Fix #39: mirror an unchanged poll status with one conditional update

README.md

@@ -192,7 +192,7 @@ php artisan migrate
 
 Set `AETHER_PERSIST_TASKS=true` in your `.env`.
 
-When enabled, Aether inserts a row into `quantum_tasks` containing the circuit, shots, and driver when a task is dispatched, and updates its `status` and `counts` as the polling job progresses. The `status` always mirrors the backend's real state. Polling problems (like exhaustion or malformed responses) are logged in `error` and `failed_at`.
+When enabled, Aether inserts a row into `quantum_tasks` containing the circuit, shots, and driver when a task is dispatched, and updates its `status` and `counts` as the polling job progresses. The `status` always mirrors the backend's real state. Polling problems (like exhaustion or malformed responses) are logged in `error` and `failed_at`. While a task is queued or running, each poll issues a single conditional update that writes nothing when the status is unchanged, so an hour of polling a queued QPU task costs one query per poll and no row churn. Those intermediate updates go straight to the query builder, so `QuantumTask` model observers only see the terminal transition (completed, failed, cancelled, or a polling error); listen to `CircuitCompleted` or query the table for anything earlier.
 
 Since persistence is strictly best-effort, a database failure never affects queue behaviour or prevents the `CircuitCompleted` event from being emitted.
 

src/Jobs/PollQuantumTask.php

@@ -126,9 +126,12 @@ public function handle(QuantumManager $manager, Dispatcher $events): void
      *
      * The status column always reflects what the backend last reported; our
      * own polling problems (exhausted budget, malformed response) only ever
-     * populate error and failed_at. Persistence is best-effort: a database
-     * failure is reported and swallowed so it can never fail the job or
-     * suppress the CircuitCompleted event.
+     * populate error and failed_at. A status-only update, the common case
+     * while a task is queued or running, is a single conditional query that
+     * touches no row when the status has not changed and does not fire the
+     * QuantumTask model events; the terminal transitions do. Persistence is
+     * best-effort: a database failure is reported and swallowed so it can
+     * never fail the job or suppress the CircuitCompleted event.
      *
      * @param  array<string, int>|null  $counts
      */
@@ -139,6 +142,20 @@ private function persist(TaskStatus $status, ?array $counts = null, ?string $err
         }
 
         try {
+            if ($counts === null && $error === null) {
+                // An intermediate poll only mirrors the status: one conditional
+                // UPDATE instead of a SELECT per poll, and no row is written
+                // (nor updated_at bumped) while the backend reports the same
+                // status as before. This bypasses the model's events; the
+                // terminal transitions below still go through save().
+                QuantumTask::query()
+                    ->where('task_arn', $this->taskArn)
+                    ->where('status', '!=', $status->value)
+                    ->update(['status' => $status->value]);
+
+                return;
+            }
+
             $task = QuantumTask::query()->where('task_arn', $this->taskArn)->first();
 
             if ($task === null) {

tests/Feature/PersistenceTest.php

@@ -135,6 +135,40 @@
     expect(QuantumTask::query()->firstOrFail()->status)->toBe(TaskStatus::Running);
 });
 
+it('issues a single conditional update and writes nothing when the status is unchanged', function () {
+    $this->device->snapshotToReturn = new TaskSnapshot(TaskStatus::Running);
+    $job = ($this->submit)()->withFakeQueueInteractions();
+
+    ($this->poll)($job);
+    $before = QuantumTask::query()->firstOrFail();
+    $this->travel(5)->seconds();
+
+    DB::enableQueryLog();
+    ($this->poll)($job);
+    $queries = DB::getQueryLog();
+    DB::disableQueryLog();
+
+    expect($queries)->toHaveCount(1)
+        ->and(strtolower($queries[0]['query']))->toStartWith('update')
+        ->and(QuantumTask::query()->firstOrFail()->updated_at->equalTo($before->updated_at))->toBeTrue();
+});
+
+it('records a changed intermediate status without reading the row first', function () {
+    $this->device->snapshotToReturn = new TaskSnapshot(TaskStatus::Queued);
+    $job = ($this->submit)()->withFakeQueueInteractions();
+    ($this->poll)($job);
+
+    $this->device->snapshotToReturn = new TaskSnapshot(TaskStatus::Running);
+
+    DB::enableQueryLog();
+    ($this->poll)($job);
+    $queries = DB::getQueryLog();
+    DB::disableQueryLog();
+
+    expect($queries)->toHaveCount(1)
+        ->and(QuantumTask::query()->firstOrFail()->status)->toBe(TaskStatus::Running);
+});
+
 it('keeps the backend status and records the error when polling is exhausted', function () {
     config()->set('aether.max_poll_attempts', 1);
     $this->device->snapshotToReturn = new TaskSnapshot(TaskStatus::Running);