Tenant-aware observability and autonomous control for multi-tenant PostgreSQL. Detect noisy neighbors, attribute costs, and auto-throttle — in a single binary.
DATABASE_URL=postgres://... ./faultwallPostgreSQL sees queries. FaultWall sees tenants.
Automatically identifies your tenant isolation pattern — schema-per-tenant, row-level, or database-per-tenant. Zero config needed.
Statistical learning builds per-tenant baselines. Z-score analysis flags deviations. No LLM, no API keys — runs locally.
Linear regression on metric trends: "acme_corp will breach threshold in ~4 minutes." Act before the outage.
Kill runaway queries, enforce per-tenant connection limits. Configurable grace periods and cancel → terminate escalation.
"Tenant X costs $338/mo of your $360 RDS bill." Proportional allocation based on actual query time consumption.
10-tool MCP server lets AI agents detect noisy neighbors and throttle them autonomously. No human in the loop.
Set DATABASE_URL and run the binary. FaultWall connects as a read-only client — it never modifies your data.
Polls pg_stat_statements every 10 seconds, identifies your isolation pattern, and starts building per-tenant metrics.
Real-time dashboard: tenant leaderboard, cost attribution, anomaly alerts, and breach predictions. Auto-refreshing.
FaultWall auto-kills runaway queries and enforces limits. Or let your AI agent handle it via MCP.
Every monitoring tool has a dashboard for humans. FaultWall is the first natively controllable by AI.
What you can't see, you can't control. We went deeper than anyone else.
FaultWall connects as a read-only client. It polls Postgres catalog views (pg_stat_statements, pg_stat_activity) every 10 seconds — the same views every monitoring tool reads. It never modifies your data, never adds queries to your workload, and never touches your application traffic.
We hook directly into the Linux kernel's scheduler and block I/O subsystem using eBPF. Every CPU nanosecond and every disk byte is attributed to the exact PostgreSQL PID — and mapped back to the tenant in real-time.
Traditional tools poll aggregated stats. FaultWall traces at the kernel level: "This INSERT from tenant acme_corp used 12ms of CPU and 480KB of disk I/O." No other PostgreSQL monitoring tool provides this precision.
FaultWall deploys an autonomous optimization loop inspired by Andrej Karpathy's AutoResearch. A genetic algorithm continuously evolves detection parameters — sensitivity thresholds, window sizes, baseline intervals — against your real workload.
The longer FaultWall runs, the better it understands your database. No LLM needed — pure statistical optimization in Go.
Our eBPF engine is available for teams running self-hosted PostgreSQL. Validated on Linux 5.8+ with PostgreSQL 14-16.
Contact Us →shreyas@faultwall.com
Open source. MIT licensed. One binary. Zero config.
Get Started — it's free →