//! Copyright (c) 2022-2026 GitDataAi All rights reserved.

//! Prometheus-compatible metrics for GitKS.
//!
//! Tracks:
//! - Request counts by gRPC method + status code
//! - Request duration histogram by method
//! - Active requests gauge
//! - Repository count
//! - Cache hits / misses
//! - Error counts by error type
//! - Git subprocess duration
//! - Cache operation duration
//! - Slow request detection
//!
//! Exposes HTTP endpoints on a configurable port (default 9100):
//! - GET /metrics          — Prometheus metrics
//! - GET /health           — Liveness probe
//! - GET /ready            — Readiness probe
//! - GET /debug/log-level  — Current log level filter
//! - PUT /debug/log-level  — Update log level filter (body: "gitks=debug")
//! - GET /debug/config     — Runtime configuration

use dashmap::DashMap;
use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
use std::sync::{Arc, OnceLock};
use std::time::{Duration, Instant};

struct MetricsInner {
    /// Counter: total requests by (method, status_code)
    request_count: DashMap<String, AtomicU64>,
    /// Histogram buckets for request duration (ms)
    duration_buckets: DashMap<String, AtomicU64>,
    /// Gauge: number of currently in-flight requests
    active_requests: AtomicU64,
    /// Gauge: total number of registered repositories
    repository_count: AtomicU64,
    /// Counter: cache hits
    cache_hits: AtomicU64,
    /// Counter: cache misses
    cache_misses: AtomicU64,
    /// Counter: errors by error kind
    error_count: DashMap<String, AtomicU64>,
    /// Start timestamp
    start_time: Instant,
    /// Slow request threshold in ms
    slow_request_threshold_ms: AtomicU64,
    /// Server ready flag
    ready: AtomicBool,
    /// Histogram buckets for git subprocess duration (ms)
    git_cmd_duration_buckets: DashMap<String, AtomicU64>,
    /// Counter: git subprocess calls by command
    git_cmd_count: DashMap<String, AtomicU64>,
    /// Histogram buckets for cache operation duration (ms)
    cache_op_duration_buckets: DashMap<String, AtomicU64>,
    /// Counter: cache operations by (cache, result)
    cache_op_count: DashMap<String, AtomicU64>,
    /// Histogram: hook execution duration (ms)
    hook_duration_buckets: DashMap<String, AtomicU64>,
    /// Counter: hook executions by (hook_type, result)
    hook_count: DashMap<String, AtomicU64>,
    /// Counter: slow requests by method
    slow_request_count: DashMap<String, AtomicU64>,
    /// Counter: cache evictions by (cause, namespace)
    cache_eviction_count: DashMap<String, AtomicU64>,
    /// Counter: cache hits by namespace
    cache_hit_by_namespace: DashMap<String, AtomicU64>,
    /// Counter: cache misses by namespace
    cache_miss_by_namespace: DashMap<String, AtomicU64>,
    /// Histogram: cache value size in bytes
    cache_value_size_buckets: DashMap<String, AtomicU64>,
    /// Counter: rate-limit rejections by repository
    rate_limit_reject_count: DashMap<String, AtomicU64>,
    /// Counter: rate-limit acquires by repository
    rate_limit_acquire_count: DashMap<String, AtomicU64>,
}

static METRICS: OnceLock<Arc<MetricsInner>> = OnceLock::new();

/// Handle for dynamic log level reload.
static LOG_RELOAD_HANDLE: OnceLock<
    Option<tracing_subscriber::reload::Handle<EnvFilter, tracing_subscriber::Registry>>,
> = OnceLock::new();

use tracing_subscriber::EnvFilter;

fn metrics() -> &'static Arc<MetricsInner> {
    METRICS.get_or_init(|| {
        Arc::new(MetricsInner {
            request_count: DashMap::new(),
            duration_buckets: DashMap::new(),
            active_requests: AtomicU64::new(0),
            repository_count: AtomicU64::new(0),
            cache_hits: AtomicU64::new(0),
            cache_misses: AtomicU64::new(0),
            error_count: DashMap::new(),
            start_time: Instant::now(),
            slow_request_threshold_ms: AtomicU64::new(5000),
            ready: AtomicBool::new(false),
            git_cmd_duration_buckets: DashMap::new(),
            git_cmd_count: DashMap::new(),
            cache_op_duration_buckets: DashMap::new(),
            cache_op_count: DashMap::new(),
            hook_duration_buckets: DashMap::new(),
            hook_count: DashMap::new(),
            slow_request_count: DashMap::new(),
            cache_eviction_count: DashMap::new(),
            cache_hit_by_namespace: DashMap::new(),
            cache_miss_by_namespace: DashMap::new(),
            cache_value_size_buckets: DashMap::new(),
            rate_limit_reject_count: DashMap::new(),
            rate_limit_acquire_count: DashMap::new(),
        })
    })
}

#[rustfmt::skip]
const DURATION_BUCKET_MS: &[u64] = &[
    5, 10, 25, 50, 100, 250, 500, 1_000,
    2_500, 5_000, 10_000, 30_000, 60_000, u64::MAX,
];

const BUCKET_INF: u64 = u64::MAX;

fn record_duration_bucket(map: &DashMap<String, AtomicU64>, key_prefix: &str, duration_ms: u64) {
    for &bound_ms in DURATION_BUCKET_MS {
        if duration_ms <= bound_ms || bound_ms == BUCKET_INF {
            let bucket_key = format!("{key_prefix}:{bound_ms}");
            map.entry(bucket_key)
                .or_insert_with(|| AtomicU64::new(0))
                .value()
                .fetch_add(1, Ordering::Relaxed);
        }
    }
}

pub fn set_slow_request_threshold(ms: u64) {
    metrics()
        .slow_request_threshold_ms
        .store(ms, Ordering::Relaxed);
}

pub fn set_ready(ready: bool) {
    metrics().ready.store(ready, Ordering::Relaxed);
}

pub fn set_log_reload_handle(
    handle: tracing_subscriber::reload::Handle<EnvFilter, tracing_subscriber::Registry>,
) {
    LOG_RELOAD_HANDLE.set(Some(handle)).ok();
}

/// Record a gRPC request.
pub fn record_request(method: &str, status_code: &str, duration: Duration) {
    let m = metrics();
    let duration_ms = duration.as_millis() as u64;

    let key = format!("{method}:{status_code}");
    m.request_count
        .entry(key)
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);

    record_duration_bucket(&m.duration_buckets, method, duration_ms);

    let threshold = m.slow_request_threshold_ms.load(Ordering::Relaxed);
    if threshold > 0 && duration_ms >= threshold {
        m.slow_request_count
            .entry(method.to_string())
            .or_insert_with(|| AtomicU64::new(0))
            .value()
            .fetch_add(1, Ordering::Relaxed);
    }
}

pub fn inc_active_requests() {
    metrics().active_requests.fetch_add(1, Ordering::Relaxed);
}

pub fn dec_active_requests() {
    metrics().active_requests.fetch_sub(1, Ordering::Relaxed);
}

pub fn set_repository_count(count: u64) {
    metrics().repository_count.store(count, Ordering::Relaxed);
}

pub fn inc_cache_hits(count: u64) {
    metrics().cache_hits.fetch_add(count, Ordering::Relaxed);
}

pub fn inc_cache_misses(count: u64) {
    metrics().cache_misses.fetch_add(count, Ordering::Relaxed);
}

pub fn inc_error(kind: &str) {
    metrics()
        .error_count
        .entry(kind.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record a git subprocess execution.
pub fn record_git_cmd(command: &str, duration: Duration) {
    let m = metrics();
    let duration_ms = duration.as_millis() as u64;

    m.git_cmd_count
        .entry(command.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);

    record_duration_bucket(&m.git_cmd_duration_buckets, command, duration_ms);
}

/// Record a cache operation.
pub fn record_cache_op(cache: &str, result: &str, duration: Duration) {
    let m = metrics();
    let duration_ms = duration.as_millis() as u64;
    let key = format!("{cache}:{result}");

    m.cache_op_count
        .entry(key)
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);

    record_duration_bucket(&m.cache_op_duration_buckets, cache, duration_ms);
}

/// Record a cache entry eviction.
pub fn record_cache_eviction(namespace: &str, cause: &str) {
    let m = metrics();
    let key = format!("{cause}:{namespace}");
    m.cache_eviction_count
        .entry(key)
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record a per-namespace cache hit.
pub fn record_cache_hit_ns(namespace: &str) {
    metrics()
        .cache_hit_by_namespace
        .entry(namespace.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record a per-namespace cache miss.
pub fn record_cache_miss_ns(namespace: &str) {
    metrics()
        .cache_miss_by_namespace
        .entry(namespace.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record a hook execution.
pub fn record_hook_execution(hook_type: &str, result: &str, duration: Duration) {
    let m = metrics();
    let duration_ms = duration.as_millis() as u64;
    let key = format!("{hook_type}:{result}");

    m.hook_count
        .entry(key)
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);

    record_duration_bucket(&m.hook_duration_buckets, hook_type, duration_ms);
}

/// Record cache value size distribution (in bytes).
pub fn record_cache_value_size(namespace: &str, size: usize) {
    let m = metrics();
    record_size_bucket(&m.cache_value_size_buckets, namespace, size as u64);
}

/// Record a rate-limit rejection event.
pub fn record_rate_limit_reject(repo: &str) {
    let m = metrics();
    m.rate_limit_reject_count
        .entry(repo.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record a rate-limit acquire event.
pub fn record_rate_limit_acquire(repo: &str) {
    let m = metrics();
    m.rate_limit_acquire_count
        .entry(repo.to_string())
        .or_insert_with(|| AtomicU64::new(0))
        .value()
        .fetch_add(1, Ordering::Relaxed);
}

/// Record size distribution buckets (log2-based: 1KB, 4KB, 16KB, ..., 1GB).
fn record_size_bucket(map: &DashMap<String, AtomicU64>, label: &str, size: u64) {
    let buckets = [1024, 4096, 16384, 65536, 262144, 1048576, 4194304, 16777216, 67108864, 268435456, 1073741824];
    for &bound in &buckets {
        let key = format!("{label}:le_{bound}");
        if size <= bound {
            map.entry(key)
                .or_insert_with(|| AtomicU64::new(0))
                .value()
                .fetch_add(1, Ordering::Relaxed);
        }
    }
}

/// Escape a string for use as a Prometheus label value.
/// Replaces `\` → `\\`, `"` → `\"`, `\n` → `\n` per the Prometheus spec.
fn prom_escape(value: &str) -> String {
    let mut out = String::with_capacity(value.len());
    for ch in value.chars() {
        match ch {
            '\\' => out.push_str("\\\\"),
            '"' => out.push_str("\\\""),
            '\n' => out.push_str("\\n"),
            _ => out.push(ch),
        }
    }
    out
}

fn render_counter_map(
    out: &mut String,
    name: &str,
    help: &str,
    map: &DashMap<String, AtomicU64>,
    labels: &[&str],
) {
    out.push_str(&format!("# HELP {name} {help}\n"));
    out.push_str(&format!("# TYPE {name} counter\n"));
    for entry in map {
        let (key, count) = (entry.key(), entry.value().load(Ordering::Relaxed));
        let parts: Vec<&str> = key.split(':').collect();
        if parts.len() == labels.len() {
            let label_str: String = labels
                .iter()
                .zip(parts.iter())
                .map(|(l, v)| format!("{l}=\"{}\"", prom_escape(v)))
                .collect::<Vec<_>>()
                .join(",");
            out.push_str(&format!("{name}{{{label_str}}} {count}\n"));
        }
    }
    out.push('\n');
}

fn render_histogram(out: &mut String, name: &str, help: &str, map: &DashMap<String, AtomicU64>) {
    out.push_str(&format!("# HELP {name} {help}\n"));
    out.push_str(&format!("# TYPE {name} histogram\n"));
    for entry in map {
        let (key_and_bound, count) = (entry.key(), entry.value().load(Ordering::Relaxed));
        if let Some((key, bound_str)) = key_and_bound.rsplit_once(':') {
            let le = if bound_str.parse::<u64>() == Ok(BUCKET_INF) {
                "+Inf".to_string()
            } else {
                bound_str.to_string()
            };
            out.push_str(&format!(
                "{name}_bucket{{method=\"{}\",le=\"{le}\"}} {count}\n",
                prom_escape(key)
            ));
        }
    }
    out.push('\n');
}

pub fn render_metrics() -> String {
    let m = metrics();
    let mut out = String::with_capacity(8192);

    let uptime = m.start_time.elapsed().as_secs();
    out.push_str("# HELP gitks_uptime_seconds Time since gitks started\n");
    out.push_str("# TYPE gitks_uptime_seconds gauge\n");
    out.push_str(&format!("gitks_uptime_seconds {uptime}\n\n"));

    let active = m.active_requests.load(Ordering::Relaxed);
    out.push_str("# HELP gitks_active_requests Currently in-flight requests\n");
    out.push_str("# TYPE gitks_active_requests gauge\n");
    out.push_str(&format!("gitks_active_requests {active}\n\n"));

    let repos = m.repository_count.load(Ordering::Relaxed);
    out.push_str("# HELP gitks_repository_count Number of registered repositories\n");
    out.push_str("# TYPE gitks_repository_count gauge\n");
    out.push_str(&format!("gitks_repository_count {repos}\n\n"));

    render_counter_map(
        &mut out,
        "gitks_requests_total",
        "Total gRPC requests by method and status",
        &m.request_count,
        &["method", "status"],
    );

    render_histogram(
        &mut out,
        "gitks_request_duration_milliseconds",
        "Request duration histogram in ms",
        &m.duration_buckets,
    );

    render_counter_map(
        &mut out,
        "gitks_slow_requests_total",
        "Slow gRPC requests by method",
        &m.slow_request_count,
        &["method"],
    );

    let hits = m.cache_hits.load(Ordering::Relaxed);
    let misses = m.cache_misses.load(Ordering::Relaxed);
    out.push_str("# HELP gitks_cache_hits_total Cache hit count\n");
    out.push_str("# TYPE gitks_cache_hits_total counter\n");
    out.push_str(&format!("gitks_cache_hits_total {hits}\n\n"));
    out.push_str("# HELP gitks_cache_misses_total Cache miss count\n");
    out.push_str("# TYPE gitks_cache_misses_total counter\n");
    out.push_str(&format!("gitks_cache_misses_total {misses}\n\n"));

    render_counter_map(
        &mut out,
        "gitks_errors_total",
        "Total errors by kind",
        &m.error_count,
        &["kind"],
    );

    render_counter_map(
        &mut out,
        "gitks_git_cmd_total",
        "Git subprocess calls by command",
        &m.git_cmd_count,
        &["command"],
    );
    render_histogram(
        &mut out,
        "gitks_git_cmd_duration_milliseconds",
        "Git subprocess duration in ms",
        &m.git_cmd_duration_buckets,
    );

    render_counter_map(
        &mut out,
        "gitks_cache_ops_total",
        "Cache operations by cache and result",
        &m.cache_op_count,
        &["cache", "result"],
    );
    render_histogram(
        &mut out,
        "gitks_cache_op_duration_milliseconds",
        "Cache operation duration in ms",
        &m.cache_op_duration_buckets,
    );

    render_counter_map(
        &mut out,
        "gitks_cache_evictions_total",
        "Cache evictions by cause and namespace",
        &m.cache_eviction_count,
        &["cause", "namespace"],
    );

    render_counter_map(
        &mut out,
        "gitks_cache_hits_by_namespace_total",
        "Cache hits by namespace",
        &m.cache_hit_by_namespace,
        &["namespace"],
    );

    render_counter_map(
        &mut out,
        "gitks_cache_misses_by_namespace_total",
        "Cache misses by namespace",
        &m.cache_miss_by_namespace,
        &["namespace"],
    );

    render_counter_map(
        &mut out,
        "gitks_hook_executions_total",
        "Hook executions by type and result",
        &m.hook_count,
        &["hook_type", "result"],
    );
    render_histogram(
        &mut out,
        "gitks_hook_duration_milliseconds",
        "Hook execution duration in ms",
        &m.hook_duration_buckets,
    );

    render_histogram(
        &mut out,
        "gitks_cache_value_size_bytes",
        "Cache value size distribution in bytes",
        &m.cache_value_size_buckets,
    );

    render_counter_map(
        &mut out,
        "gitks_rate_limit_rejects_total",
        "Rate-limit rejections by repository",
        &m.rate_limit_reject_count,
        &["repo"],
    );
    render_counter_map(
        &mut out,
        "gitks_rate_limit_acquires_total",
        "Rate-limit acquires by repository",
        &m.rate_limit_acquire_count,
        &["repo"],
    );

    out
}

use bytes::Bytes;
use http_body_util::Full;
use hyper::body::Incoming;
use hyper::service::Service;
use hyper::{Method, Request, Response};
use std::convert::Infallible;
use std::future::Future;
use std::net::SocketAddr;
use std::pin::Pin;

/// Global cancel token for the HTTP server, set from main.
static HTTP_CANCEL: OnceLock<tokio_util::sync::CancellationToken> = OnceLock::new();

pub fn set_http_cancel_token(token: tokio_util::sync::CancellationToken) {
    HTTP_CANCEL.set(token).ok();
}

struct Router;

impl Service<Request<Incoming>> for Router {
    type Response = Response<Full<Bytes>>;
    type Error = Infallible;
    type Future = Pin<Box<dyn Future<Output = Result<Self::Response, Self::Error>> + Send>>;

    fn call(&self, req: Request<Incoming>) -> Self::Future {
        Box::pin(handle_request(req))
    }
}

fn json_response(status: u16, body: &str) -> Response<Full<Bytes>> {
    match Response::builder()
        .status(status)
        .header("Content-Type", "application/json")
        .header("Connection", "close")
        .body(Full::new(Bytes::from(body.to_string())))
    {
        Ok(response) => response,
        Err(err) => {
            tracing::error!(error = %err, "failed to build JSON response");
            Response::new(Full::new(Bytes::from_static(
                br#"{"error":"response build failed"}"#,
            )))
        }
    }
}

fn text_response(status: u16, content_type: &str, body: String) -> Response<Full<Bytes>> {
    match Response::builder()
        .status(status)
        .header("Content-Type", content_type)
        .header("Connection", "close")
        .body(Full::new(Bytes::from(body)))
    {
        Ok(response) => response,
        Err(err) => {
            tracing::error!(error = %err, "failed to build text response");
            Response::new(Full::new(Bytes::from_static(b"response build failed")))
        }
    }
}

async fn handle_request(req: Request<Incoming>) -> Result<Response<Full<Bytes>>, Infallible> {
    let method = req.method().clone();
    let path = req.uri().path().to_owned();

    tracing::debug!(%method, %path, "HTTP request");

    let response = match (method, path.as_str()) {
        (Method::GET, "/metrics") => {
            let body = render_metrics();
            text_response(200, "text/plain; version=0.0.4; charset=utf-8", body)
        }
        (Method::GET, "/health") => json_response(200, r#"{"status":"healthy"}"#),
        (Method::GET, "/ready") => {
            if metrics().ready.load(Ordering::Relaxed) {
                json_response(200, r#"{"status":"ready"}"#)
            } else {
                json_response(503, r#"{"status":"not_ready"}"#)
            }
        }
        (Method::GET, "/debug/log-level") => {
            let msg = if LOG_RELOAD_HANDLE.get().is_some() {
                "use PUT to change"
            } else {
                "dynamic log level not available"
            };
            json_response(200, &format!(r#"{{"log_level":"{msg}"}}"#))
        }
        (Method::PUT, "/debug/log-level") => handle_log_level_update(req)
            .await
            .unwrap_or_else(|e| json_response(400, &format!(r#"{{"error":"{e}"}}"#))),
        (Method::GET, "/debug/config") => {
            let threshold = metrics().slow_request_threshold_ms.load(Ordering::Relaxed);
            let ready = metrics().ready.load(Ordering::Relaxed);
            json_response(
                200,
                &format!(
                    r#"{{"slow_request_threshold_ms":{},"ready":{}}}"#,
                    threshold, ready
                ),
            )
        }
        _ => json_response(404, r#"{"error":"not found"}"#),
    };

    Ok(response)
}

async fn handle_log_level_update(req: Request<Incoming>) -> Result<Response<Full<Bytes>>, String> {
    use http_body_util::BodyExt;

    let body_bytes = req
        .collect()
        .await
        .map_err(|e| format!("failed to read body: {e}"))?
        .to_bytes();

    let new_filter =
        String::from_utf8(body_bytes.to_vec()).map_err(|e| format!("invalid UTF-8: {e}"))?;
    let new_filter = new_filter.trim().to_string();

    if new_filter.is_empty() {
        return Ok(json_response(400, r#"{"error":"empty filter"}"#));
    }

    match LOG_RELOAD_HANDLE.get() {
        Some(Some(handle)) => match new_filter.parse::<EnvFilter>() {
            Ok(filter) => {
                if let Err(e) = handle.reload(filter) {
                    Ok(json_response(500, &format!(r#"{{"error":"{e}"}}"#)))
                } else {
                    tracing::info!(new_filter = %new_filter, "log level updated via HTTP");
                    Ok(json_response(
                        200,
                        &format!(r#"{{"status":"ok","filter":"{}"}}"#, new_filter),
                    ))
                }
            }
            Err(e) => Ok(json_response(
                400,
                &format!(r#"{{"error":"invalid filter: {e}"}}"#),
            )),
        },
        _ => Ok(json_response(
            501,
            r#"{"error":"dynamic log level not configured"}"#,
        )),
    }
}

/// Start the HTTP server (metrics + health + debug) using hyper 1.x.
pub fn start_metrics_server(port: u16) -> tokio::task::JoinHandle<()> {
    tokio::spawn(async move {
        let listener =
            match tokio::net::TcpListener::bind(SocketAddr::from(([0, 0, 0, 0], port))).await {
                Ok(l) => l,
                Err(e) => {
                    tracing::error!(port, error = %e, "failed to bind HTTP server");
                    return;
                }
            };
        tracing::info!(port, "HTTP server started (metrics + health + debug)");

        let cancel = HTTP_CANCEL
            .get()
            .cloned()
            .unwrap_or_else(tokio_util::sync::CancellationToken::new);

        loop {
            tokio::select! {
                result = listener.accept() => {
                    match result {
                        Ok((stream, _peer)) => {
                            let io = hyper_util::rt::TokioIo::new(stream);
                            tokio::spawn(async move {
                                let builder = hyper::server::conn::http1::Builder::new();
                                if let Err(e) = builder
                                    .serve_connection(io, Router)
                                    .with_upgrades()
                                    .await
                                {
                                    tracing::debug!(error = %e, "HTTP connection error");
                                }
                            });
                        }
                        Err(e) => {
                            tracing::error!(error = %e, "HTTP accept error");
                        }
                    }
                }
                _ = cancel.cancelled() => {
                    tracing::info!("HTTP server shutting down");
                    break;
                }
            }
        }
    })
}

pub struct RequestMetrics {
    method: &'static str,
    start: Instant,
}

impl RequestMetrics {
    pub fn new(method: &'static str) -> Self {
        inc_active_requests();
        Self {
            method,
            start: Instant::now(),
        }
    }

    /// Record the outcome. Idempotent — safe to call before each return.
    pub fn record(&self, status: &str) {
        let duration = self.start.elapsed();
        let duration_ms = duration.as_millis() as u64;
        record_request(self.method, status, duration);

        let threshold = metrics().slow_request_threshold_ms.load(Ordering::Relaxed);
        if threshold > 0 && duration_ms >= threshold {
            tracing::warn!(
                method = self.method,
                status,
                elapsed_ms = duration_ms,
                threshold_ms = threshold,
                "slow gRPC request detected"
            );
        }
    }
}

impl Drop for RequestMetrics {
    fn drop(&mut self) {
        dec_active_requests();
    }
}

/// Convenience: record an error from a tonic Status.
pub fn record_rpc_error(m: &RequestMetrics, status: &tonic::Status) {
    let kind = status.code().description();
    inc_error(kind);
    m.record(kind);
}