Zero-footprint, predictive observability for Solana validators.

An 8MB sidecar that detects node degradation before it impacts your epoch returns. Zero external API fees, zero bloat.

The Motivation: The SFDP Validator Prune

The Solana Foundation Delegation Program is actively pruning underperforming nodes. Bad performance now equals immediate removal and loss of delegated stake. SentinelSOL was engineered to give operators a predictive edge, catching hardware degradation and network exhaustion before it results in on-chain delinquency and stake-stripping.

Read the Blockworks Report ↗

View on GitHub Read the Docs

sentinel-dashboard - Grafana

SentinelSOL Grafana dashboard showing validator observability metrics

Why SentinelSOL

The old way costs you vote credits.

The Trap

Reactive Monitoring is a Trap.

Relying on external third-party monitors means by the time an alert fires, you have already lost vote credits. Latency inherent to cloud-based polling puts your performance metrics behind reality by minutes, minutes that cost you.

Alert latency: 60–300s

Vote credits lost: unrecoverable

The Solution

Predictive Z-Score Math.

SentinelSOL monitors absolute vote velocity locally, catching hardware exhaustion 3 standard deviations before total failure. No polling lag. No external API round-trips. Just raw signal, processed on-node, in real time.

Detection window: < 5s

Degradation threshold: σ = 3.0

Built for operators

Engineering-first by design.

go binary

8MB

Peak Footprint

Written in raw Go, idling at 2MB. No JVM, no interpreter, no runtime overhead stealing cycles from your validator.

local-first

Egress Fees

Local RPC scraping means no cloud bandwidth charges. Your telemetry never leaves the machine unless you want it to.

docker-compose

Deploy Command

One docker-compose command deploys the Extractor, Prometheus, and Grafana. Fully wired, zero manual config.

Source-verifiable

Real Go. No abstractions hiding the work.

The extractor polls your local RPC endpoint on a tight ticker, computes a rolling z-score over vote-velocity deltas, and pushes metrics to a local Prometheus exporter - all in a single self-contained binary you can audit line-by-line.

go 1.25.3 · linux/amd64

MIT License

extractor/main.go

1package main
2
3import (
4	"net/http"
5	"time"
6)
7
8// httpClient is shared across all RPC polling goroutines.
9// A single instance with a hard timeout prevents goroutine leaks
10// under degraded network conditions.
11var httpClient = &http.Client{
12	Timeout: 5 * time.Second,
13}

About the Project

Why this exists.

Built by Hamza - SRE & DevOps Engineer. When analyzing the Solana ecosystem from a reliability perspective, a critical gap became apparent: existing validator observability is entirely reactive. SentinelSOL was engineered as a proactive, out-of-band capable telemetry layer. By predicting hardware exhaustion and monitoring ShredStream latency, it protects operators from stake pruning and Jito MEV loss before failures occur. Built for the Colosseum Hackathon 2026.