Memory leak in `destr` package, Nuxt 3 production usage

[mrauhu]

Hello, @pi0.

I found a memory leak, that affects Nuxt 3 framework, because the destr package is used as default parseResponse option in the ohmyfetch, that used in the Nuxt 3 useFetch() function as $fetch.

Memory leak

Flamegraph

Table

Memory usage

Temporary fix for Nuxt 3

1. Create a local TypeScript file (lib/use-fetch.ts) with copy of useFetch() function.

2. Add custom parseResponse() function:

   import type { FetchOptions } from 'ohmyfetch';
   
   /**
    * Parse response
    * @description Replace of `destr()` function to prevent memory leak
    * @param responseText
    */
   const parseResponse: FetchOptions['parseResponse'] = (responseText) => {
     // Send empty string if zero length
     if (responseText.length === 0) {
       return '';
     }
   
     try {
       return JSON.parse(responseText);
     } catch (_e) {
       return responseText;
     }
   };
   

3. Replace _fetchOptions constant in the copy of useFetch() function:

     const _fetchOptions = {
       ...opts,
       parseResponse,
       cache: typeof opts.cache === 'boolean' ? undefined : opts.cache
     }
   

4. Use patched useFetch() function instead of build-in.

Because, right now there is no way to provide custom parseResponse() function as option of the useFetch() without error.

Best wishes, Sergey.

[pi0]

Hi @mrauhu thanks for making the issue and providing a workaround but are you sure the memory leak originates from destr and memory is not back after a while with GC? Can you please provide a minimal reproduction? 🙏🏼 (without nuxt but a direct Node.js repo using destr)

[pi0]

Any update on this?

[mrauhu]

Any update on this?

@pi0 unfortunately, I can't create a reproduction in the isolated environment only for destr package.

Changes after replacing destr() to JSON.parse() function for the production server, a week of observations:

$fetch parseResponse option	Max memory usage, MB
destr()	509
JSON.parse()	112

Memory usage

Flamegraph

Table

[pi0]

Higher memory usage is likely the overhead of regex checkers over a heavy payload. What happens after max memory usage? Do you auto restart? Also what is prod Node.js version?

[mrauhu]

Higher memory usage is likely the overhead of regex checkers over a heavy payload.

API responses is nested tree JSON documents, 50—100 KB size.

What happens after max memory usage?

Restart of a container.

Do you auto restart?

Yes.

Also what is prod Node.js version?

Node.js 16.14.2, a Docker image based on the https://github.com/astefanutti/scratch-node.

[dargmuesli]

@mrauhu with which application did you examine the memory leak with?

[mrauhu]

@dargmuesli it's part of the Datadog APM.

The dd-trace-js package was used for tracing. Under the hood it's using the @datadog/pprof package, fork of pprof.

The Datadog Continuous Profiler used for visualization.

[mcstover]

@mrauhu Might I ask how you implemented the dd-trace-js in your application? Did you use a module or a server plugin? Thanks!

[pi0]

Hi. Kindly this issue still needs a minimal reproduction. Please ping to reopen if you still believe there is a possible memory leak in destr itself.