Preserve live-query re-evaluation when writes race _execOverDatabase() (#8536)

* fix: replace infinite retry loop in _execOverDatabase with counterBefore signal for eventual consistency

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/e3b3eb19-a9bb-4a2e-ade7-4a0d97dab92f

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

* test: add regression test for #7067 first-emission stall under concurrent writes

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/18b5bac4-27b1-4b4f-a295-1f30c3108f7a

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

* fix: restore fixed wait in query rerun path

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/e5ba458c-8692-4a6e-9c4c-fc4c9a0c0251

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

* test: remove complex #7067 regression case

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/7f3e4920-b04a-47bf-9b2c-a369a840a9ee

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

* docs: add changelog entry for issue 8444

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/e10a28bf-0e68-41a2-ad64-b8fc905af251

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

* docs: correct changelog wording for issue 8444

Agent-Logs-Url: https://github.com/pubkey/rxdb/sessions/698d5cce-d56a-45ec-8dd7-34d782157358

Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: pubkey <8926560+pubkey@users.noreply.github.com>

Author: Copilot

orga/changelog/fix-live-query-rerun-race-8444.md

@@ -0,0 +1 @@
+- FIX live-query subscriptions being delayed under concurrent writes, by using a fixed `promiseWait(20)` in the `_execOverDatabase()` rerun path instead of increasing the wait time with each rerun [#8444](https://github.com/pubkey/rxdb/issues/8444)

src/rx-query.ts

@@ -283,14 +283,9 @@ export class RxQueryBase<
         };
 
         /**
-         * @performance
-         * Instead of subscribing to eventBulks$ to detect concurrent writes,
-         * we snapshot the change event counter before and after the query.
-         * If the counter changed, a write happened during execution and
-         * we must re-run the query to ensure correct results.
-         * This avoids the overhead of RxJS Subject subscribe/unsubscribe per query.
-         *
-         * @link https://github.com/pubkey/rxdb/issues/7067
+         * Snapshot the counter before running the query.
+         * If a write commits during the query, the counter will be higher
+         * when we check it below, and we need to signal re-evaluation.
          */
         const counterBefore = this.collection._changeEventBuffer.getCounter();
 
@@ -343,8 +338,22 @@ export class RxQueryBase<
             };
         }
 
+        /**
+         * If a write committed concurrently during our query, our result may be
+         * stale: the write's change event was already buffered so getCounter()
+         * inside queryCollection() advanced _latestChangeEvent past the event,
+         * which would cause subsequent _ensureEqual() calls to see
+         * _isResultsInSync()=true and skip re-evaluation entirely.
+         *
+         * Re-run the query when a write commits concurrently. We still yield
+         * with a fixed wait so non-async storages do not monopolize the event
+         * loop, but we avoid the old growing backoff that could delay
+         * convergence more and more under write pressure.
+         *
+         * @link https://github.com/pubkey/rxdb/issues/7067
+         */
         if (this.collection._changeEventBuffer.getCounter() !== counterBefore) {
-            await promiseWait(rerunCount * 20);
+            await promiseWait(20);
             return this._execOverDatabase(rerunCount + 1);
         }