Slow queries on larger result sets

[ceit-each]

Hi there,

I am new to this addon after working with emberfire for a good while (looking to move on due to lack of maintenance). Appreciate the work and the addon looks great! However, one thing I've noticed is that when using query with a result set in the 1000s there is a fairly significant difference in the amount of time it takes for both addons to return the documents.

In my simple test, to retrieve 2000 documents using this.store.query, both addons took roughly the following amount of time to return: ember-cloud-firestore-adapter: ~ 18 seconds emberfire: ~ 4-5 seconds

I also would periodically come across the following error:

[2022-01-11T14:25:46.640Z]  @firebase/firestore: Firestore (9.6.2): FirebaseError: [code=resource-exhausted]: Too many outstanding requests.
[2022-01-11T14:25:46.641Z]  @firebase/firestore: Firestore (9.6.2): Using maximum backoff delay to prevent overloading the backend.

I have only briefly looked at the code of the addon, but is the individual fetching of docs in query() -> findQueryRecords() in adapters/cloud-firestore-modular.ts the cause of these issues?

[mikkopaderes]

Yes, there are individual fetches after a query but if you notice, the query uses onSnapshot and does the individual fetching inside of it. Then at the very end, it does a unsubscribe to it.

Because we kept a listener running, the individual fetches will attempt to fetch from the cache making it resolve the request significantly faster.

To be honest, it's been a while since I did this simple benchmarking and I'm not sure if it's still the case or if Firestore has changed that behavior already. Maybe it's time for me to check that out soon.

[mikkopaderes]

I've checked this again, and it does seem to not act as I would've remembered then. I'm gonna try to revisit the implementation to improve the querying.

[ceit-each]

Great, thanks for looking into it!

[mikkopaderes]

Yes, there are individual fetches after a query but if you notice, the query uses onSnapshot and does the individual fetching inside of it. Then at the very end, it does a unsubscribe to it.

Because we kept a listener running, the individual fetches will attempt to fetch from the cache making it resolve the request significantly faster.

To be honest, it's been a while since I did this simple benchmarking and I'm not sure if it's still the case or if Firestore has changed that behavior already. Maybe it's time for me to check that out soon.

After revisiting this again, I'm going to backtrack my previous statement. It does seem like the scenario I mentioned above is still the same. If you already have a query onSnapshot, any document onSnapshot that's part of that query will be fetched from the cache. However, a getDoc will not. Here's an example:

const colRef = collection(db, 'users');
const queryRef = query(colRef, where('field', '==', 'foo'));

onSnapshot(queryRef, (querySnapshot) => {
  // querySnapshot contains /users/12345
});

// If after some time, you do a doc fetch via `onSnapshot` for a document 
// already fetched by `onSnapshot` above, it will fetch from cache and finish the request immediately.
const docRef = doc(db, 'users', '12345');

onSnapshot(docRef, (docSnapshot) => { });

// However, doing this doesn't fetch from cache so it will take a little bit longer to finish this request
getDoc(docRef);

I'm not sure how you did your benchmark in order to cause that output. It would help if you could share a repo. Nevertheless, I'm still keen on pursuing the refactor because looking at my code now, I feel it's so crafty with the supposed optimizations that it's hard to understand what's happening underneath at first glance.

Reference

[mikkopaderes]

237 introduced the refactor