Refactor rollback ratio calculation and thresholds.

Introduced a new helper function `_rollback_ratio_recent` to calculate the rollback ratio over the last 15 minutes, ensuring meaningful evaluation only when a minimum transaction threshold is met. Adjusted warning and alert thresholds for rollback ratio and added a contextual metric for transaction volume in the past 15 minutes.

backend/app/services/alerts.py

@@ -151,6 +151,20 @@ async def _metric_delta(db: AsyncSession, target_id: int, metric_name: str, minu
     return max(0.0, float(latest) - float(oldest))
 
 
+async def _rollback_ratio_recent(
+    db: AsyncSession, target_id: int, minutes: int, min_total_transactions: int
+) -> tuple[float | None, float]:
+    commit_delta = await _metric_delta(db, target_id, "xact_commit", minutes=minutes)
+    rollback_delta = await _metric_delta(db, target_id, "xact_rollback", minutes=minutes)
+    if commit_delta is None or rollback_delta is None:
+        return None, 0.0
+    tx_total = commit_delta + rollback_delta
+    if tx_total < float(min_total_transactions):
+        # Too little traffic in window, ratio would be noisy and misleading.
+        return None, tx_total
+    return (rollback_delta / tx_total) if tx_total > 0 else None, tx_total
+
+
 async def _latest_query_snapshot_max(db: AsyncSession, target_id: int, column_name: str) -> float | None:
     latest_ts = await db.scalar(select(func.max(QueryStat.ts)).where(QueryStat.target_id == target_id))
     if latest_ts is None:
@@ -248,8 +262,6 @@ async def _build_standard_rules(db: AsyncSession, target: Target) -> tuple[list[
     total_connections = await _latest_metric_value(db, target.id, "connections_total")
     cache_hit_ratio = await _latest_metric_value(db, target.id, "cache_hit_ratio")
     lock_count = await _latest_metric_value(db, target.id, "locks_total")
-    xact_commit = await _latest_metric_value(db, target.id, "xact_commit")
-    xact_rollback = await _latest_metric_value(db, target.id, "xact_rollback")
     checkpoints_req_delta = await _metric_delta(db, target.id, "checkpoints_req", minutes=15)
     deadlocks_delta = await _metric_delta(db, target.id, "deadlocks", minutes=60)
     slowest_query_mean = await _latest_query_snapshot_max(db, target.id, "mean_time")
@@ -259,11 +271,9 @@ async def _build_standard_rules(db: AsyncSession, target: Target) -> tuple[list[
     if active_connections is not None and total_connections and total_connections > 0:
         active_ratio = active_connections / total_connections
 
-    rollback_ratio = None
-    if xact_commit is not None and xact_rollback is not None:
-        tx_total = xact_commit + xact_rollback
-        if tx_total > 0:
-            rollback_ratio = xact_rollback / tx_total
+    rollback_ratio_15m, tx_total_15m = await _rollback_ratio_recent(
+        db, target.id, minutes=15, min_total_transactions=20
+    )
 
     rules.extend(
         [
@@ -307,11 +317,11 @@ async def _build_standard_rules(db: AsyncSession, target: Target) -> tuple[list[
             _RuleInput(
                 key="rollback_ratio",
                 name="Rollback Ratio",
-                description="Fraction of rolled back transactions over all transactions.",
+                description="Fraction of rolled back transactions in the last 15 minutes (evaluated only when at least 20 transactions occurred).",
                 category="transactions",
-                value=rollback_ratio,
-                warning_threshold=0.05,
-                alert_threshold=0.15,
+                value=rollback_ratio_15m,
+                warning_threshold=0.10,
+                alert_threshold=0.25,
             ),
             _RuleInput(
                 key="deadlocks_60m",
@@ -342,6 +352,18 @@ async def _build_standard_rules(db: AsyncSession, target: Target) -> tuple[list[
             ),
         ]
     )
+    # Expose transaction volume as contextual metric for UI/debugging.
+    rules.append(
+        _RuleInput(
+            key="rollback_tx_volume_15m",
+            name="Rollback Ratio Evaluation Volume",
+            description="Total transactions in the last 15 minutes used for rollback-ratio evaluation.",
+            category="transactions",
+            value=tx_total_15m,
+            warning_threshold=None,
+            alert_threshold=None,
+        )
+    )
     return rules, forced_items