firestore: add experimental option "sendWriteRequestsDelayMs" to enable write request batching in fewer http requests

[dconeybe]

This is a proposed fix to reduce the number of HTTP POST requests sent from the browser when the write pipeline is being replenished after each write acknowledgement. This is done by batching writes into a single HTTP request that get added into the pipeline due to fast write acknowledgements within a 250ms window. This fix is subject to change or being reverted entirely. It needs thorough review and testing.

Fixes: #5971

[changeset-bot]

⚠️ No Changeset found

Latest commit: 74d6f2e

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[github-actions]

Vertex AI Mock Responses Check ⚠️

A newer major version of the mock responses for Vertex AI unit tests is available. update_vertexai_responses.sh should be updated to clone the latest version of the responses: v8.0

[google-oss-bot]

Size Report ¹

Affected Products

• @firebase/firestore

  Type	Base (1df3d26)	Merge (8afab38)	Diff
  browser	382 kB	383 kB	+1.67 kB (+0.4%)
  main	591 kB	594 kB	+2.88 kB (+0.5%)
  module	382 kB	383 kB	+1.67 kB (+0.4%)
  react-native	382 kB	384 kB	+1.67 kB (+0.4%)

• @firebase/firestore-lite

  Type	Base (1df3d26)	Merge (8afab38)	Diff
  browser	113 kB	114 kB	+968 B (+0.9%)
  main	155 kB	157 kB	+1.34 kB (+0.9%)
  module	113 kB	114 kB	+968 B (+0.9%)
  react-native	113 kB	114 kB	+968 B (+0.9%)

• bundle

  15 size changes

  Type	Base (1df3d26)	Merge (8afab38)	Diff
  firestore (CSI Auto Indexing Disable and Delete)	272 kB	274 kB	+1.41 kB (+0.5%)
  firestore (CSI Auto Indexing Enable)	272 kB	274 kB	+1.41 kB (+0.5%)
  firestore (Persistence)	304 kB	305 kB	+1.41 kB (+0.5%)
  firestore (Query Cursors)	250 kB	251 kB	+1.03 kB (+0.4%)
  firestore (Query)	248 kB	249 kB	+1.03 kB (+0.4%)
  firestore (Read data once)	236 kB	237 kB	+1.03 kB (+0.4%)
  firestore (Read Write w Persistence)	328 kB	330 kB	+1.41 kB (+0.4%)
  firestore (Realtime updates)	238 kB	239 kB	+1.03 kB (+0.4%)
  firestore (Transaction)	215 kB	216 kB	+1.03 kB (+0.5%)
  firestore (Write data)	214 kB	216 kB	+1.41 kB (+0.7%)
  firestore-lite (Query Cursors)	104 kB	105 kB	+844 B (+0.8%)
  firestore-lite (Query)	99.9 kB	101 kB	+844 B (+0.8%)
  firestore-lite (Read data once)	75.4 kB	76.2 kB	+844 B (+1.1%)
  firestore-lite (Transaction)	101 kB	101 kB	+844 B (+0.8%)
  firestore-lite (Write data)	85.0 kB	85.8 kB	+844 B (+1.0%)

• firebase

  Type	Base (1df3d26)	Merge (8afab38)	Diff
  firebase-compat.js	794 kB	796 kB	+1.29 kB (+0.2%)
  firebase-firestore-compat.js	340 kB	341 kB	+1.29 kB (+0.4%)
  firebase-firestore-lite.js	131 kB	132 kB	+962 B (+0.7%)
  firebase-firestore.js	440 kB	442 kB	+1.67 kB (+0.4%)

Test Logs

• Base (1df3d26): https://github.com/firebase/firebase-js-sdk/actions/runs/14272240626
• Merge (8afab38): https://github.com/firebase/firebase-js-sdk/actions/runs/14318862174

1. https://storage.googleapis.com/firebase-sdk-metric-reports/qhRIt5a53G.html

[google-oss-bot]

Size Analysis Report ¹

This report is too large (297,347 characters) to be displayed here in a GitHub comment. Please use the below link to see the full report on Google Cloud Storage.

Test Logs

• Base (1df3d26): https://github.com/firebase/firebase-js-sdk/actions/runs/14272240626
• Merge (8afab38): https://github.com/firebase/firebase-js-sdk/actions/runs/14318862174

1. https://storage.googleapis.com/firebase-sdk-metric-reports/yuHB84Eq6d.html

[dconeybe]

I did a quick performance test to see if there were any overall improvements to batching the write requests into fewer physical http requests. tl;dr there were only improvements noted on very slow ("3G") networks; the benefit on high-quality networks is negligible.

To test this I called disableNetwork(), called setDoc() 30 times, then called enableNetwork(). I tested by throttling the network connection in Chrome Developer Tools to the network profiles "4G (slow)" and "3G". Here are the results:

4G (slow)

Without batching: 2.4 seconds
With batching: 2.8 seconds
Batching improved runtime by 0.4 seconds (14% improvement)

3G

Without batching: 6.7 seconds
With batching: 5.8 seconds
Batching improved runtime by 0.9 seconds (13% improvement)

Not surprisingly, it seems that the slower the network the greater the benefit of batching. An improvement of almost 1 second in 3G networks could be quite noticeable to a person sitting there waiting for operations to complete. This improvement would be magnified as the number of write requests increases and as the network performance decreases.

[tormjens]

@dconeybe This is great news. Thank you for your efforts here.