feat: add node_id and run_id labels to request metrics for multi-run isolation

[cbaugus]

Problem

When a worker node runs more than one test (e.g. idle → new POST /config → running again),
all runs produce metrics with identical label sets (region, tenant). There is no way
to distinguish test 1 from test 2 in Prometheus/GMP queries or dashboards.

Additionally node_id (the hostname of the worker) is only present on the
rust_loadtest_node_info gauge, not on the per-request counters and histograms,
so per-node breakdowns require joining an info metric.

Proposed changes

1. Add run_id to YamlMetadata (src/yaml_config.rs)

metadata:
  name: "Acme checkout load test"
  tenant: "acme"
  run_id: "run-20260321-001234"   # generated by control plane (webload-gui#83)

If run_id is absent from the YAML (e.g. standalone env-var boot), auto-generate
a value at POST /config time:

let run_id = yaml_cfg.metadata.run_id
    .unwrap_or_else(|| format!("auto-{}", chrono::Utc::now().timestamp()));

2. Add node_id and run_id label dimensions to all request metrics (src/metrics.rs)

Affected metrics:

Metric	Current labels	New labels
rust_loadtest_requests_total	region, tenant	region, tenant, node_id, run_id
rust_loadtest_request_errors_by_category	category, region, tenant	+ node_id, run_id
rust_loadtest_request_duration_seconds (histogram)	region, tenant	+ node_id, run_id
Per-step histograms (step, scenario)	step, scenario, region, tenant	+ node_id, run_id

node_id is read from CLUSTER_NODE_ID env var (already available at startup).

3. Thread node_id and run_id through WorkerConfig

pub struct WorkerConfig {
    // existing fields ...
    pub node_id: String,
    pub run_id: String,
}

4. Expose active run_id in GET /health

{
  "node_id": "worker-run-20260321-001",
  "run_id":  "run-20260321-001234",
  ...
}

Allows the control plane to confirm which run is active and correlate GMP metrics.

5. run_id in TestState

struct TestState {
    // existing fields ...
    run_id: String,
}

Reset to a new auto-generated value on each POST /config, same as generation.

Example GMP queries after change

# Total RPS for a specific run
sum(rate(rust_loadtest_requests_total{run_id="run-20260321-001234"}[1m]))

# Compare two runs on the same node
sum(rate(rust_loadtest_requests_total{node_id="worker-1"}[1m])) by (run_id)

# Error rate per run across all nodes for a tenant
sum(rate(rust_loadtest_request_errors_by_category{tenant="acme"}[5m])) by (run_id, node_id)

Acceptance criteria

• run_id field added to YamlMetadata; auto-generated (timestamp-based) if absent
• node_id and run_id present on requests_total, errors_by_category, duration histograms, and per-step histograms
• WorkerConfig carries node_id and run_id; passed through to all metric label calls
• GET /health response includes active run_id
• TestState stores run_id; reset on each new POST /config
• Existing tests updated for new label dimensions
• Integration test: two sequential POST /config calls produce metrics with different run_id values

Related

• webload-gui#83 — control plane generates and includes run_id in dispatched YAML
• Depends on no other issues

[cbaugus]

Implemented in commit 29c0751 (feat) + 7448e84 (rustfmt fix), included in PR #107.

All 12 metrics now carry node_id and run_id labels:

• requests_total, request_duration_seconds, concurrent_requests, request_errors_by_category, requests_status_codes_total
• scenario_runs_total, scenario_duration_seconds, scenario_steps_total, scenario_step_duration_seconds, scenario_step_status_codes_total, scenario_assertions_total, scenario_throughput_bytes_total

run_id is read from metadata.run_id in the posted YAML, or auto-generated as run-<unix_timestamp> if absent. It is also surfaced in GET /health.