Implement DNS querying cache management API

[Ethan-Arrowood]

What is the problem this feature will solve?

Follow up to #57640 , we should create an API that allows users to specify the DNS querying cache TTL.

With the addition of this API, we should restore the default behavior to match the (sensible) defaults of c-ares (i.e. remove the qcache_max_ttl=0`).

I will contribute this in the coming days.

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

API I have in my head is to add a ttl option to https://nodejs.org/api/dns.html#resolveroptions and to all of the shortcuts too i.e. https://nodejs.org/api/dns.html#dnsresolvehostname-rrtype-callback

Something like dns.resolve('...', { ttl: 100 }, () => {}); or new dns.Resolver({ ttl: 100 });

What alternatives have you considered?

No response

[ronag]

Should we somehow automatically clear the cache if we detect a connection error to a cached IP? Does that make sense?

[Ethan-Arrowood]

I noticed that the shorthands already have a ttl option: https://nodejs.org/api/dns.html#dnsresolve4hostname-options-callback

So this is a little confusing. Should I call the new thing maxTTL or something else?

[Ethan-Arrowood]

@ronag I'm trying to demonstrate the cache timings, but with or without the fix in #57640 I'm seeing DNS caching. I believe this is my system level cache and not the one in Node?

Do you have or anyone in @nodejs/dns know how I can reliably test the c-ares cache only?

This is the naive script I'm using:

import { PerformanceObserver } from 'node:perf_hooks';
import { resolve4 } from 'node:dns/promises';
import { setTimeout } from 'node:timers/promises';

let c = 1;
const obs = new PerformanceObserver((items) => {
	items.getEntries().forEach((item) => {
		console.log(`DNS query ${c++} took ${item.duration}ms`);
	});
});
obs.observe({ entryTypes: ['dns'] });

console.log('Cache Test of github.com');

await resolve4('github.com', { ttl: true })
let result = await resolve4('github.com', { ttl: true });

// Sleep for TTL of dns query 2
await setTimeout((result[0].ttl + 1) * 1000);

await resolve4('github.com', { ttl: true });
await resolve4('github.com', { ttl: true });

I've ran this with a local build of main branch (which contains the qcache_max_ttl=0 fix), but I'm still observing caching happening.

Cache Test of github.com
DNS query 1 took 6.98ms
DNS query 2 took 0.1665840000000003ms
DNS query 3 took 49.486541999998735ms
DNS query 4 took 1.5781669999996666ms

Which query caching disabled, I'd expect for each query to take about the same amount of time. But as my script demonstrates, only if you wait for the returned ttl to expire will it be a new hit.

[metcoder95]

Might be that you are hitting the system level caching; depending what OS you use you might/might-not be able to disable it completely.

For Mac, it is a service that keeps the records in memory. So you might need to kill such service upon request (I guess, not sure if it will work correctly), but c-ares cache should be bypassed with the PR you shared (AFAICT)

[Ethan-Arrowood]

Okay, but how could I add a unit test to Node for this?

[ronag]

I see the same on linux + 22.15.0... which is concerning...

[Ethan-Arrowood]

I spent some time trying to go around my system cache, but I couldn't get anything to show different results.

Based on the c-ares docs I do agree the fix was appropriate and correct, I just cannot test it in anyway 😅

[augjoh]

@Ethan-Arrowood, @ronag : The flag ARES_OPT_QUERY_CACHE is missing in ares_init_options(3) call and the fix in #57640 is incomplete. Perhaps this should be added as well.