Place zcashd.debug.* metrics behind a -debugmetrics config option

Cargo.lock

@@ -50,6 +50,15 @@ dependencies = [
  "version_check",
 ]
 
+[[package]]
+name = "aho-corasick"
+version = "0.7.19"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b4f55bd91a0978cbfd91c457a164bab8b4001c833b7f323132c0a4e1922dd44e"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "anyhow"
 version = "1.0.65"
@@ -964,6 +973,7 @@ dependencies = [
  "memuse",
  "metrics",
  "metrics-exporter-prometheus",
+ "metrics-util",
  "nonempty",
  "orchard",
  "rand 0.8.5",
@@ -977,6 +987,7 @@ dependencies = [
  "subtle",
  "thiserror",
  "time",
+ "tokio",
  "tracing",
  "tracing-appender",
  "tracing-core",
@@ -1108,6 +1119,7 @@ version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f7d24dc2dbae22bff6f1f9326ffce828c9f07ef9cc1e8002e5279f845432a30a"
 dependencies = [
+ "aho-corasick",
  "crossbeam-epoch",
  "crossbeam-utils",
  "hashbrown",

Cargo.toml

@@ -77,6 +77,8 @@ rayon = "1.5"
 ipnet = "2"
 metrics = "0.20"
 metrics-exporter-prometheus = "0.11"
+metrics-util = { version = "0.14", default-features = false, features = ["layer-filter"] }
+tokio = { version = "1", features = ["rt", "net", "time"] }
 
 # General tool dependencies
 gumdrop = "0.8"

qa/rpc-tests/show_help.py

@@ -343,6 +343,9 @@ Monitoring options:
        any request path. Use -metricsallowip and -metricsbind to control
        access.
 
+  -debugmetrics
+       Include debug metrics in exposed node metrics.
+
 Debugging/Testing options:
 
   -debug=<category>

qa/supply-chain/config.toml

@@ -51,6 +51,10 @@ criteria = "safe-to-deploy"
 version = "0.7.6"
 criteria = "safe-to-deploy"
 
+[[exemptions.aho-corasick]]
+version = "0.7.19"
+criteria = "safe-to-deploy"
+
 [[exemptions.anyhow]]
 version = "1.0.56"
 criteria = "safe-to-deploy"

src/init.cpp

@@ -438,6 +438,7 @@ std::string HelpMessage(HelpMessageMode mode)
     strUsage += HelpMessageOpt("-prometheusport=<port>", _("Expose node metrics in the Prometheus exposition format. "
             "An HTTP listener will be started on <port>, which responds to GET requests on any request path. "
             "Use -metricsallowip and -metricsbind to control access."));
+    strUsage += HelpMessageOpt("-debugmetrics", _("Include debug metrics in exposed node metrics."));
 
     strUsage += HelpMessageGroup(_("Debugging/Testing options:"));
     if (showDebug)
@@ -1471,10 +1472,12 @@ bool AppInit2(boost::thread_group& threadGroup, CScheduler& scheduler)
             metricsBindCstr = metricsBind.c_str();
         }
 
+        bool debugMetrics = GetBoolArg("-debugmetrics", false);
+
         // Start up the metrics runtime. This spins off a Rust thread that runs
         // the Prometheus exporter. We just let this thread die at process end.
         LogPrintf("metrics thread start");
-        if (!metrics_run(metricsBindCstr, vAllowCstr.data(), vAllowCstr.size(), prometheusPort)) {
+        if (!metrics_run(metricsBindCstr, vAllowCstr.data(), vAllowCstr.size(), prometheusPort, debugMetrics)) {
             return InitError(strprintf(_("Failed to start Prometheus metrics exporter")));
         }
     }

src/rust/include/rust/metrics.h

@@ -24,7 +24,8 @@ bool metrics_run(
     const char* bind_address,
     const char* const* allow_ips,
     size_t allow_ips_len,
-    uint16_t prometheus_port);
+    uint16_t prometheus_port,
+    bool debug_metrics);
 
 struct MetricsCallsite;
 typedef struct MetricsCallsite MetricsCallsite;

src/rust/src/metrics_ffi.rs

@@ -1,18 +1,48 @@
 use libc::{c_char, c_double};
 use metrics::{try_recorder, Key, Label};
-use metrics_exporter_prometheus::PrometheusBuilder;
+use metrics_exporter_prometheus::{BuildError, PrometheusBuilder};
+use metrics_util::layers::{FilterLayer, Stack};
+
 use std::ffi::CStr;
 use std::net::{IpAddr, SocketAddr};
 use std::ptr;
 use std::slice;
+use std::thread;
+
 use tracing::error;
 
+/// Builds the recorder and exporter, applies the given filters to the recorder, and
+/// installs both of them globally.
+fn metrics_install(builder: PrometheusBuilder, filters: &[&str]) -> Result<(), BuildError> {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .map_err(|e| BuildError::FailedToCreateRuntime(e.to_string()))?;
+
+    let (recorder, exporter) = {
+        let _g = runtime.enter();
+        builder.build()?
+    };
+
+    thread::Builder::new()
+        .name("zc-metrics".to_string())
+        .spawn(move || runtime.block_on(exporter))
+        .map_err(|e| BuildError::FailedToCreateRuntime(e.to_string()))?;
+
+    Stack::new(recorder)
+        .push(FilterLayer::from_patterns(filters))
+        .install()?;
+
+    Ok(())
+}
+
 #[no_mangle]
 pub extern "C" fn metrics_run(
     bind_address: *const c_char,
     allow_ips: *const *const c_char,
     allow_ips_len: usize,
     prometheus_port: u16,
+    debug_metrics: bool,
 ) -> bool {
     // Convert the C string IPs to Rust strings.
     let allow_ips = unsafe { slice::from_raw_parts(allow_ips, allow_ips_len) };
@@ -54,6 +84,13 @@ pub extern "C" fn metrics_run(
         prometheus_port,
     );
 
+    // Metrics matching any of these filters will be discarded.
+    let filters = if debug_metrics {
+        &[][..]
+    } else {
+        &["zcashd.debug."]
+    };
+
     allow_ips
         .into_iter()
         .try_fold(
@@ -66,7 +103,7 @@ pub extern "C" fn metrics_run(
             },
         )
         .and_then(|builder| {
-            builder.install().map_err(|e| {
+            metrics_install(builder, filters).map_err(|e| {
                 error!("Could not install Prometheus exporter: {}", e);
                 e
             })