Performance improvements

[wzrdtales]

Currently the library is hardly limited by the performance of the adapter and constraints around it.

The reason for this is here

node-oidc-provider/lib/models/client.js

Lines 638 to 645 in 8355886

	const properties = await getAdapter().find(id);
	if (!properties) {
	return undefined;
	}
	const propHash = fingerprint(properties);
	let client = dynamicCache.get(propHash);

The dynamicCache is only used to cache the instance, but it actually goes to a pretty expensive operation of talking to the db first, the creating a fingerprint (actually why not use the "updatedAt" timestamp?) and then get things from the dynamicCache.

There are multiple strategies to make this of better performance:

1. Embrace config desync and use ttl timeout's for cached client configs. (This can be synchronized a little bit by writing the date of ttl to a shared store)
2. Apply late synchronization, by setting the update of the clients record to not become active before time XYZ, which can be in turn synchronized with the local ttl of the cache.

[panva]

3. cache client data before hitting your expensive DB operation in the adapter. ✅ as intended. if you hardly ever change client data, cache them with a memoizer/lru/ttl in the adapter find method.

[wzrdtales]

fair enough, still questionmark around why the expensive fingerprint is needed if there is an updatedAt which can be checked way cheaper than computing a hash?

[wzrdtales]

even if you want to do the fingerprint, wouldn't do that before checking the updatedAt

[panva]

There may be updatedAt in your db model, there isn't one in the provider's tho.

[panva]

And the fingerprint is there to ensure unchanged client adapter metadata doesn't result in the expensive (but less so than the fingerprint) instantiation of a Client.

[wzrdtales]

ok no worries, we will change it in our hardfork, still giving some of our thoughts back to you in the hope we don't diverge everywhere. these changes are necessary though and it is not enough on the adapter level in that case.

as always thank you for your work and time