Open-source performance diagnostics and root-cause analysis for relational databases.
culpritsql.com · Guide · Architecture · Coding Agents (MCP) · Collector & ADX · Releases
| Mode | How it works |
|---|---|
| Direct Desktop · CLI · MCP |
Connect directly to a database (or open a saved snapshot offline) to inspect top queries, ranked findings, execution plans, live blocking, virtual indexes, and SQL migrations in CI or coding agents. |
| Fleet Collector ADX · ClickHouse |
Deploy a bounded container in your database network and push verified telemetry and synthesized root-cause findings to Azure Data Explorer (ADX) or ClickHouse (with Blob or S3 recovery). |
- Top queries by database time, wait attribution & ranked findings
- Execution plan regressions, parameter sensitivity & side-by-side plan diffs
@TVPcardinality skew, memory-grant starvation & disk spills- Page-latch (
PAGELATCH) hot pages &tempdballocation contention - Blocking chains, DDL metadata-lock sandwiches & deadlocks
- Worker-thread (
THREADPOOL) & compiler-gateway starvation - PostgreSQL subtransaction (
SubtransSLRU) &XIDwraparound cliffs - Transaction-log stalls, slow client consumers & replica redo blocks
- Version-store (
PVS/ vacuum) retention & cloud tier throttling - Missing/unused indexes & unsafe SQL migrations in pull requests
Point an Azure Monitor Scheduled Query Rule or Grafana alert at ADX to page on-call with the culprit and fix already attached:
// Runs every 1m in Azure Monitor / Grafana -> PagerDuty, Slack, or Teams
CulpritIncidentsV1
| where UpdatedAt > ago(5m) and Status == "active" and Severity == "high"
| project StartedAt, DatabaseName, CulpritKind, Headline, CulpritSql, DeveloperAction