Database CPU utilization was spiking due to expensive notification COUNT
queries. Analysis revealed two critical issues:
1. Missing database index: Notification count queries were performing
table scans without proper indexing
2. Duplicate WHERE clauses: SQL queries contained redundant read_at IS
NULL conditions, causing unnecessary query complexity
### Root Cause Analysis
The expensive queries were:
```
-- 41.61 calls/sec with duplicate condition
SELECT COUNT(*) FROM "notifications"
WHERE "notifications"."user_id" = $1
AND "notifications"."account_id" = $2
AND "notifications"."snoozed_until" IS NULL
AND "notifications"."read_at" IS NULL
AND "notifications"."read_at" IS NULL -- Duplicate!
```
This was caused by a logic error in NotificationFinder#unread_count
introduced in commit cd06b2b33 (PR #8907). The method assumed
@notifications contained all notifications, but @notifications was
already filtered to unread notifications in most cases.
### The Default Query Flow:
1. Frontend calls: NotificationsAPI.getUnreadCount() →
/notifications/unread_count
2. No parameters sent, so params = {}
3. NotificationFinder setup:
- find_all_notifications: WHERE user_id = ? AND account_id = ?
- filter_snoozed_notifications: WHERE snoozed_until IS NULL
- filter_read_notifications: WHERE read_at IS NULL (because
type_included?('read') is false)
4. unread_count called: Adds another WHERE read_at IS NULL
----
### Solution
1. Added Missing Database Index
- Index: (user_id, account_id, snoozed_until, read_at)
2. Fixed Duplicate WHERE Clause Logic
