An application flight recorder for Rust: structured execution history, scoped outcomes, and logging views.
eventline records structured execution history (events + scopes) first. Console logs, text files, JSONL files, and replay helpers are views over that journal.
- Append-only journal of structured records (events + scope exits)
- RAII scopes with duration tracking and success/failure/aborted outcomes
- Per-outcome exit messages for clean
done:lines - Emission gating (sink log levels) that does NOT affect recording
- Structured fields on events
- JSONL rendering/export for machine-readable journals
- Console and file sinks (toggleable, configurable formatting)
- Automatic log rotation with configurable size limits and backup retention
- Thread-local runtime scope stacks so concurrent threads do not share active scope state
- Dropped async-writer record counters and observable writer setup errors
- Replay helpers for scope trees, failed scopes, slowest scopes, and filtered records
- Optional
logandtracingbridges behind feature flags - Run headers written as raw bytes at the start of each log session
eventline is a small, predictable execution journal built for real systems software. It should be treated as a local flight recorder: the journal answers "what happened?", while logging output answers "what should I show right now?"
Core rule:
Record structured history independently from output sinks. Only gate emission (console/file output) with log levels and sink toggles.
This makes eventline useful for post-mortems, debugging, and deterministic "what happened?" replay without turning every debug line into console noise.
Add it to your project:
[dependencies]
eventline = "0.9.0"
Initialize once early:
eventline::init_sync();
eventline::runtime::init().await is still available when async startup code prefers that shape.
Log events:
eventline::info!("starting");
eventline::debug!("debug details: {}", 123);
eventline::warn!("something looks off");
eventline::error!("something failed: {}", "oops");
Attach structured fields:
eventline::info!("user login", user_id = 42, method = "oauth");
Create a scope:
eventline::scope!("config", {
eventline::info!("loading config");
// work...
});
Scopes with exit messages:
eventline::scope!(
"config",
success="loaded",
failure="failed",
aborted="aborted",
{
// work...
});
This produces a final line like:
done: config#12 loaded [success] (3.2ms)
Default text output uses a calm, aligned lowercase style:
debug: loading config {path="config.toml"}
info: server listening {addr="127.0.0.1:8080"}
warn: retrying request {attempt=2}
error: request failed {status=500}
done: request#12 served [success] (18.4ms)
(Exact formatting depends on runtime settings.)
- Recording happens independently from output sinks — by default all built-in levels are recorded.
- Emission is controlled by per-sink log levels and enabled sinks.
You get full fidelity history without spamming stdout.
If you need to reduce formatting/recording overhead, set an explicit recording threshold:
eventline::set_record_level(eventline::LogLevel::Info);
The in-memory runtime journal is bounded, so long-running processes should use a file sink or JSONL export if they need durable history.
Set runtime journal retention:
eventline::set_journal_retention(10_000);
Console output:
eventline::runtime::enable_console_output(true);
eventline::runtime::enable_console_color(true);
eventline::runtime::enable_console_scope_labels(false);
eventline::runtime::enable_console_scope_exits(false);
eventline::runtime::enable_console_timestamp(false);
eventline::runtime::enable_console_duration(false);
Per-sink levels:
eventline::runtime::set_console_level(eventline::runtime::LogLevel::Info);
eventline::runtime::set_file_level(eventline::runtime::LogLevel::Debug);
File output (append, no rotation):
eventline::runtime::enable_file_output("/tmp/app.log")?;
JSONL file output:
eventline::runtime::enable_file_output_jsonl("/tmp/app.jsonl")?;
Export the current in-memory journal as JSONL lines:
let lines = eventline::records_jsonl();
File output with automatic rotation:
use eventline::{LogPolicy, RunHeader};
eventline::runtime::enable_file_output_rotating(
"logs/app.log",
LogPolicy::default(), // 5 MiB max, 5 backups
Some(RunHeader::new("my-app")), // with PID annotation
)?;
Rotation renames the active log to app.log.1, shifts older backups up, and
opens a fresh app.log. The oldest backup beyond keep_backups is silently
dropped.
Disable all output (still records):
eventline::runtime::disable_all_output();
Set global output fast-path level (normally derived from enabled sinks):
eventline::runtime::set_log_level(eventline::runtime::LogLevel::Debug);
Inspect writer health:
let dropped = eventline::dropped_writer_records();
let last_error = eventline::last_writer_error();
LogPolicy controls when and how rotation happens:
use eventline::LogPolicy;
// Custom: 10 MiB max, keep 3 backups
let policy = LogPolicy::new(10 * 1024 * 1024, 3);
// Or use the defaults (5 MiB, 5 backups)
let policy = LogPolicy::default();
Rotation is triggered automatically before a record is written that would push
the file past max_bytes. Set keep_backups to 0 to simply delete the log
on rotation rather than keeping any history.
The journal can be inspected directly:
let records = eventline::records();
let scopes = eventline::scopes();
let tree = eventline::replay::render_scope_tree(&scopes);
let failed = eventline::replay::failed_scopes(&records, &scopes);
let slowest = eventline::replay::slowest_scopes(&records, &scopes, 10);
This is the main difference between eventline and ordinary logging: log output is only one way to view the execution history.
Core eventline stays dependency-light. Existing instrumentation can be bridged with optional features:
[dependencies]
eventline = { version = "0.9.0", features = ["log", "tracing"] }
log facade bridge:
eventline::integrations::log::init(log::LevelFilter::Debug)?;
tracing event bridge:
use tracing_subscriber::prelude::*;
let subscriber = tracing_subscriber::registry()
.with(eventline::integrations::tracing::EventlineLayer);
tracing::subscriber::set_global_default(subscriber)?;
The tracing layer currently records tracing events into eventline. Native eventline scopes remain the stronger scoped-outcome model.
A RunHeader is a single decorated line written as raw bytes at the start of a
log session, before structured records begin. It is always visible in the file
regardless of log level.
use eventline::runtime::RunHeader;
// With PID:
// ==================== my-daemon (pid=18432) ====================
RunHeader::new("my-daemon")
// Without PID:
// ====================== my-daemon ======================
RunHeader::without_pid("my-daemon")
// Custom width:
RunHeader::new("my-daemon").with_width(80)
When appending to an existing non-empty log file, a blank separator line is inserted automatically before the header so run boundaries are clearly visible.
- Records are append-only and never rewritten
- Scopes are created on enter and finalized exactly once on exit
- The journal is the canonical execution history
- Rendering and runtime only affect presentation, not data
- Runtime scope context is thread-local
- In-memory retention is bounded and configurable
- Writer queues are bounded; dropped output records are counted
- Rotation and headers operate on raw bytes and bypass the record pipeline intentionally
core/— data types only (Record, Scope, ids, Outcome, guards)journal/— append-only buffer, writers, rotation logic, structured fieldsrender/— canonical formatting for console and file outputruntime/— global config, sinks, filtering policy, run headersreplay— replay helpers for scope trees and summariesintegrations— optionallogandtracingbridgesmacros—info!,debug!,warn!,error!,scope!
MIT