Genarrative/server-rs/crates/shared-logging/src/lib.rs

use std::io;

use opentelemetry::{KeyValue, global, trace::TracerProvider};
use opentelemetry_appender_tracing::layer::OpenTelemetryTracingBridge;
use opentelemetry_otlp::WithExportConfig;
use opentelemetry_sdk::{
    Resource,
    logs::SdkLoggerProvider,
    metrics::SdkMeterProvider,
    trace::SdkTracerProvider,
};
use tracing::warn;
use tracing_subscriber::{
    EnvFilter, Layer, filter::LevelFilter, fmt, layer::SubscriberExt, util::SubscriberInitExt,
};

#[derive(Clone, Copy, Debug, Default)]
pub struct OtelConfig {
    pub enabled: bool,
}

// 统一解析工作区日志过滤器，优先环境变量，其次回落到调用方传入的默认值。
pub fn resolve_env_filter(default_filter: &str) -> EnvFilter {
    EnvFilter::try_from_default_env()
        .or_else(|_| EnvFilter::try_new(default_filter))
        .unwrap_or_else(|_| EnvFilter::new("info"))
}

// 统一初始化 tracing subscriber，避免各入口重复散落相同配置。
pub fn init_tracing(default_filter: &str, otel_config: OtelConfig) -> Result<(), io::Error> {
    let env_filter = resolve_env_filter(default_filter);
    let fmt_layer = fmt::layer().with_target(true).with_ansi(false).compact();

    if !otel_config.enabled {
        return tracing_subscriber::registry()
            .with(env_filter)
            .with(fmt_layer)
            .try_init()
            .map_err(|error| io::Error::other(format!("初始化 tracing subscriber 失败：{error}")));
    }

    let Some(otel) = build_otel_pipeline() else {
        return tracing_subscriber::registry()
            .with(env_filter)
            .with(fmt_layer)
            .try_init()
            .map_err(|error| io::Error::other(format!("初始化 tracing subscriber 失败：{error}")));
    };

    tracing_subscriber::registry()
        .with(env_filter)
        .with(fmt_layer)
        .with(
            tracing_opentelemetry::layer()
                .with_tracer(otel.tracer_provider.tracer("genarrative-api")),
        )
        .with(
            OpenTelemetryTracingBridge::new(&otel.logger_provider).with_filter(LevelFilter::INFO),
        )
        .try_init()
        .map_err(|error| io::Error::other(format!("初始化 tracing subscriber 失败：{error}")))
}

struct OtelPipeline {
    tracer_provider: SdkTracerProvider,
    _meter_provider: SdkMeterProvider,
    logger_provider: SdkLoggerProvider,
}

fn build_otel_pipeline() -> Option<OtelPipeline> {
    let resource = Resource::builder()
        .with_service_name(read_env_or_default("OTEL_SERVICE_NAME", "genarrative-api"))
        .with_attribute(KeyValue::new("service.namespace", "genarrative"))
        .build();

    let span_exporter = match opentelemetry_otlp::SpanExporter::builder()
        .with_http()
        .with_endpoint(resolve_otlp_http_signal_endpoint(
            "OTEL_EXPORTER_OTLP_TRACES_ENDPOINT",
            "/v1/traces",
        ))
        .build()
    {
        Ok(exporter) => exporter,
        Err(error) => {
            warn!(%error, "OpenTelemetry span exporter 初始化失败，已回退为本地日志");
            return None;
        }
    };

    let metric_exporter = match opentelemetry_otlp::MetricExporter::builder()
        .with_http()
        .with_endpoint(resolve_otlp_http_signal_endpoint(
            "OTEL_EXPORTER_OTLP_METRICS_ENDPOINT",
            "/v1/metrics",
        ))
        .build()
    {
        Ok(exporter) => exporter,
        Err(error) => {
            warn!(%error, "OpenTelemetry metric exporter 初始化失败，已回退为本地日志");
            return None;
        }
    };

    let log_exporter = match opentelemetry_otlp::LogExporter::builder()
        .with_http()
        .with_endpoint(resolve_otlp_http_signal_endpoint(
            "OTEL_EXPORTER_OTLP_LOGS_ENDPOINT",
            "/v1/logs",
        ))
        .build()
    {
        Ok(exporter) => exporter,
        Err(error) => {
            warn!(%error, "OpenTelemetry log exporter 初始化失败，已回退为本地日志");
            return None;
        }
    };

    let tracer_provider = SdkTracerProvider::builder()
        .with_resource(resource.clone())
        .with_batch_exporter(span_exporter)
        .build();
    let meter_provider = SdkMeterProvider::builder()
        .with_resource(resource)
        .with_periodic_exporter(metric_exporter)
        .build();
    let logger_provider = SdkLoggerProvider::builder()
        .with_resource(Resource::builder()
            .with_service_name(read_env_or_default("OTEL_SERVICE_NAME", "genarrative-api"))
            .with_attribute(KeyValue::new("service.namespace", "genarrative"))
            .build())
        .with_batch_exporter(log_exporter)
        .build();

    global::set_tracer_provider(tracer_provider.clone());
    global::set_meter_provider(meter_provider.clone());

    Some(OtelPipeline {
        tracer_provider,
        _meter_provider: meter_provider,
        logger_provider,
    })
}

fn read_env_or_default(key: &str, default_value: &str) -> String {
    std::env::var(key)
        .ok()
        .filter(|value| !value.trim().is_empty())
        .unwrap_or_else(|| default_value.to_string())
}

fn resolve_otlp_http_signal_endpoint(signal_key: &str, signal_path: &str) -> String {
    if let Ok(value) = std::env::var(signal_key)
        && !value.trim().is_empty()
    {
        return value;
    }

    append_otlp_signal_path(
        &read_env_or_default("OTEL_EXPORTER_OTLP_ENDPOINT", "http://127.0.0.1:4318"),
        signal_path,
    )
}

fn append_otlp_signal_path(base_endpoint: &str, signal_path: &str) -> String {
    let base_endpoint = base_endpoint.trim_end_matches('/');
    let signal_path = signal_path.trim_start_matches('/');
    format!("{base_endpoint}/{signal_path}")
}

#[cfg(test)]
mod tests {
    use std::sync::{Mutex, OnceLock};

    use super::resolve_otlp_http_signal_endpoint;

    const OTEL_ENDPOINT_ENV_KEYS: [&str; 4] = [
        "OTEL_EXPORTER_OTLP_ENDPOINT",
        "OTEL_EXPORTER_OTLP_TRACES_ENDPOINT",
        "OTEL_EXPORTER_OTLP_METRICS_ENDPOINT",
        "OTEL_EXPORTER_OTLP_LOGS_ENDPOINT",
    ];

    fn env_lock() -> std::sync::MutexGuard<'static, ()> {
        static LOCK: OnceLock<Mutex<()>> = OnceLock::new();
        LOCK.get_or_init(|| Mutex::new(())).lock().unwrap()
    }

    fn clear_otel_endpoint_env() {
        unsafe {
            for key in OTEL_ENDPOINT_ENV_KEYS {
                std::env::remove_var(key);
            }
        }
    }

    #[test]
    fn generic_otlp_http_endpoint_expands_to_signal_paths() {
        let _guard = env_lock();
        clear_otel_endpoint_env();
        unsafe {
            std::env::set_var("OTEL_EXPORTER_OTLP_ENDPOINT", "http://127.0.0.1:4318");
        }

        assert_eq!(
            resolve_otlp_http_signal_endpoint("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT", "/v1/traces"),
            "http://127.0.0.1:4318/v1/traces"
        );
        assert_eq!(
            resolve_otlp_http_signal_endpoint("OTEL_EXPORTER_OTLP_METRICS_ENDPOINT", "/v1/metrics"),
            "http://127.0.0.1:4318/v1/metrics"
        );
        assert_eq!(
            resolve_otlp_http_signal_endpoint("OTEL_EXPORTER_OTLP_LOGS_ENDPOINT", "/v1/logs"),
            "http://127.0.0.1:4318/v1/logs"
        );

        clear_otel_endpoint_env();
    }
}