScaling and both memory upscaling and downscaling; done means the percentage is reapplied after runtime limit changes without a restart.