Monitoring stack: Prometheus + Grafana dashboards

[x3c41a]

Fourth slice of #267: the dashboards that make a stress run readable at a glance.

Stacked on #272 (the domain-state exporter), which provides most of the metrics these dashboards chart; base retargets to dev once #272 merges.

What it adds

monitoring/ — a docker-compose Prometheus + Grafana stack with provisioned dashboards.

Scrape targets (all on the host, reached via host.docker.internal):

• web3-storage-chain — the domain-state exporter (chain-exporter.ts), :9700
• parachain-node — the collator's built-in Substrate metrics, :9615
• relay-node — the relay validator's metrics, :9616 (optional)

The collator and relay Prometheus ports are pinned in zombienet.toml (zombienet otherwise assigns a random port each run, which a static scrape config can't target).

Three provisioned dashboards (Grafana folder "Web3 Storage"):

• Chain Health — block height (best vs finalized), finality lag, block production rate, txpool, peers, and per-class block weight. The mandatory-class PoV panel is where the challenge-deadline sweep spikes (5 MiB budget line), tying directly to the Challenge-saturation stress prototype #271 finding.
• Economics — provider stake, free/reserved balance, cumulative slashes, checkpoint rewards and pool.
• Protocol Activity — agreements active vs expired (the Scheduling & lazy processing: move unbounded on_finalize work to on_poll / bounded lazy cleanup #177 lazy-cleanup residual), pending challenges, challenge lifecycle (created/defended/slashed) and rate, checkpoints submitted/missed, provider/bucket counts.

Dashboard JSON is generated by grafana/gen-dashboards.mjs (dashboards-as-code) rather than hand-edited.

Also here

A small addition to the exporter (on the #272 branch): per-class block_weight_ref_time / block_weight_proof_size gauges from System.BlockWeight. Substrate's own Prometheus doesn't expose block weight, and it's the signal the Chain Health dashboard needs to show the deadline sweep spike.

Tested

Docker isn't available on my machine, so I verified the stack by running the real Prometheus and Grafana binaries natively against a live zombienet + exporter:

• promtool check config passes on prometheus/prometheus.yml.
• Prometheus scrapes parachain-node (:9615) and web3-storage-chain (:9700) — both targets up; every metric the dashboards query returns data (e.g. substrate_block_height{status="best"}, web3storage_block_weight_proof_size).
• Grafana provisions the datasource (health check: "Successfully queried the Prometheus API") and all three dashboards, and a range query through Grafana's datasource proxy returns a live series.
• The collator Prometheus-port pin was confirmed live (spawn command shows --prometheus-port 9615); the relay pin uses the identical zombienet mechanism.
• All YAML/JSON config files parse.

docker compose up itself hasn't been run here (no Docker on this box); the compose file is standard and the config it mounts is the same one verified natively.