Metrics

The framework collects metrics at multiple levels: opcode, session, node, and network-wide. These are available via support::session_metrics on each connection and aggregated into support::node_report for the entire node.

---

metric_counter — Atomic Counter

The basic building block for metrics is metric_counter:

struct metric_counter {
    void add(uint64_t value);              // Increment by value
    // Tracks total, current_period, and last_time
};

It provides lock-free counting with period tracking for windowed aggregation.

---

opcode_metrics — Per-Opcode Counters

Each opcode within a connection tracks its own read/write counters:

struct opcode_metrics {
    metric_counter read_bytes_;
    metric_counter write_bytes_;
    metric_counter read_count_;
    metric_counter write_count_;

    void reset_metrics_window();
};

Note: opcode_report (used in node_report) has the same structure but with plain uint64_t fields instead of metric_counter — it is a snapshot, not a live counter.

---

session_metrics — Per-Connection Metrics

struct session_metrics {
    metric_counter total_read_bytes_;
    metric_counter total_write_bytes_;
    int64_t connected_at_;
    std::array<opcode_metrics, 50> opcodes_;
    latency_metrics latency_;
    metric_counter quota_warnings_;
    metric_counter quota_rejects_;

    void reset_metrics_window();
};

---

latency_metrics — Round-Trip Latency

struct latency_metrics {
    uint64_t last_;    // Last measured latency
    uint64_t min_;     // Minimum observed
    uint64_t high_;    // Maximum observed
    uint64_t avg_;     // Running average

    void add(uint64_t value, size_t history_size);
};

---

Node Reports

Aggregated metrics across the entire node:

auto report = app.get_state()->get_total_network_metrics();
// report.total_read_bytes()
// report.total_write_bytes()
// report.rtt_latency_ns_

The node_report struct contains:

struct node_report {
    uint64_t http_read_bytes_;
    uint64_t http_write_bytes_;
    uint64_t ws_read_bytes_;
    uint64_t ws_write_bytes_;
    uint64_t node_read_bytes_;
    uint64_t node_write_bytes_;
    std::array<opcode_report, 50> opcodes_;  // snapshot of opcode counters
    uint64_t mailbox_queue_size_;
    uint64_t rtt_latency_ns_;

    uint64_t total_read_bytes() const;
    uint64_t total_write_bytes() const;
};

opcode_report is the snapshot version of opcode_metrics — same fields, plain integers.

---

Metrics Collection Configuration

config.node_.metrics_interval_ms_.store(60000);  // Aggregate and report every 60s

// Reset all metrics windows
app.get_state()->reset_metrics_window();

---

Latency Tracking Configuration

config.latency_.enabled_.store(true);
config.latency_.interval_s_.store(5);       // Measure every 5 seconds
config.latency_.timeout_s_.store(60);       // Timeout after 60s
config.latency_.history_size_.store(10);     // Keep last 10 readings