Commit alert state before firing notifications

@@ -182,34 +182,40 @@ class AlertsMixin:        """        is_currently_up = current_status_code == 200        # Lock the property row so concurrent checks can't both observe the        # same alert_state and double-fire transitions.        # Commit the state transition inside the atomic block BEFORE firing        # notifications. If the save raises (e.g. SQLite "database is locked"        # from a concurrent writer), the transaction rolls back and nothing        # is emailed — the next check will retry from the same state. Sending        # first would mean a failed save leaves us firing the same alert on        # every subsequent check.        transition = None        with transaction.atomic():            locked = Property.objects.select_for_update().get(pk=self.pk)            if is_currently_up and locked.alert_state == "down":                self.send_recovery_email()                self.send_recovery_discord_message()                locked.alert_state = "up"                locked.last_alert_sent = timezone.now()                locked.save(update_fields=["alert_state", "last_alert_sent"])                self.alert_state = locked.alert_state                self.last_alert_sent = locked.last_alert_sent                transition = "recovery"            elif not is_currently_up and locked.alert_state == "up":                # Require at least 2 consecutive failures to avoid false positives.                checks = self.statuses.order_by("-created_at")[:2]                if (                    len(checks) >= 2                    and checks[0].status_code != 200                    and checks[1].status_code != 200                ):                    self.send_down_email()                    self.send_down_discord_message()                    locked.alert_state = "down"                    locked.last_alert_sent = timezone.now()                    locked.save(update_fields=["alert_state", "last_alert_sent"])                    self.alert_state = locked.alert_state                    self.last_alert_sent = locked.last_alert_sent                    transition = "down"            if transition is not None:                locked.alert_state = "up" if transition == "recovery" else "down"                locked.last_alert_sent = timezone.now()                locked.save(update_fields=["alert_state", "last_alert_sent"])                self.alert_state = locked.alert_state                self.last_alert_sent = locked.last_alert_sent        if transition == "recovery":            self.send_recovery_email()            self.send_recovery_discord_message()        elif transition == "down":            self.send_down_email()            self.send_down_discord_message()class CrawlerMixin:

@@ -48,6 +48,16 @@ DATABASES = {    "default": {        "ENGINE": "django.db.backends.sqlite3",        "NAME": "/data/db/db.sqlite3",        "OPTIONS": {            # WAL lets readers run concurrently with a writer; the scheduler            # has several worker threads, so the default rollback journal            # yields frequent "database is locked" errors. A 30s busy timeout            # gives contending writers a chance to serialize rather than            # fail, which previously stranded alert state mid-transition.            "timeout": 30,            "init_command": "PRAGMA journal_mode=WAL; PRAGMA synchronous=NORMAL;",            "transaction_mode": "IMMEDIATE",        },    }}