feat: Add basic Grafana/Prometheus infra (#12)

- Add grafana/prometheus to run locally
- Add declarative dashboard generator
- Fix a bunch of issue with exported metrics

Author: popzxc

ethereum_prover/configs/ethproofs_prod.yaml

@@ -8,3 +8,4 @@ eth_prover:
   block_mod: 100
   prover_id: 0
   on_failure: "continue"
+  prometheus_port: 9898

ethereum_prover/configs/ethproofs_staging.yaml

@@ -8,3 +8,4 @@ eth_prover:
   block_mod: 10
   prover_id: 0
   on_failure: "continue"
+  prometheus_port: 9898

ethereum_prover/configs/local_debug.yaml

@@ -7,5 +7,6 @@ eth_prover:
   ethproofs_submission: "off"
   block_mod: 1
   prover_id: 0
-  on_failure: "exit"
+  on_failure: "continue"
+  prometheus_port: 9898
 

ethereum_prover/src/clients/ethproofs/mod.rs

@@ -93,7 +93,7 @@ impl EthproofsClient {
         payload: &T,
         context: &'static str,
     ) -> anyhow::Result<()> {
-        let latency = METRICS.ethproofs_request_duration_seconds.start();
+        let latency = METRICS.ethproofs_request_duration.start();
         for attempt in 1..=MAX_ATTEMPTS {
             let response = self
                 .client

ethereum_prover/src/metrics.rs

@@ -12,18 +12,18 @@ pub struct ProverMetrics {
     pub witness_success_total: Counter<u64>,
     pub witness_failure_total: Counter<u64>,
     #[metrics(buckets = Buckets::LATENCIES, unit = Unit::Seconds)]
-    pub witness_duration_seconds: Histogram<Duration>,
+    pub witness_duration: Histogram<Duration>,
     pub inflight_witness_tasks: Gauge<u64>,
     pub proof_success_total: Counter<u64>,
     pub proof_failure_total: Counter<u64>,
     #[metrics(buckets = Buckets::LATENCIES, unit = Unit::Seconds)]
-    pub proof_duration_seconds: Histogram<Duration>,
+    pub proof_duration: Histogram<Duration>,
     pub inflight_proof_tasks: Gauge<u64>,
     pub last_processed_block: Gauge<u64>,
     pub ethproofs_request_success_total: Counter<u64>,
     pub ethproofs_request_failure_total: Counter<u64>,
     #[metrics(buckets = Buckets::LATENCIES, unit = Unit::Seconds)]
-    pub ethproofs_request_duration_seconds: Histogram<Duration>,
+    pub ethproofs_request_duration: Histogram<Duration>,
 }
 
 #[vise::register]

ethereum_prover/src/tasks/cpu_witness.rs

@@ -55,7 +55,7 @@ impl CpuWitnessTask {
             );
             let block_number = witness.block_header.number;
             let _inflight = InflightGuard::new(&METRICS.inflight_witness_tasks);
-            let latency = METRICS.witness_duration_seconds.start();
+            let latency = METRICS.witness_duration.start();
             match self.process_block(witness).await {
                 Ok(cpu_witness) => {
                     tracing::info!("Generated CPU witness for block {}", block_number);

ethereum_prover/src/tasks/gpu_prove.rs

@@ -43,7 +43,7 @@ impl GpuProveTask {
             );
             let block_number = witness.block_header.number;
             let _inflight = InflightGuard::new(&METRICS.inflight_proof_tasks);
-            let latency = METRICS.proof_duration_seconds.start();
+            let latency = METRICS.proof_duration.start();
             self.command_sender
                 .send(CalculationUpdate::ProofQueued { block_number })
                 .await

infra/README.md

@@ -0,0 +1,44 @@
+# Infra: Local Metrics Stack
+
+This folder contains a local Prometheus + Grafana setup for `ethereum_prover`, plus a
+TypeScript (Grafana Foundation SDK) dashboard generator.
+
+## Layout
+
+- `infra/docker-compose.yml`: Prometheus + Grafana services.
+- `infra/prometheus/prometheus.yml`: Prometheus scrape config (targets host at `9898`).
+- `infra/grafana/provisioning/`: Grafana datasource + dashboard provisioning.
+- `infra/grafana/dashboards/`: Generated dashboard JSON output.
+- `infra/dashboards/`: Dashboard source (TypeScript) and build script.
+
+## Prerequisites
+
+- `docker` + `docker compose`
+- `node` + `yarn` (for dashboard generation)
+- `ethereum_prover` running on host with metrics enabled:
+  - `eth_prover_prometheus_port=9898`
+
+## Generate dashboards
+
+From repo root:
+
+```sh
+./infra/dashboards/build.sh
+```
+
+This writes `infra/grafana/dashboards/ethereum-prover.json`.
+
+## Run Prometheus + Grafana
+
+```sh
+docker compose -f infra/docker-compose.yml up
+```
+
+Grafana: http://localhost:3000
+
+Prometheus: http://localhost:9090
+
+## Notes
+
+- Prometheus scrapes `host.docker.internal:9898`. On Linux, `host.docker.internal`
+  is provided via `extra_hosts: host-gateway` in the compose file.

infra/dashboards/.gitignore

@@ -0,0 +1,3 @@
+/dist
+/node_modules
+/wasm

infra/dashboards/build.sh

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "$script_dir"
+
+yarn install
+yarn build