feat: add jolt-profiling crate by markosg04 · Pull Request #1364 · a16z/jolt

markosg04 · 2026-03-25T19:05:29Z

Summary

Changes

Testing

Ran tests for modified crates
cargo clippy and cargo fmt pass

Security Considerations

Breaking Changes

None

markosg04 · 2026-03-25T19:05:48Z

feat: add jolt-host crate #1369
feat: add jolt-crypto crate #1368
feat: add jolt-poly crate #1367
feat: add jolt-instructions crate #1366
feat: add jolt-transcript crate #1365
feat: add jolt-profiling crate #1364 👈 (View in Graphite)
feat: add jolt-field crate #1363
chore: workspace scaffolding for modular crates #1362
main

This stack of pull requests is managed by Graphite. Learn more about stacking.

github-actions · 2026-04-03T20:54:35Z

Warning

This PR has more than 500 changed lines and does not include a spec.

Large features and architectural changes benefit from a spec-driven workflow.
See CONTRIBUTING.md for details on how to create a spec.

If this PR is a bug fix, refactor, or doesn't warrant a spec, feel free to ignore this message.

Remove duplicate [workspace.lints.clippy] section introduced by rebase, add workspace lint inheritance to jolt-profiling, bump sysinfo to workspace version, and fix all lint violations.

0xAndoroid

7 issues found after deduplication across semantic, bug, tech-debt, and security review passes. Two earlier issues (missing workspace lints, sysinfo version mismatch) were fixed in b003fba.

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/monitor.rs

+                let interval = Duration::from_millis((interval_secs * 1000.0) as u64);
+                let mut system = System::new_all();
+
+                thread::sleep(sysinfo::MINIMUM_CPU_UPDATE_INTERVAL);
+
+                while !stop.load(Ordering::Relaxed) {
+                    system.refresh_all();


System::new_all() initializes disk, network, component, and process lists — none of which are read. refresh_all() refreshes all of them every tick. For a monitor that only needs CPU and memory this is significant unnecessary overhead per sample.

Suggested change

let interval = Duration::from_millis((interval_secs * 1000.0) as u64);

let mut system = System::new_all();

thread::sleep(sysinfo::MINIMUM_CPU_UPDATE_INTERVAL);

while !stop.load(Ordering::Relaxed) {

system.refresh_all();

let interval = Duration::from_millis((interval_secs * 1000.0) as u64);

let mut system = System::new();

thread::sleep(sysinfo::MINIMUM_CPU_UPDATE_INTERVAL);

while !stop.load(Ordering::Relaxed) {

system.refresh_cpu_all();

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/units.rs

+///
+/// Uses GiB for values >= 1.0, otherwise MiB.
+pub fn format_memory_size(gib: f64) -> String {
+    if gib >= 1.0 {


Memory deltas from memory.rs can be negative (memory freed during a span). A delta of -2.0 GiB fails the >= 1.0 check, falls into the MiB branch, and formats as "-2048.00 MiB" instead of "-2.00 GiB".

Suggested change

if gib >= 1.0 {

if gib.abs() >= 1.0 {

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/flamegraph.rs

+    let mut opts = Options::default();
+    opts.color_diffusion = true;
+    opts.count_name = String::from("MiB");
+    opts.factor = 1.0 / BYTES_PER_GIB * 1024.0;


The intent is bytes→MiB (matching count_name = "MiB"), but the expression takes a detour through GiB. BYTES_PER_MIB is already defined in units.

Suggested change

opts.factor = 1.0 / BYTES_PER_GIB * 1024.0;

opts.factor = 1.0 / BYTES_PER_MIB;

Also update the import on line 8 to bring in BYTES_PER_MIB instead of (or in addition to) BYTES_PER_GIB.

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/memory.rs

+    let mut memory_usage_map = MEMORY_USAGE_MAP.lock().unwrap();
+    let memory_gib_start = memory_usage_map
+        .remove(label)
+        .unwrap_or_else(|| panic!("no open memory span: {label}"));
+
+    let delta = memory_gib_end - memory_gib_start;
+    let mut memory_delta_map = MEMORY_DELTA_MAP.lock().unwrap();
+    assert_eq!(memory_delta_map.insert(label, delta), None);


MEMORY_USAGE_MAP is held while acquiring MEMORY_DELTA_MAP. If the assert_eq! on line 67 panics, the first mutex is poisoned and every subsequent call to start_memory_tracing_span/end_memory_tracing_span panics with PoisonError. Drop the first lock before acquiring the second:

Suggested change

let mut memory_usage_map = MEMORY_USAGE_MAP.lock().unwrap();

let memory_gib_start = memory_usage_map

.remove(label)

.unwrap_or_else(|| panic!("no open memory span: {label}"));

let delta = memory_gib_end - memory_gib_start;

let mut memory_delta_map = MEMORY_DELTA_MAP.lock().unwrap();

assert_eq!(memory_delta_map.insert(label, delta), None);

let memory_gib_start = {

let mut map = MEMORY_USAGE_MAP.lock().unwrap();

map.remove(label)

.unwrap_or_else(|| panic!("no open memory span: {label}"))

};

let delta = memory_gib_end - memory_gib_start;

let mut memory_delta_map = MEMORY_DELTA_MAP.lock().unwrap();

assert_eq!(memory_delta_map.insert(label, delta), None);

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/monitor.rs

+        let handle = thread::Builder::new()
+            .name("metrics-monitor".to_string())
+            .spawn(move || {
+                let interval = Duration::from_millis((interval_secs * 1000.0) as u64);


(negative_f64 * 1000.0) as u64 saturates to 0, and NaN also casts to 0. This creates a Duration::ZERO sleep, turning the loop into a CPU spin. Clamp to a sane minimum.

Suggested change

let interval = Duration::from_millis((interval_secs * 1000.0) as u64);

let interval = Duration::from_millis(((interval_secs * 1000.0) as u64).max(50));

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/lib.rs

+#[cfg(feature = "allocative")]
+pub mod flamegraph;


Every other module's public functions are re-exported at the crate root (memory::*, monitor::MetricsMonitor, setup::*). flamegraph is the exception — callers must use jolt_profiling::flamegraph::write_flamegraph_svg. Consider adding:

#[cfg(feature = "allocative")] pub use flamegraph::{print_data_structure_heap_usage, write_flamegraph_svg};

0xAndoroid · 2026-04-03T21:07:40Z

crates/jolt-profiling/src/monitor.rs

+                while !stop.load(Ordering::Relaxed) {
+                    system.refresh_all();
+
+                    let memory_gib = memory_stats()
+                        .map(|s| s.physical_mem as f64 / BYTES_PER_GIB)
+                        .unwrap_or(0.0);
+                    let cpu_percent = system.global_cpu_usage();
+                    let cores_active_avg = cpu_percent / 100.0 * (system.cpus().len() as f32);
+                    let active_cores = system
+                        .cpus()
+                        .iter()
+                        .filter(|cpu| cpu.cpu_usage() > 0.1)
+                        .count();
+
+                    #[cfg(target_os = "linux")]
+                    let active_threads = std::fs::read_dir("/proc/self/task")
+                        .map(|entries| entries.count())
+                        .unwrap_or(0);
+
+                    #[cfg(not(target_os = "linux"))]
+                    let active_threads = 0_usize;
+
+                    tracing::debug!(
+                        counters.memory_gib = memory_gib,
+                        counters.cpu_percent = cpu_percent,
+                        counters.cores_active_avg = cores_active_avg,
+                        counters.cores_active = active_cores,
+                        counters.thread_count = active_threads,
+                    );
+
+                    thread::sleep(interval);
+                }
+
+                tracing::info!("MetricsMonitor stopping");
+            })
+            .expect("Failed to spawn metrics monitor thread");
+
+        MetricsMonitor {
+            handle: Some(handle),
+            stop_flag,
+        }
+    }
+}
+
+impl Drop for MetricsMonitor {
+    fn drop(&mut self) {
+        self.stop_flag.store(true, Ordering::Relaxed);


Ordering::Relaxed on both the load (line 49) and store (line 95) is technically safe — join() provides the final synchronization — but on weakly-ordered architectures the thread may run extra iterations after Drop signals stop. Standard pattern for stop flags is Release (store) / Acquire (load). Minor, but it's a one-word change.

github-actions · 2026-04-03T21:07:51Z

Benchmark comparison (crates)

No baseline found — skipping comparison.

This was referenced Mar 25, 2026

chore: workspace scaffolding for modular crates #1362

Merged

feat: add jolt-field crate #1363

Merged

This was referenced Mar 25, 2026

feat: add jolt-transcript crate #1365

Open

feat: add jolt-instructions crate #1366

Open

feat: add jolt-poly crate #1367

Draft

feat: add jolt-crypto crate #1368

Draft

feat: add jolt-host crate #1369

Draft

markosg04 marked this pull request as ready for review April 2, 2026 15:33

markosg04 requested review from 0xAndoroid and moodlezoup as code owners April 2, 2026 15:33

0xAndoroid force-pushed the jolt-v2/jolt-field branch from 74305c7 to 17e1d5d Compare April 3, 2026 18:38

0xAndoroid requested a review from sagar-a16z as a code owner April 3, 2026 18:38

0xAndoroid force-pushed the jolt-v2/jolt-field branch 3 times, most recently from 15bbf97 to 327de42 Compare April 3, 2026 19:21

Base automatically changed from jolt-v2/jolt-field to main April 3, 2026 20:51

markosg04 added 2 commits April 3, 2026 16:52

chore: workspace scaffolding for modular crates

eeeabac

feat: add jolt-profiling crate

240e60c

0xAndoroid force-pushed the jolt-v2/jolt-profiling branch from 0a58fef to 240e60c Compare April 3, 2026 20:54

github-actions bot added the no-spec PR has no spec file label Apr 3, 2026

fix: align jolt-profiling with workspace lints after rebase

b003fba

Remove duplicate [workspace.lints.clippy] section introduced by rebase, add workspace lint inheritance to jolt-profiling, bump sysinfo to workspace version, and fix all lint violations.

0xAndoroid reviewed Apr 3, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: add jolt-profiling crate#1364

feat: add jolt-profiling crate#1364
markosg04 wants to merge 3 commits intomainfrom
jolt-v2/jolt-profiling

markosg04 commented Mar 25, 2026

Uh oh!

markosg04 commented Mar 25, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Apr 3, 2026

Uh oh!

0xAndoroid left a comment

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

0xAndoroid Apr 3, 2026

Uh oh!

github-actions bot commented Apr 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	opts.factor = 1.0 / BYTES_PER_GIB * 1024.0;
	opts.factor = 1.0 / BYTES_PER_MIB;

	let interval = Duration::from_millis((interval_secs * 1000.0) as u64);
	let interval = Duration::from_millis(((interval_secs * 1000.0) as u64).max(50));

Conversation

markosg04 commented Mar 25, 2026

Summary

Changes

Testing

Security Considerations

Breaking Changes

Uh oh!

markosg04 commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Apr 3, 2026

Uh oh!

0xAndoroid left a comment

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

0xAndoroid Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Apr 3, 2026

Benchmark comparison (crates)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

markosg04 commented Mar 25, 2026 •

edited

Loading