You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Proposal (Firestore): better default behaviour for the collectionData and docData functions (for the underlying SDK's handling of optimistic updates)
#84
Make a technicallybreaking change to the collectionData and docData functions to always have includeMetadataChanges set to false when listening for snapshots.
Background
The underlying Firebase SDK for Firestore performs optimistic updates on the client, with a hasPendingWrites metadata flag available on snapshots to tell you when a server-side update has been committed (only if you've called the underlying onSnapshot with includeMetadataChanges: true, which is what rxfire currently does). From the docs:
Local writes in your app will invoke snapshot listeners immediately. This is because of an important feature called "latency compensation." When you perform a write, your listeners will be notified with the new data before the data is sent to the backend.
Retrieved documents have a metadata.hasPendingWrites property that indicates whether the document has local changes that haven't been written to the backend yet.
This behaviour results in a very common "why am I seeing two updates trigger for my snapshot?" (though only when includeMetadataChanges is set to true). The idea is that you could filter out and choose which of the updates to respond to in your implementation (or something more elaborate, e.g. to handle offline scenarios).
Note: it's not clear (at least, to me) what happens if a server-side error occurs (e.g. security rules fail) but I'll assume for now that a failure always results in a new snapshot being emitted.
Problem: double updates emitted when using the rxfire collectionData and docData functions with no way to filter out
In the current rxfire implementation, the collectionData and docData functions end up setting includeMetadataChanges: true when setting up the snapshot listener. So if an update is made to any document in the collection (or to the single document) being observed then two whole emissions of the collection (or document) are made in the observable – one for the client-side optimistic update and another once the server-side update has committed. This has been reported previously in: #50.
This would normally be okay except the emitted value of both functions do not have the hasPendingWrites metadata flag, so it's impossible to choose when to filter out one of the emitted results. And since there's no way to explicitly set includeMetadataChanges to false when listening to the snapshot we have to live with this behaviour (with some folks resorting to workarounds).
Note: the same behviour is true of the collection and data functions, but in those cases we do have access to the hasPendingWrites flag so we have control over which we filter out.
Proposed solution
An attempt was made to allow configuring the includeMetadataChanges option in #12.
Instead of the approach taken there, this proposal suggests:
Changing the default behaviour of the collectionData and docData rxfire functions to always have includeMetadataChanges set to false when setting up the snapshot listener.
This is technically a breaking change but may not have a big impact as the net effect is that the same results will be emitted once, rather than twice. So it's only a "breaking" change if folks were relying on this double update behaviour (as opposed to treating it like a bug).
Rationale: these methods are essentially convenience methods on top of the underlying SDK, which simplify usage (e.g. setting the ID field). So it could make sense to also abstract away the metadata change and use the default behaviour of the underlying SDK.
And as a result of this implementation change (see below), allowing the collection and doc rxfire functions to be passed in a value for includeMetadataChanges, defaulting to true (as it is now).
Implementation details
The current implementation chain goes like (shown for collection, but the same for doc):
Hi @davideast 👋🏽 I expect you're busy with a million things right now — do you think there's a possibility we could pick up this change and works towards a new (breaking change) version of rxfire with improvements like this? Thanks!
tl;dr
Make a technically breaking change to the
collectionData
anddocData
functions to always haveincludeMetadataChanges
set tofalse
when listening for snapshots.Background
The underlying Firebase SDK for Firestore performs optimistic updates on the client, with a
hasPendingWrites
metadata flag available on snapshots to tell you when a server-side update has been committed (only if you've called the underlyingonSnapshot
withincludeMetadataChanges: true
, which is what rxfire currently does). From the docs:This behaviour results in a very common "why am I seeing two updates trigger for my snapshot?" (though only when
includeMetadataChanges
is set totrue
). The idea is that you could filter out and choose which of the updates to respond to in your implementation (or something more elaborate, e.g. to handle offline scenarios).Note: it's not clear (at least, to me) what happens if a server-side error occurs (e.g. security rules fail) but I'll assume for now that a failure always results in a new snapshot being emitted.
Problem: double updates emitted when using the rxfire
collectionData
anddocData
functions with no way to filter outIn the current rxfire implementation, the
collectionData
anddocData
functions end up settingincludeMetadataChanges: true
when setting up the snapshot listener. So if an update is made to any document in the collection (or to the single document) being observed then two whole emissions of the collection (or document) are made in the observable – one for the client-side optimistic update and another once the server-side update has committed. This has been reported previously in: #50.This would normally be okay except the emitted value of both functions do not have the
hasPendingWrites
metadata flag, so it's impossible to choose when to filter out one of the emitted results. And since there's no way to explicitly setincludeMetadataChanges
tofalse
when listening to the snapshot we have to live with this behaviour (with some folks resorting to workarounds).Note: the same behviour is true of the
collection
anddata
functions, but in those cases we do have access to thehasPendingWrites
flag so we have control over which we filter out.Proposed solution
An attempt was made to allow configuring the
includeMetadataChanges
option in #12.Instead of the approach taken there, this proposal suggests:
collectionData
anddocData
rxfire functions to always haveincludeMetadataChanges
set tofalse
when setting up the snapshot listener.collection
anddoc
rxfire functions to be passed in a value forincludeMetadataChanges
, defaulting totrue
(as it is now).Implementation details
The current implementation chain goes like (shown for collection, but the same for doc):
Proposed solution:
/cc @davideast
The text was updated successfully, but these errors were encountered: