search

firebase / flutterfire

🔥 A collection of Firebase plugins for Flutter apps.
https://firebase.google.com/docs/flutter/setup
BSD 3-Clause "New" or "Revised" License
8.62k stars 3.95k forks source link

[cloud_firestore] streaming query inefficient #837

Closed kroikie closed 3 years ago

kroikie commented 4 years ago

The firestore plugin serializes the entire document list for each query update. Wouldn't it be more efficient to only serializes the deltas and apply them in the dart code? Especially for large queries.

kroikie commented 4 years ago

@ribrdb

The issue at https://github.com/flutter/flutter/issues/17247 has been closed and moved here. Future collaboration on this issue will be done here.

iapicca commented 4 years ago

Hi @ribrdb If you are still experiencing this issue with the latest version of Flutter and FlutterFire plugin can you please provide your updated flutter doctor -v and your and your flutter run --verbose/flutter build --verbose ? Also, to better address the issue, would be helpful if you could post a self contained app on github or the steps to reproduce it. Thank you

ribrdb commented 4 years ago

I don't have an app, I was reading your source code to see if I want to use it. You still do the same thing of serializing every document for every QuerySnapshot even if only a single document is added.

kroikie commented 4 years ago

@ribrdb Thanks for bringing this up, we could indeed reduce the amount of serialization however that would also require significant book keeping and additional memory usage for caching documents on the Dart side. We would need to find the right balance, we are open to suggestions and PRs if you have a solution in mind.

animeshjain commented 4 years ago

This is indeed a problem in non-trivial realtime kind of use cases. For instance, if the result set is more then a few documents (say near to 50-60 documents) and a realtime listener is setup to keep a track of changes in this result set, there is a huge hit on the CPU and a bunch of frames start dropping.

I tried to do a very unscientific benchmark to compare Firebase Realtime DB to Cloud Firestore. If you look at the Firestore flame graph, most of the time is spent on decodeMethodCall doing a bunch of deserializations firestore_flame_graph

On the other hand, realtime database does not break a sweat realtimedb_flame_graph

These were rough benchmarks by me clicking around and generating events in a dummy app at roughly the same speed. But I think the difference is evident and it can be clearly felt in skipped frames in the UI in case of Firestore.

To be honest, I am a bit surprised that this is not higher priority. I feel a lot of users who complain about Firebase (Firestore) being slow might be getting affected by something like this unknowingly too.

Salakar commented 3 years ago

Given the age of this issue and that Firestore has had a major rework since and has had a few improvements around serialisation and use of threading to improve performance; I'm going to close this issue in favour of trying the latest release.