feat: replace jemalloc with mimalloc (#4174)

* feat: replace jemalloc with mimalloc

* docs: record allocator rounds5 retest

* fix(replication): satisfy clippy unwrap lints

* docs: keep allocator migration plan local only

* feat(profiling): rely on pyroscope cpu profiling

* refactor(profiling): centralize unsupported pprof responses

* chore(deps): update s3s revision
This commit is contained in:
houseme
2026-07-02 19:03:38 +08:00
committed by GitHub
parent 54496f2796
commit 70e1d79dfd
17 changed files with 217 additions and 1529 deletions
+1 -2
View File
@@ -28,7 +28,7 @@ documentation = "https://docs.rs/rustfs-obs/latest/rustfs_obs/"
[features]
default = []
gpu = ["dep:nvml-wrapper"]
pyroscope = ["dep:jemalloc_pprof", "dep:pyroscope"]
pyroscope = ["dep:pyroscope"]
[lints]
workspace = true
@@ -83,7 +83,6 @@ nvml-wrapper = { workspace = true, optional = true }
[target.'cfg(any(target_os = "macos", all(target_os = "linux", target_env = "gnu", target_arch = "x86_64")))'.dependencies]
pyroscope = { workspace = true, features = ["backend-pprof-rs"], optional = true }
jemalloc_pprof = { workspace = true, optional = true }
[dev-dependencies]
-30
View File
@@ -39,10 +39,6 @@ pub(crate) type ProfilingAgent = pyroscope::PyroscopeAgent<pyroscope::pyroscope:
any(target_os = "macos", all(target_os = "linux", target_env = "gnu", target_arch = "x86_64"))
)))]
pub(crate) type ProfilingAgent = ();
#[cfg(all(feature = "pyroscope", target_os = "linux", target_env = "gnu", target_arch = "x86_64"))]
pub(crate) type MemoryProfilingAgent = pyroscope::PyroscopeAgent<pyroscope::pyroscope::PyroscopeAgentRunning>;
#[cfg(not(all(feature = "pyroscope", target_os = "linux", target_env = "gnu", target_arch = "x86_64")))]
pub(crate) type MemoryProfilingAgent = ();
const LOG_COMPONENT_OBS: &str = "obs";
const LOG_SUBSYSTEM_GUARD: &str = "guard";
@@ -67,7 +63,6 @@ pub struct OtelGuard {
/// Optional logger provider for OTLP log export.
pub(crate) logger_provider: Option<SdkLoggerProvider>,
pub(crate) profiling_agent: Option<ProfilingAgent>,
pub(crate) memory_profiling_agent: Option<MemoryProfilingAgent>,
/// Handle to the background log-cleanup task; aborted on drop.
pub(crate) cleanup_handle: Option<tokio::task::JoinHandle<()>>,
/// Worker guard that keeps the non-blocking `tracing_appender` thread
@@ -84,7 +79,6 @@ impl std::fmt::Debug for OtelGuard {
.field("meter_provider", &self.meter_provider.is_some())
.field("logger_provider", &self.logger_provider.is_some())
.field("profiling_agent", &self.profiling_agent.is_some())
.field("memory_profiling_agent", &self.memory_profiling_agent.is_some())
.field("cleanup_handle", &self.cleanup_handle.is_some())
.field("tracing_guard", &self.tracing_guard.is_some())
.field("stdout_guard", &self.stdout_guard.is_some())
@@ -161,30 +155,6 @@ impl Drop for OtelGuard {
}
}
#[cfg(all(feature = "pyroscope", target_os = "linux", target_env = "gnu", target_arch = "x86_64"))]
if let Some(agent) = self.memory_profiling_agent.take() {
match agent.stop() {
Err(err) => {
if tracing::dispatcher::has_been_set() {
error!(
event = EVENT_OBS_GUARD_SHUTDOWN,
component = LOG_COMPONENT_OBS,
subsystem = LOG_SUBSYSTEM_GUARD,
resource = "memory_profiling_agent",
result = "shutdown_failed",
error = %err,
"observability guard shutdown failed"
);
} else {
eprintln!("{}", format_guard_shutdown_stderr_message("memory_profiling_agent", err));
}
}
Ok(stopped) => {
stopped.shutdown();
}
}
}
if let Some(handle) = self.cleanup_handle.take() {
debug!(
event = EVENT_OBS_GUARD_SHUTDOWN,
-2
View File
@@ -271,7 +271,6 @@ fn init_stdout_only(_config: &OtelConfig, logger_level: &str, is_production: boo
meter_provider: None,
logger_provider: None,
profiling_agent: None,
memory_profiling_agent: None,
tracing_guard: Some(guard),
stdout_guard: None,
cleanup_handle: None,
@@ -389,7 +388,6 @@ fn init_file_logging_internal(
meter_provider: None,
logger_provider: None,
profiling_agent: None,
memory_profiling_agent: None,
tracing_guard: Some(guard),
stdout_guard,
cleanup_handle: Some(cleanup_handle),
+1 -62
View File
@@ -40,7 +40,7 @@ use crate::cleaner::types::FileMatchMode;
use crate::config::OtelConfig;
use crate::global::set_observability_metric_enabled;
use crate::telemetry::filter::build_env_filter;
use crate::telemetry::guard::{MemoryProfilingAgent, OtelGuard, ProfilingAgent};
use crate::telemetry::guard::{OtelGuard, ProfilingAgent};
use crate::telemetry::local::{build_json_log_layer, spawn_cleanup_task};
use crate::telemetry::recorder::Recorder;
use crate::telemetry::resource::build_resource;
@@ -173,7 +173,6 @@ pub(super) fn init_observability_http(
let meter_provider = build_meter_provider(&metric_ep, config, res.clone(), &service_name, use_stdout)?;
let profiling_agent = init_profiler(config);
let memory_profiling_agent = init_memory_profiler(config);
// ── Logger Logic ──────────────────────────────────────────────────────────
// Logging is the only signal that may intentionally route to either OTLP
@@ -317,7 +316,6 @@ pub(super) fn init_observability_http(
meter_provider,
logger_provider,
profiling_agent,
memory_profiling_agent,
tracing_guard,
stdout_guard,
cleanup_handle,
@@ -563,65 +561,6 @@ fn init_profiler(_config: &OtelConfig) -> Option<ProfilingAgent> {
None
}
/// Initialise a Pyroscope agent for continuous **memory** profiling via jemalloc.
///
/// This is only available on `linux + gnu + x86_64` where tikv-jemallocator
/// is the global allocator and `jemalloc_pprof::PROF_CTL` is accessible.
///
/// Returns `None` when profiling export is disabled, the endpoint is missing,
/// jemalloc profiling is not activated, or the agent fails to build/start.
#[cfg(all(feature = "pyroscope", target_os = "linux", target_env = "gnu", target_arch = "x86_64"))]
fn init_memory_profiler(config: &OtelConfig) -> Option<MemoryProfilingAgent> {
use pyroscope::backend::jemalloc_backend;
use pyroscope::pyroscope::PyroscopeAgentBuilder;
use rustfs_config::VERSION;
if !config
.profiling_export_enabled
.unwrap_or(rustfs_config::DEFAULT_OBS_PROFILING_EXPORT_ENABLED)
{
return None;
}
let endpoint = config.profiling_endpoint.as_ref()?.as_str();
if endpoint.is_empty() {
return None;
}
// Verify jemalloc profiling is available and activated
{
let prof_ctl = jemalloc_pprof::PROF_CTL.as_ref()?;
let ctl = prof_ctl.try_lock().ok()?;
if !ctl.activated() {
eprintln!("Memory profiling skipped: jemalloc profiling is not activated");
return None;
}
}
let backend = jemalloc_backend();
let service_name = config.service_name.as_deref().unwrap_or(APP_NAME);
let version = config.service_version.as_deref().unwrap_or(VERSION);
let sample_rate = 100;
let agent = PyroscopeAgentBuilder::new(endpoint, service_name, sample_rate, "pyroscope-rs", "1.0.1", backend)
.tags(vec![("version", version), ("profile_type", "memory")])
.build()
.ok()?;
match agent.start() {
Ok(agent) => Some(agent),
Err(err) => {
eprintln!("Memory profiling agent start error: {err:?}");
None
}
}
}
#[cfg(not(all(feature = "pyroscope", target_os = "linux", target_env = "gnu", target_arch = "x86_64")))]
fn init_memory_profiler(_config: &OtelConfig) -> Option<MemoryProfilingAgent> {
None
}
/// Create a stdout periodic metrics reader for the given interval.
///
/// This helper is primarily used for local development and diagnostics when