Increase heap memory for jsii engine

[bverhoeve]

🚀 Feature Request

When using JSII for small AWS CDK applications, the JSII kernel will normally not run out of heap memory. However, for larger apps (>800 resources, multiple stacks), depending on the complexity, you can encounter the following error: FATAL ERROR: Ineffective mark-compacts near heap limit Allocation failed - JavaScript heap out of memory.

Is there already a way to increase the Javascript heap memory beyond the default value?

Affected Languages

• TypeScript or Javascript
• Python
• Java
• .NET (C#, F#, ...)

General Information

• JSII Version: v0.15
• Platform: UNIX

• I may be able to implement this feature request

• This feature might incur a breaking change

Description

Add a setting in the cdk.json file which specifies the maximum allowed heap memory that jsii can allocate. Set the maximum heap memory then to this setting.

Proposed Solution

[RomainMuller]

I don't think we can host this setting in cdk.json (because JSII can be used in contexts other than cdk).

Ability to customize the node runtime is however a very interesting idea and we should dig into this further. We should also consider what other aspects of the node runtime we could/should allow users to control.

[RomainMuller]

Note that in the meantime, you can pass additional arguments to the node process by setting the NODE_OPTIONS environment variable. In particular, controlling the max heap size can be achieved using:

# Sets the maximum heap to 4GiB
NODE_OPTIONS='--max-old-space-size=4096'

[bverhoeve]

Does jsii always use the system node installation to run? Or does it use another kernel?

If the system node is the case, the NODE_OPTIONS is quite nice already as a workaround :)

[bverhoeve]

@RomainMuller the NODE_OPTIONS variable works for increasing the max heap size, but unfortunately not the call stack size. This can be changed by using node with the --stack_size option set to a certain value. Unfortunately, the --stack_size option can't be set yet using NODE_OPTIONS.

[RomainMuller]

Ah - an alternate way to solve this until we actually support acting on those settings via configuration is to create a wrapper around node that is ahead of the real node in your $PATH, something along the lines of (warning: I haven't tested it):

#!/bin/bash

/usr/local/bin/node --stack_size=4096 $@

[ansgarm]

We recently encountered this as well in the CDK for Terraform when generating provider bindings for the AWS Terraform Provider for Go.

Here is a small example which can be used to reproduce the issue or to do some profiling.
For now we went the route of setting NODE_OPTIONS on process.env so it is made available to the child process for jsii-pakmak which is spawned by jsii-srcmak (which we use). While this seems like an okayish fix, I'm still a bit confused why the memory usage is that heavy.

Do you have any pointers where one might start digging, @RomainMuller? We're experiencing this particularly with Go (to this 7 GB extent 🏋️😄)

[github-actions]

This issue has not received any attention in 2 years. If you want to keep this issue open, please leave a comment below and auto-close will be canceled.