gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-21 20:06:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor(logging): standardize protocol and observability events (#3419)

Browse Source 
* refactor(logging): standardize object capacity events

* refactor(logging): standardize protocol server events

* refactor(logging): standardize swift protocol events

* refactor(logging): standardize observability events

* refactor(logging): move masking helper and extend guardrails
This commit is contained in:
houseme
2026-06-14 07:14:45 +08:00
committed by GitHub
parent 22460243bf
commit efa89a98ed
48 changed files with 2976 additions and 434 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/obs/src/telemetry/dial9.rs
+49 -7
View File
@@ -39,6 +39,10 @@ use rustfs_utils::get_env_usize;
use std::path::PathBuf;
use tracing::{info, warn};
const LOG_COMPONENT_OBS: &str = "obs";
const LOG_SUBSYSTEM_DIAL9: &str = "dial9";
const EVENT_DIAL9_STATE: &str = "dial9_state";
/// Configuration for dial9 Tokio telemetry.
#[derive(Debug, Clone)]
pub struct Dial9Config {
@@ -131,25 +135,51 @@ impl Dial9SessionGuard {
/// Returns `Ok(None)` if dial9 is disabled.
pub async fn new(config: Dial9Config) -> Result<Option<Self>, TelemetryError> {
if !config.enabled {
info!("Dial9 telemetry disabled");
info!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
state = "disabled",
"dial9 state changed"
);
return Ok(None);
}
info!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
state = "validating",
output_dir = %config.output_dir,
file_prefix = %config.file_prefix,
sampling_rate = config.sampling_rate,
"Validating dial9 telemetry configuration"
"dial9 state changed"
);
// Only create directory; writer will be created in build_traced_runtime
if let Err(e) = tokio::fs::create_dir_all(&config.output_dir).await {
warn!("Failed to create dial9 output directory '{}': {}", config.output_dir, e);
warn!("Continuing without dial9 telemetry");
warn!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
result = "output_dir_create_failed",
output_dir = %config.output_dir,
error = %e,
fallback = "disabled",
"dial9 state changed"
);
return Ok(None);
}
info!("Dial9 telemetry configuration validated successfully");
info!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
state = "validated",
output_dir = %config.output_dir,
file_prefix = %config.file_prefix,
"dial9 state changed"
);
Ok(Some(Self { _guard: None, config }))
}
@@ -168,7 +198,13 @@ impl Dial9SessionGuard {
/// Flush any pending telemetry data.
pub async fn shutdown(&self) {
if let Some(_guard) = &self._guard {
info!("Dial9 telemetry data will be flushed on drop");
info!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
state = "shutdown_requested",
"dial9 state changed"
);
// TelemetryGuard handles flushing automatically when dropped
}
}
@@ -178,7 +214,13 @@ impl Drop for Dial9SessionGuard {
fn drop(&mut self) {
if let Some(_guard) = &self._guard {
// TelemetryGuard flushes automatically when dropped
info!("Dial9 telemetry guard dropped, data flushed");
info!(
event = EVENT_DIAL9_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_DIAL9,
state = "flushed",
"dial9 state changed"
);
}
}
}
crates/obs/src/telemetry/local.rs
+36 -5
View File
@@ -60,6 +60,11 @@ use tracing_subscriber::{
util::SubscriberInitExt,
};
const LOG_COMPONENT_OBS: &str = "obs";
const LOG_SUBSYSTEM_LOCAL_LOGGING: &str = "local_logging";
const EVENT_LOCAL_LOGGING_STATE: &str = "local_logging_state";
const EVENT_LOG_CLEANER_STATE: &str = "log_cleaner_state";
pub(super) fn build_json_log_layer<S, W>(writer: W, enable_ansi: bool, span_events: FmtSpan) -> impl tracing_subscriber::Layer<S>
where
S: Subscriber + for<'span> LookupSpan<'span>,
@@ -161,12 +166,16 @@ fn init_stdout_only(_config: &OtelConfig, logger_level: &str, is_production: boo
set_observability_metric_enabled(false);
counter!("rustfs_start_total").increment(1);
info!(
event = EVENT_LOCAL_LOGGING_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_LOCAL_LOGGING,
state = "initialized",
backend = "local",
sink = "stdout",
output_format = "json",
logger_level,
is_production,
"Initialized local logging"
"local logging state changed"
);
OtelGuard {
@@ -271,6 +280,10 @@ fn init_file_logging_internal(
let cleanup_handle = spawn_cleanup_task(config, log_directory, log_filename, keep_files);
info!(
event = EVENT_LOCAL_LOGGING_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_LOCAL_LOGGING,
state = "initialized",
backend = "local",
sink = "file",
output_format = "json",
@@ -280,7 +293,7 @@ fn init_file_logging_internal(
stdout_mirror_enabled = stdout_guard.is_some(),
is_production,
logger_level,
"Initialized local logging"
"local logging state changed"
);
Ok(OtelGuard {
@@ -459,13 +472,17 @@ pub fn spawn_cleanup_task(
);
info!(
event = EVENT_LOG_CLEANER_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_LOCAL_LOGGING,
state = "configured",
compression_algorithm = %compression_algorithm,
parallel_compress,
parallel_workers,
zstd_level,
zstd_fallback_to_gzip,
zstd_workers,
"log cleaner compression profile configured"
"log cleaner state changed"
);
tokio::spawn(async move {
@@ -483,11 +500,25 @@ pub fn spawn_cleanup_task(
}
Ok(Err(e)) => {
counter!(METRIC_LOG_CLEANER_RUN_FAILURES_TOTAL).increment(1);
tracing::warn!("Log cleanup failed: {}", e);
tracing::warn!(
event = EVENT_LOG_CLEANER_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_LOCAL_LOGGING,
result = "cleanup_failed",
error = %e,
"log cleaner state changed"
);
}
Err(e) => {
counter!(METRIC_LOG_CLEANER_RUN_FAILURES_TOTAL).increment(1);
tracing::warn!("Log cleanup task panicked: {}", e);
tracing::warn!(
event = EVENT_LOG_CLEANER_STATE,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_LOCAL_LOGGING,
result = "cleanup_task_panicked",
error = %e,
"log cleaner state changed"
);
}
}
}
crates/obs/src/telemetry/recorder.rs
+7 -3
View File
@@ -30,6 +30,10 @@ use std::{
};
use tracing::error;
const LOG_COMPONENT_OBS: &str = "obs";
const LOG_SUBSYSTEM_RECORDER: &str = "recorder";
const EVENT_RECORDER_STATE: &str = "recorder_state";
macro_rules! configure_builder {
($builder:expr, $metadata:expr) => {{
let mut builder = $builder;
@@ -151,7 +155,7 @@ impl Recorder {
let cache = match lock.read() {
Ok(g) => g,
Err(e) => {
error!("{} cache read lock poisoned: {}", metric_type, e);
error!(event = EVENT_RECORDER_STATE, component = LOG_COMPONENT_OBS, subsystem = LOG_SUBSYSTEM_RECORDER, metric_type = %metric_type, result = "cache_read_lock_poisoned", error = %e, "recorder state changed");
e.into_inner()
}
};
@@ -162,7 +166,7 @@ impl Recorder {
let mut cache = match lock.write() {
Ok(g) => g,
Err(e) => {
error!("{} cache write lock poisoned: {}", metric_type, e);
error!(event = EVENT_RECORDER_STATE, component = LOG_COMPONENT_OBS, subsystem = LOG_SUBSYSTEM_RECORDER, metric_type = %metric_type, result = "cache_write_lock_poisoned", error = %e, "recorder state changed");
e.into_inner()
}
};
@@ -179,7 +183,7 @@ impl Recorder {
F: FnOnce(&mut HashMap<KeyName, MetricMetadata>) -> R,
{
let mut guard = self.metrics_metadata.lock().unwrap_or_else(|e| {
error!("metrics_metadata lock poisoned: {}", e);
error!(event = EVENT_RECORDER_STATE, component = LOG_COMPONENT_OBS, subsystem = LOG_SUBSYSTEM_RECORDER, result = "metrics_metadata_lock_poisoned", error = %e, "recorder state changed");
e.into_inner()
});
f(&mut guard)