metrics: Use prometheus-client for proxy_build_info (#2551)

In preparation to add metrics that report float values, and ultimately in
service of replacing Linkerd's custom metrics implementation with an
upstream library, this change adds an integration with the
prometheus-client crate.

This is accomplished by leaving the existing types and traits in place
and adding a FmtMetrics implementation for the prometheus-client
Registry type, minimizing the change needed to support this new library.

New metrics can use the upstream registry and metric types. The
proxy_build_info metric has been ported to use the new API to validate
the new library. Now, the metrics dump ends with:

    # HELP proxy_build_info Proxy build info.
    # TYPE proxy_build_info gauge
    proxy_build_info{date="2023-12-05T19:52:40Z",git_sha="215c32d3f",profile="release",vendor="code@ver-sea",version="0.0.0-dev.215c32d3f"} 1
    # EOF

Cargo.lock

@@ -465,6 +465,12 @@ dependencies = [
  "tower",
 ]
 
+[[package]]
+name = "dtoa"
+version = "1.0.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dcbb2bf8e87535c23f7a8a321e364ce21462d0ff10cb6407820e8e96dfff6653"
+
 [[package]]
 name = "either"
 version = "1.8.1"
@@ -1143,6 +1149,7 @@ dependencies = [
  "linkerd-transport-metrics",
  "parking_lot",
  "pin-project",
+ "prometheus-client",
  "quickcheck",
  "regex",
  "semver",
@@ -1623,6 +1630,7 @@ dependencies = [
  "hyper",
  "linkerd-stack",
  "parking_lot",
+ "prometheus-client",
  "quickcheck",
  "tokio",
  "tracing",
@@ -2550,6 +2558,29 @@ dependencies = [
  "rustix 0.36.16",
 ]
 
+[[package]]
+name = "prometheus-client"
+version = "0.22.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "510c4f1c9d81d556458f94c98f857748130ea9737bbd6053da497503b26ea63c"
+dependencies = [
+ "dtoa",
+ "itoa",
+ "parking_lot",
+ "prometheus-client-derive-encode",
+]
+
+[[package]]
+name = "prometheus-client-derive-encode"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "440f724eba9f6996b75d63681b0a92b06947f1457076d503a4d2e2c8f56442b8"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.39",
+]
+
 [[package]]
 name = "prost"
 version = "0.12.1"

linkerd/app/core/Cargo.toml

@@ -58,6 +58,7 @@ linkerd-transport-header = { path = "../../transport-header" }
 linkerd-transport-metrics = { path = "../../transport-metrics" }
 linkerd-tls = { path = "../../tls" }
 linkerd-trace-context = { path = "../../trace-context" }
+prometheus-client = "0.22"
 regex = "1"
 serde_json = "1"
 thiserror = "1"

linkerd/app/core/src/build_info.rs

@@ -0,0 +1,28 @@
+use linkerd_metrics::prom::{self, encoding::EncodeLabelSet};
+
+pub const BUILD_INFO: BuildInfo = BuildInfo {
+    date: env!("LINKERD2_PROXY_BUILD_DATE"),
+    git_sha: env!("GIT_SHA"),
+    profile: env!("PROFILE"),
+    vendor: env!("LINKERD2_PROXY_VENDOR"),
+    version: env!("LINKERD2_PROXY_VERSION"),
+};
+
+#[derive(Copy, Clone, Debug, Default, Hash, PartialEq, Eq, EncodeLabelSet)]
+pub struct BuildInfo {
+    pub date: &'static str,
+    pub git_sha: &'static str,
+    pub profile: &'static str,
+    pub vendor: &'static str,
+    pub version: &'static str,
+}
+
+impl BuildInfo {
+    pub fn metric(&self) -> prom::Family<BuildInfo, prom::ConstGauge> {
+        let fam = prom::Family::<Self, prom::ConstGauge>::new_with_constructor(|| {
+            prom::ConstGauge::new(1)
+        });
+        let _ = fam.get_or_create(self);
+        fam
+    }
+}

linkerd/app/core/src/lib.rs

@@ -13,6 +13,7 @@
 
 use thiserror::Error;
 
+mod build_info;
 pub mod classify;
 pub mod config;
 pub mod control;
@@ -27,6 +28,7 @@ pub mod svc;
 pub mod telemetry;
 pub mod transport;
 
+pub use self::build_info::{BuildInfo, BUILD_INFO};
 pub use drain;
 pub use ipnet::{IpNet, Ipv4Net, Ipv6Net};
 pub use linkerd_addr::{self as addr, Addr, AddrMatch, IpMatch, NameAddr, NameMatch};

linkerd/app/core/src/metrics.rs

@@ -9,7 +9,7 @@
 pub use crate::transport::labels::{TargetAddr, TlsAccept};
 use crate::{
     classify::Class,
-    control, http_metrics, http_metrics as metrics, opencensus, profiles, stack_metrics,
+    control, http_metrics, opencensus, profiles, stack_metrics,
     svc::Param,
     telemetry, tls,
     transport::{self, labels::TlsConnect},
@@ -39,6 +39,9 @@ pub struct Metrics {
     pub proxy: Proxy,
     pub control: ControlHttp,
     pub opencensus: opencensus::metrics::Registry,
+
+    // Global prometheus registry for new metrics.
+    pub registry: prom::Registry,
 }
 
 #[derive(Clone, Debug)]
@@ -147,36 +150,42 @@ impl Metrics {
         retain_idle: Duration,
         start_time: telemetry::StartTime,
     ) -> (Self, impl FmtMetrics + Clone + Send + 'static) {
+        let registry = prom::Registry::default();
+
         let process = telemetry::process::Report::new(start_time);
 
-        let build_info = telemetry::build_info::Report::default();
+        registry.write().register(
+            "proxy_build_info",
+            "Proxy build info",
+            crate::BUILD_INFO.metric(),
+        );
 
         let (control, control_report) = {
-            let m = metrics::Requests::<ControlLabels, Class>::default();
+            let m = http_metrics::Requests::<ControlLabels, Class>::default();
             let r = m.clone().into_report(retain_idle).with_prefix("control");
             (m, r)
         };
 
         let (http_endpoint, endpoint_report) = {
-            let m = metrics::Requests::<EndpointLabels, Class>::default();
+            let m = http_metrics::Requests::<EndpointLabels, Class>::default();
             let r = m.clone().into_report(retain_idle);
             (m, r)
         };
 
         let (http_profile_route, profile_route_report) = {
-            let m = metrics::Requests::<ProfileRouteLabels, Class>::default();
+            let m = http_metrics::Requests::<ProfileRouteLabels, Class>::default();
             let r = m.clone().into_report(retain_idle).with_prefix("route");
             (m, r)
         };
 
         let (http_profile_route_retry, retry_report) = {
-            let m = metrics::Retries::<ProfileRouteLabels>::default();
+            let m = http_metrics::Retries::<ProfileRouteLabels>::default();
             let r = m.clone().into_report(retain_idle).with_prefix("route");
             (m, r)
         };
 
         let (http_profile_route_actual, actual_report) = {
-            let m = metrics::Requests::<ProfileRouteLabels, Class>::default();
+            let m = http_metrics::Requests::<ProfileRouteLabels, Class>::default();
             let r = m
                 .clone()
                 .into_report(retain_idle)
@@ -203,6 +212,7 @@ impl Metrics {
             proxy,
             control,
             opencensus,
+            registry: registry.clone(),
         };
 
         let report = endpoint_report
@@ -214,7 +224,9 @@ impl Metrics {
             .and_report(opencensus_report)
             .and_report(stack)
             .and_report(process)
-            .and_report(build_info);
+            // The prom registry reports an "# EOF" at the end of its export, so
+            // it should be emitted last.
+            .and_report(registry);
 
         (metrics, report)
     }

linkerd/app/core/src/telemetry.rs

@@ -1,3 +1,3 @@
-pub mod build_info;
 pub mod process;
+
 pub use self::process::StartTime;

linkerd/metrics/Cargo.toml

@@ -16,6 +16,7 @@ http = "0.2"
 hyper = { version = "0.14", features = ["http1", "http2"] }
 linkerd-stack = { path = "../stack", optional = true }
 parking_lot = "0.12"
+prometheus-client = "0.22"
 tokio = { version = "1", features = ["time"] }
 tracing = "0.1"
 

linkerd/metrics/src/counter.rs

@@ -1,5 +1,5 @@
 use super::{
-    prom::{FmtLabels, FmtMetric},
+    fmt::{FmtLabels, FmtMetric},
     Factor,
 };
 use std::fmt::{self, Display};

linkerd/metrics/src/gauge.rs

@@ -1,4 +1,4 @@
-use super::prom::{FmtLabels, FmtMetric};
+use super::fmt::{FmtLabels, FmtMetric};
 use std::fmt::{self, Display};
 use std::sync::atomic::{AtomicU64, Ordering};
 

linkerd/metrics/src/lib.rs

@@ -4,26 +4,55 @@
 //! Utilities for exposing metrics to Prometheus.
 
 mod counter;
+mod fmt;
 mod gauge;
 mod histogram;
 pub mod latency;
 #[cfg(feature = "linkerd-stack")]
 mod new_metrics;
-mod prom;
 mod serve;
 mod store;
 
 #[cfg(feature = "linkerd-stack")]
 pub use self::new_metrics::NewMetrics;
 pub use self::{
     counter::Counter,
+    fmt::{FmtLabels, FmtMetric, FmtMetrics, Metric},
     gauge::Gauge,
     histogram::Histogram,
-    prom::{FmtLabels, FmtMetric, FmtMetrics, Metric},
     serve::Serve,
     store::{LastUpdate, SharedStore, Store},
 };
 
+/// Integration with the [`prometheus_client`]` crate.
+///
+/// This should be used for all new metrics.
+pub mod prom {
+    use parking_lot::RwLock;
+    use std::sync::Arc;
+
+    pub use prometheus_client::{
+        metrics::{
+            counter::{ConstCounter, Counter},
+            family::Family,
+            gauge::{ConstGauge, Gauge},
+            histogram::Histogram,
+            info::Info,
+        },
+        *,
+    };
+
+    /// New metrics should use the prometheus-client Registry.
+    pub type Registry = Arc<RwLock<registry::Registry>>;
+
+    impl crate::FmtMetrics for Registry {
+        #[inline]
+        fn fmt_metrics(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+            encoding::text::encode(f, &self.read())
+        }
+    }
+}
+
 #[macro_export]
 macro_rules! metrics {
     { $( $name:ident : $kind:ty { $help:expr } ),+ } => {

linkerd2-proxy/src/main.rs

@@ -12,10 +12,7 @@ compile_error!(
 );
 
 use linkerd_app::{
-    core::{
-        telemetry::{build_info, StartTime},
-        transport::BindTcp,
-    },
+    core::{telemetry::StartTime, transport::BindTcp, BUILD_INFO},
     trace, Config,
 };
 use linkerd_signal as signal;
@@ -42,11 +39,11 @@ fn main() {
 
     info!(
         "{profile} {version} ({sha}) by {vendor} on {date}",
-        date = build_info::DATE,
-        sha = build_info::GIT_SHA,
-        version = build_info::VERSION,
-        profile = build_info::PROFILE,
-        vendor = build_info::VENDOR,
+        date = BUILD_INFO.date,
+        sha = BUILD_INFO.git_sha,
+        version = BUILD_INFO.version,
+        profile = BUILD_INFO.profile,
+        vendor = BUILD_INFO.vendor,
     );
 
     // Load configuration from the environment without binding ports.