Switch allocator to jemalloc (#3697)

## Proposed Changes

Another `tree-states` motivated PR, this adds `jemalloc` as the default allocator, with an option to use the system allocator by compiling with `FEATURES="" make`.

- [x] Metrics
- [x] Test on Windows
- [x] Test on macOS
- [x] Test with `musl`
- [x] Metrics dashboard on `lighthouse-metrics` (https://github.com/sigp/lighthouse-metrics/pull/37)


Co-authored-by: Michael Sproul <micsproul@gmail.com>

common/malloc_utils/Cargo.toml

@@ -4,13 +4,21 @@ version = "0.1.0"
 authors = ["Paul Hauner <paul@paulhauner.com>"]
 edition = "2021"
 
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
 [dependencies]
 lighthouse_metrics = { path = "../lighthouse_metrics" }
 lazy_static = "1.4.0"
 libc = "0.2.79"
 parking_lot = "0.12.0"
+jemalloc-ctl = { version = "0.5.0", optional = true }
+
+# Jemalloc's background_threads feature requires Linux (pthreads).
+[target.'cfg(target_os = "linux")'.dependencies]
+jemallocator = { version = "0.5.0", optional = true, features = ["stats", "background_threads"] }
+
+[target.'cfg(not(target_os = "linux"))'.dependencies]
+jemallocator = { version = "0.5.0", optional = true, features = ["stats"] }
 
 [features]
 mallinfo2 = []
+jemalloc = ["jemallocator", "jemalloc-ctl"]
+jemalloc-profiling = ["jemallocator/profiling"]

common/malloc_utils/src/jemalloc.rs

@@ -0,0 +1,52 @@
+//! Set the allocator to `jemalloc`.
+//!
+//! Due to `jemalloc` requiring configuration at compile time or immediately upon runtime
+//! initialisation it is configured via a Cargo config file in `.cargo/config.toml`.
+//!
+//! The `jemalloc` tuning can be overriden by:
+//!
+//! A) `JEMALLOC_SYS_WITH_MALLOC_CONF` at compile-time.
+//! B) `_RJEM_MALLOC_CONF` at runtime.
+use jemalloc_ctl::{arenas, epoch, stats, Error};
+use lazy_static::lazy_static;
+use lighthouse_metrics::{set_gauge, try_create_int_gauge, IntGauge};
+
+#[global_allocator]
+static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
+
+// Metrics for jemalloc.
+lazy_static! {
+    pub static ref NUM_ARENAS: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_num_arenas", "The number of arenas in use");
+    pub static ref BYTES_ALLOCATED: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_allocated", "Equivalent to stats.allocated");
+    pub static ref BYTES_ACTIVE: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_active", "Equivalent to stats.active");
+    pub static ref BYTES_MAPPED: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_mapped", "Equivalent to stats.mapped");
+    pub static ref BYTES_METADATA: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_metadata", "Equivalent to stats.metadata");
+    pub static ref BYTES_RESIDENT: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_resident", "Equivalent to stats.resident");
+    pub static ref BYTES_RETAINED: lighthouse_metrics::Result<IntGauge> =
+        try_create_int_gauge("jemalloc_bytes_retained", "Equivalent to stats.retained");
+}
+
+pub fn scrape_jemalloc_metrics() {
+    scrape_jemalloc_metrics_fallible().unwrap()
+}
+
+pub fn scrape_jemalloc_metrics_fallible() -> Result<(), Error> {
+    // Advance the epoch so that the underlying statistics are updated.
+    epoch::advance()?;
+
+    set_gauge(&NUM_ARENAS, arenas::narenas::read()? as i64);
+    set_gauge(&BYTES_ALLOCATED, stats::allocated::read()? as i64);
+    set_gauge(&BYTES_ACTIVE, stats::active::read()? as i64);
+    set_gauge(&BYTES_MAPPED, stats::mapped::read()? as i64);
+    set_gauge(&BYTES_METADATA, stats::metadata::read()? as i64);
+    set_gauge(&BYTES_RESIDENT, stats::resident::read()? as i64);
+    set_gauge(&BYTES_RETAINED, stats::retained::read()? as i64);
+
+    Ok(())
+}

common/malloc_utils/src/lib.rs

@@ -2,18 +2,18 @@
 //!
 //! ## Conditional Compilation
 //!
-//! Presently, only configuration for "The GNU Allocator" from `glibc` is supported. All other
-//! allocators are ignored.
+//! This crate can be compiled with different feature flags to support different allocators:
 //!
-//! It is assumed that if the following two statements are correct then we should expect to
-//! configure `glibc`:
+//! - Jemalloc, via the `jemalloc` feature.
+//! - GNU malloc, if no features are set and the system supports it.
+//! - The system allocator, if no features are set and the allocator is not GNU malloc.
+//!
+//! It is assumed that if Jemalloc is not in use, and the following two statements are correct then
+//! we should expect to configure `glibc`:
 //!
 //! - `target_os = linux`
 //! - `target_env != musl`
 //!
-//! In all other cases this library will not attempt to do anything (i.e., all functions are
-//! no-ops).
-//!
 //! If the above conditions are fulfilled but `glibc` still isn't present at runtime then a panic
 //! may be triggered. It is understood that there's no way to be certain that a compatible `glibc`
 //! is present: https://github.com/rust-lang/rust/issues/33244.
@@ -24,18 +24,42 @@
 //! detecting `glibc` are best-effort. If this crate throws errors about undefined external
 //! functions, then try to compile with the `not_glibc_interface` module.
 
-#[cfg(all(target_os = "linux", not(target_env = "musl")))]
+#[cfg(all(
+    target_os = "linux",
+    not(target_env = "musl"),
+    not(feature = "jemalloc")
+))]
 mod glibc;
 
+#[cfg(feature = "jemalloc")]
+mod jemalloc;
+
 pub use interface::*;
 
-#[cfg(all(target_os = "linux", not(target_env = "musl")))]
+#[cfg(all(
+    target_os = "linux",
+    not(target_env = "musl"),
+    not(feature = "jemalloc")
+))]
 mod interface {
     pub use crate::glibc::configure_glibc_malloc as configure_memory_allocator;
     pub use crate::glibc::scrape_mallinfo_metrics as scrape_allocator_metrics;
 }
 
-#[cfg(any(not(target_os = "linux"), target_env = "musl"))]
+#[cfg(feature = "jemalloc")]
+mod interface {
+    #[allow(dead_code)]
+    pub fn configure_memory_allocator() -> Result<(), String> {
+        Ok(())
+    }
+
+    pub use crate::jemalloc::scrape_jemalloc_metrics as scrape_allocator_metrics;
+}
+
+#[cfg(all(
+    any(not(target_os = "linux"), target_env = "musl"),
+    not(feature = "jemalloc")
+))]
 mod interface {
     #[allow(dead_code, clippy::unnecessary_wraps)]
     pub fn configure_memory_allocator() -> Result<(), String> {