Bug: `axios.waiting` grows when the max entries is set + etag handling

[bukowskiadam]

What happened?

Hey Arthur,

Recently we've migrated our code to axios and we've used your package to handle caching. Thank you for your work here ❤️ After the migration, we've noticed that memory consumption of our app presents a "lovely" toothsaw pattern 😬 I've spent quite a lot of time trying to figure out what's wrong.

Here is the recap - sorry for it being long.

Etag / last-modified

This is not a bug, but that caused us to store many entries forever.

Every resource with an etag header (or last-modified) in the response (the default express app adds it), is kept forever, and there is no simple way to disable it (or I haven't found one).

I've mitigated it with the following code that removes those headers and assign default stale ttl:

import axios from 'axios';
import {
  setupCache,
  buildMemoryStorage,
  defaultHeaderInterpreter,
} from 'axios-cache-interceptor';

const DEFAULT_TTL_MS = 5 * 60 * 1000;
const DEFAULT_STALE_MS = 60 * 1000;

const CLONE_DATA = false;
const CLEANUP_INTERVAL_MS = 5 * 60 * 1000;
// set to unlimited to avoid the hanging promise in the `waiting` object
const MAX_ENTRIES = undefined;

const axiosCache = setupCache(axiosCacheInstance, {
  ttl: DEFAULT_TTL_MS,
  storage: buildMemoryStorage(CLONE_DATA, CLEANUP_INTERVAL_MS, MAX_ENTRIES),
  headerInterpreter: (headers) => {
    // remove unwanted headers that force the cache entry to be stalled/cached forever
    headers && delete headers.etag;
    headers && delete headers['last-modified'];

    let ret = defaultHeaderInterpreter(headers);

    if (typeof ret === 'object' && 'cache' in ret) {
      // we have the cache and stale time or possibly null. make sure it's our default value
      ret.stale = DEFAULT_STALE_MS;
    } else if (typeof ret === 'number') {
      // we have the ttl number, but we have to add the stale time
      ret = {
        cache: ret,
        stale: DEFAULT_STALE_MS,
      };
    }

    if (ret === 'not enough headers') {
      // by default it would use the DEFAULT_TTL_MS, but `stale` would be `undefined`
      // in order to override `stale` we need to return an object
      ret = {
        cache: DEFAULT_TTL_MS,
        stale: DEFAULT_STALE_MS,
      };
    }

    return ret;
  },
});

waiting

To mitigate the problem with stalled entries kept forever, we've tried using MAX_ENTRIES in the memory storage, but that didn't improve the situation but the problem was different.

Where we have many parallel requests fired, and we exceed the max entries limit, the memory adapter removes some of them. But they are kept in the axios.waiting object. After the request is resolved they are not removed from there. Probably until you try to hit the same request.id again (the same url, method, query).

I believe the problem comes from this part of the code: https://github.com/arthurfiorette/axios-cache-interceptor/blob/f0ba5ad92ef333ba146a4378780a0b27262be1c0/src/interceptors/response.ts#L93-L108

Because when you're removing the entry from the storage and it's not there, the returned value is

{ state: 'empty' }

so this if is true, so we end the response interceptor, and we do not remove / resolve other waiting requests.

In our very specific case, the URL passed to Axios was sliced from a very long document, which caused Node to keep the reference to the original string. (slice / substring / regex match, does not create the copy of a string, but points to the fragment of the original string).

proof of concept

Here is my repo with test cases describing the two issues we had: https://github.com/bukowskiadam/axios-cache-interceptor-memory-issue

I hope it gives some working example so you can validate the problems.

What's next

(In other words: what is this issue about)

1. Fixing axios.waiting memory leak
2. Maybe provide an easy way like etag: false to disable handling ETag headers.

Best regards, Adam

axios-cache-interceptor version

v1.5.2

Node / Browser Version

Node v20.12.2

Axios Version

v1.6.8

What storage is being used

Memory Storage

Relevant debugging log output

Output too long - see the linked project

[arthurfiorette]

https://axios-cache-interceptor.js.org/config/request-specifics#cache-etag disallows etag

[bukowskiadam]

Hi, as you described in the docs (and I checked the code), the only valid options are true | string. false does nothing. I've checked that here: https://github.com/bukowskiadam/axios-cache-interceptor-memory-issue/blob/e7286464f8aa036fc6eace6c35baee89824dd6b0/test/etag.js#L29

Imo this code: https://github.com/arthurfiorette/axios-cache-interceptor/blob/f0ba5ad92ef333ba146a4378780a0b27262be1c0/src/interceptors/response.ts#L135-L137 allows only to override etag, but not to delete it 🤷🏻

[arthurfiorette]

Hey @bukowskiadam, you're right!

I do not have enough spare time to work in this project right now... Anyone up to a PR?

[gotdibbs]

Just a quick note here that we've also run into this memory leak in production and it caused us to attempt to reshape how we scale our services that were planning on using this library. We ended up having to roll back to axios-cache-adapter to balance effort/value of the swap. If/when we have time to revisit will come back to this issue.