Cache process/thread id getters and use them in TError origin capture

## Motivation
Profiling the YT master Automaton thread showed TOriginAttributes::Capture (run on every non-OK TError) spending ~60% of its time in a getpid() syscall — uncached on glibc >= 2.25. NYT::GetCurrentThreadId() (gettid) feeds hot thread-affinity / log-manager checks on the same thread.

## Changes
- New library/cpp/yt/system/process_id.* with cached GetProcessId(); GetSystemThreadId() now caches the kernel tid in TLS. Both caches reset in the child after fork.
- Moved thread_name.{h,cpp} from misc to system.
- Removed GetCurrentProcessId/GetCurrentThreadId shims from yt/yt/core/misc/proc.{h,cpp}; migrated all callers to NYT::GetProcessId / NYT::GetSystemThreadId.
- TOriginAttributes::Capture uses the cached getters; recorded Tid is now the real kernel tid (matches perf/ps).
- Added microbenchmarks (library/cpp/yt/system/benchmarks, yt/yt/core/benchmarks/error.cpp).

## Microbenchmarks (release)
| | before | after |
|---|---|---|
| getpid | 101 ns | 0.33 ns |
| gettid | 102 ns | 1.64 ns |
| Capture | 161 ns | 50 ns |
| failed TError | 221 ns | 74 ns |
commit_hash:ee37ae57d61a5a2dd33daee935270f4bb93b7ff9

19 files changed, 261 insertions, 10 deletions

diff --git a/library/cpp/yt/error/origin_attributes.cpp b/library/cpp/yt/error/origin_attributes.cpp
index 18582bf3e83..2198912d419 100644
--- a/library/cpp/yt/error/origin_attributes.cpp
+++ b/library/cpp/yt/error/origin_attributes.cpp
@@ -3,12 +3,13 @@
 
 #include <library/cpp/yt/assert/assert.h>
 
-#include <library/cpp/yt/misc/thread_name.h>
 #include <library/cpp/yt/misc/tls.h>
 
 #include <library/cpp/yt/string/format.h>
 
-#include <util/system/thread.h>
+#include <library/cpp/yt/system/process_id.h>
+#include <library/cpp/yt/system/thread_id.h>
+#include <library/cpp/yt/system/thread_name.h>
 
 namespace NYT {
 
@@ -65,8 +66,8 @@ void TOriginAttributes::Capture()
     }
 
     Datetime = TInstant::Now();
-    Pid = GetPID();
-    Tid = TThread::CurrentThreadId();
+    Pid = GetProcessId();
+    Tid = GetSystemThreadId();
     ThreadName = GetCurrentThreadName();
     ExtensionData = NDetail::GetExtensionData();
 }
diff --git a/library/cpp/yt/error/origin_attributes.h b/library/cpp/yt/error/origin_attributes.h
index 8804cdf2569..335e2110dcc 100644
--- a/library/cpp/yt/error/origin_attributes.h
+++ b/library/cpp/yt/error/origin_attributes.h
@@ -7,7 +7,8 @@
 #include <library/cpp/yt/memory/ref.h>
 
 #include <library/cpp/yt/misc/guid.h>
-#include <library/cpp/yt/misc/thread_name.h>
+
+#include <library/cpp/yt/system/thread_name.h>
 
 #include <library/cpp/yt/threading/public.h>
 
diff --git a/library/cpp/yt/error/ya.make b/library/cpp/yt/error/ya.make
index 3fe4f88bed7..716af327361 100644
--- a/library/cpp/yt/error/ya.make
+++ b/library/cpp/yt/error/ya.make
@@ -8,6 +8,7 @@ PEERDIR(
     library/cpp/yt/global
     library/cpp/yt/memory
     library/cpp/yt/misc
+    library/cpp/yt/system
     library/cpp/yt/threading
     library/cpp/yt/string
     library/cpp/yt/logging # TODO(arkady-e1ppa): Consider logging error_code crashes to stderr and drop this dep.
diff --git a/library/cpp/yt/logging/logger.cpp b/library/cpp/yt/logging/logger.cpp
index e0af30382d6..458d01d8133 100644
--- a/library/cpp/yt/logging/logger.cpp
+++ b/library/cpp/yt/logging/logger.cpp
@@ -4,7 +4,7 @@
 
 #include <library/cpp/yt/cpu_clock/clock.h>
 
-#include <library/cpp/yt/misc/thread_name.h>
+#include <library/cpp/yt/system/thread_name.h>
 
 #include <util/system/compiler.h>
 #include <util/system/thread.h>
diff --git a/library/cpp/yt/logging/logger.h b/library/cpp/yt/logging/logger.h
index e3e6f58d5a3..4cb6021e53f 100644
--- a/library/cpp/yt/logging/logger.h
+++ b/library/cpp/yt/logging/logger.h
@@ -12,7 +12,7 @@
 
 #include <library/cpp/yt/misc/guid.h>
 
-#include <library/cpp/yt/misc/thread_name.h>
+#include <library/cpp/yt/system/thread_name.h>
 
 #include <library/cpp/yt/memory/leaky_singleton.h>
 
diff --git a/library/cpp/yt/logging/ya.make b/library/cpp/yt/logging/ya.make
index e611c2e554c..8b338ed3b77 100644
--- a/library/cpp/yt/logging/ya.make
+++ b/library/cpp/yt/logging/ya.make
@@ -10,6 +10,7 @@ PEERDIR(
     library/cpp/yt/assert
     library/cpp/yt/memory
     library/cpp/yt/misc
+    library/cpp/yt/system
     library/cpp/yt/yson_string
 )
 
diff --git a/library/cpp/yt/misc/ya.make b/library/cpp/yt/misc/ya.make
index 3135887e95e..5d2d6d67f33 100644
--- a/library/cpp/yt/misc/ya.make
+++ b/library/cpp/yt/misc/ya.make
@@ -5,7 +5,6 @@ INCLUDE(${ARCADIA_ROOT}/library/cpp/yt/ya_cpp.make.inc)
 SRCS(
     guid.cpp
     source_location.cpp
-    thread_name.cpp
 )
 
 PEERDIR(
diff --git a/library/cpp/yt/system/benchmarks/process.cpp b/library/cpp/yt/system/benchmarks/process.cpp
new file mode 100644
index 00000000000..74b4b6b1e13
--- /dev/null
+++ b/library/cpp/yt/system/benchmarks/process.cpp
@@ -0,0 +1,35 @@
+#include <benchmark/benchmark.h>
+
+#include <library/cpp/yt/system/process_id.h>
+
+#include <util/system/getpid.h>
+
+namespace NYT {
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////
+
+void BM_GetProcessId(benchmark::State& state)
+{
+    // Cached getpid: only the first call hits the kernel.
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(GetProcessId());
+    }
+}
+
+BENCHMARK(BM_GetProcessId);
+
+void BM_RawGetPid(benchmark::State& state)
+{
+    // Uncached getpid syscall (uncached on glibc >= 2.25), for comparison.
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(::GetPID());
+    }
+}
+
+BENCHMARK(BM_RawGetPid);
+
+////////////////////////////////////////////////////////////////////////////////
+
+} // namespace
+} // namespace NYT
diff --git a/library/cpp/yt/system/benchmarks/thread.cpp b/library/cpp/yt/system/benchmarks/thread.cpp
new file mode 100644
index 00000000000..5aa5f7dc2d9
--- /dev/null
+++ b/library/cpp/yt/system/benchmarks/thread.cpp
@@ -0,0 +1,55 @@
+#include <benchmark/benchmark.h>
+
+#include <library/cpp/yt/system/thread_id.h>
+#include <library/cpp/yt/system/thread_name.h>
+
+#include <util/system/thread.h>
+
+namespace NYT {
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////
+
+void BM_GetSystemThreadId(benchmark::State& state)
+{
+    // Cached gettid: only the first call per thread hits the kernel.
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(GetSystemThreadId());
+    }
+}
+
+BENCHMARK(BM_GetSystemThreadId);
+
+void BM_GetSequentialThreadId(benchmark::State& state)
+{
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(GetSequentialThreadId());
+    }
+}
+
+BENCHMARK(BM_GetSequentialThreadId);
+
+void BM_RawGetTid(benchmark::State& state)
+{
+    // Uncached gettid syscall, for comparison.
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(::TThread::CurrentThreadNumericId());
+    }
+}
+
+BENCHMARK(BM_RawGetTid);
+
+void BM_GetCurrentThreadName(benchmark::State& state)
+{
+    // TLS-cached thread name (also read by TOriginAttributes::Capture).
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(GetCurrentThreadName());
+    }
+}
+
+BENCHMARK(BM_GetCurrentThreadName);
+
+////////////////////////////////////////////////////////////////////////////////
+
+} // namespace
+} // namespace NYT
diff --git a/library/cpp/yt/system/benchmarks/ya.make b/library/cpp/yt/system/benchmarks/ya.make
new file mode 100644
index 00000000000..874d8df323d
--- /dev/null
+++ b/library/cpp/yt/system/benchmarks/ya.make
@@ -0,0 +1,12 @@
+G_BENCHMARK()
+
+SRCS(
+    process.cpp
+    thread.cpp
+)
+
+PEERDIR(
+    library/cpp/yt/system
+)
+
+END()
diff --git a/library/cpp/yt/system/process_id-inl.h b/library/cpp/yt/system/process_id-inl.h
new file mode 100644
index 00000000000..9f0ef98100e
--- /dev/null
+++ b/library/cpp/yt/system/process_id-inl.h
@@ -0,0 +1,35 @@
+#ifndef PROCESS_ID_INL_H_
+#error "Direct inclusion of this file is not allowed, include process_id.h"
+// For the sake of sane code completion.
+#include "process_id.h"
+#endif
+
+#include <atomic>
+
+#include <util/system/compiler.h>
+
+namespace NYT {
+
+////////////////////////////////////////////////////////////////////////////////
+
+namespace NDetail {
+
+TProcessId GetProcessIdImpl();
+
+} // namespace NDetail
+
+extern std::atomic<TProcessId> CachedProcessId;
+
+inline TProcessId GetProcessId()
+{
+    auto cachedProcessId = CachedProcessId.load(std::memory_order::relaxed);
+    if (cachedProcessId == InvalidProcessId) [[unlikely]] {
+        cachedProcessId = NDetail::GetProcessIdImpl();
+        CachedProcessId.store(cachedProcessId, std::memory_order::relaxed);
+    }
+    return cachedProcessId;
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+} // namespace NYT
diff --git a/library/cpp/yt/system/process_id.cpp b/library/cpp/yt/system/process_id.cpp
new file mode 100644
index 00000000000..0be6dfd2913
--- /dev/null
+++ b/library/cpp/yt/system/process_id.cpp
@@ -0,0 +1,37 @@
+#include "process_id.h"
+
+#ifdef _unix_
+#include <library/cpp/yt/misc/static_initializer.h>
+
+#include <pthread.h>
+#endif
+
+namespace NYT {
+
+////////////////////////////////////////////////////////////////////////////////
+
+std::atomic<TProcessId> CachedProcessId = InvalidProcessId;
+
+namespace NDetail {
+
+TProcessId GetProcessIdImpl()
+{
+    return ::GetPID();
+}
+
+} // namespace NDetail
+
+#ifdef _unix_
+// The pid is stable for the lifetime of a process, so we cache it to avoid the
+// |getpid| syscall on each call. After a |fork|, however, the child runs with a
+// fresh pid, so the cache must be invalidated there.
+YT_STATIC_INITIALIZER(
+    ::pthread_atfork(
+        /*prepare*/ nullptr,
+        /*parent*/  nullptr,
+        /*child*/   [] { CachedProcessId.store(InvalidProcessId, std::memory_order::relaxed); }));
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+
+} // namespace NYT
diff --git a/library/cpp/yt/system/process_id.h b/library/cpp/yt/system/process_id.h
new file mode 100644
index 00000000000..4dcc89b13ab
--- /dev/null
+++ b/library/cpp/yt/system/process_id.h
@@ -0,0 +1,22 @@
+#pragma once
+
+#include <util/system/getpid.h>
+
+namespace NYT {
+
+////////////////////////////////////////////////////////////////////////////////
+
+constexpr TProcessId InvalidProcessId = TProcessId(-1);
+
+//! Returns the OS process id (|getpid|).
+//! The value is cached process-wide, so only the first call hits the kernel.
+//! The cache is reset in the child after |fork|.
+TProcessId GetProcessId();
+
+////////////////////////////////////////////////////////////////////////////////
+
+} // namespace NYT
+
+#define PROCESS_ID_INL_H_
+#include "process_id-inl.h"
+#undef PROCESS_ID_INL_H_
diff --git a/library/cpp/yt/system/thread_id-inl.h b/library/cpp/yt/system/thread_id-inl.h
index 916d70436ed..6139ff863c2 100644
--- a/library/cpp/yt/system/thread_id-inl.h
+++ b/library/cpp/yt/system/thread_id-inl.h
@@ -14,6 +14,25 @@ namespace NYT {
 
 ////////////////////////////////////////////////////////////////////////////////
 
+namespace NDetail {
+
+TSystemThreadId GetSystemThreadIdImpl();
+
+} // namespace NDetail
+
+YT_DECLARE_THREAD_LOCAL(TSystemThreadId, CachedSystemThreadId);
+
+inline TSystemThreadId GetSystemThreadId()
+{
+    auto& cachedSystemThreadId = CachedSystemThreadId();
+    if (Y_UNLIKELY(cachedSystemThreadId == InvalidSystemThreadId)) {
+        cachedSystemThreadId = NDetail::GetSystemThreadIdImpl();
+    }
+    return cachedSystemThreadId;
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
 YT_DECLARE_THREAD_LOCAL(TSequentialThreadId, CachedSequentialThreadId);
 extern std::atomic<TSequentialThreadId> SequentialThreadIdGenerator;
 
diff --git a/library/cpp/yt/system/thread_id.cpp b/library/cpp/yt/system/thread_id.cpp
index a1971dd0dd8..978c735df25 100644
--- a/library/cpp/yt/system/thread_id.cpp
+++ b/library/cpp/yt/system/thread_id.cpp
@@ -2,19 +2,43 @@
 
 #include <util/system/thread.h>
 
+#ifdef _unix_
+#include <library/cpp/yt/misc/static_initializer.h>
+
+#include <pthread.h>
+#endif
+
 namespace NYT {
 
 ////////////////////////////////////////////////////////////////////////////////
 
+YT_DEFINE_THREAD_LOCAL(TSystemThreadId, CachedSystemThreadId, InvalidSystemThreadId);
+
 YT_DEFINE_THREAD_LOCAL(TSequentialThreadId, CachedSequentialThreadId, InvalidSequentialThreadId);
 std::atomic<TSequentialThreadId> SequentialThreadIdGenerator = InvalidSequentialThreadId;
 
-TSystemThreadId GetSystemThreadId()
+namespace NDetail {
+
+TSystemThreadId GetSystemThreadIdImpl()
 {
     static_assert(std::is_same_v<TSystemThreadId, ::TThread::TId>);
     return ::TThread::CurrentThreadNumericId();
 }
 
+} // namespace NDetail
+
+#ifdef _unix_
+// The kernel tid is stable for the lifetime of a thread, so we cache it to
+// avoid the |gettid| syscall on each call. After a |fork|, however, the
+// surviving (calling) thread of the child gets a fresh kernel tid, so the
+// cache must be invalidated there.
+YT_STATIC_INITIALIZER(
+    ::pthread_atfork(
+        /*prepare*/ nullptr,
+        /*parent*/  nullptr,
+        /*child*/   [] { CachedSystemThreadId() = InvalidSystemThreadId; }));
+#endif
+
 ////////////////////////////////////////////////////////////////////////////////
 
-} // namespace NYT::NThreading
+} // namespace NYT
diff --git a/library/cpp/yt/system/thread_id.h b/library/cpp/yt/system/thread_id.h
index 1f75067015d..75960d036e1 100644
--- a/library/cpp/yt/system/thread_id.h
+++ b/library/cpp/yt/system/thread_id.h
@@ -10,6 +10,9 @@ namespace NYT {
 
 using TSystemThreadId = size_t;
 constexpr TSystemThreadId InvalidSystemThreadId = Max<TSystemThreadId>();
+//! Returns the OS thread id (e.g. |gettid| on Linux).
+//! The value is cached in TLS, so only the first call per thread hits the kernel.
+//! The cache is reset in the child after |fork|.
 TSystemThreadId GetSystemThreadId();
 
 using TSequentialThreadId = ui32;
diff --git a/library/cpp/yt/misc/thread_name.cpp b/library/cpp/yt/system/thread_name.cpp
index 11e20abcf5c..11e20abcf5c 100644
--- a/library/cpp/yt/misc/thread_name.cpp
+++ b/library/cpp/yt/system/thread_name.cpp
diff --git a/library/cpp/yt/misc/thread_name.h b/library/cpp/yt/system/thread_name.h
index 35153880831..35153880831 100644
--- a/library/cpp/yt/misc/thread_name.h
+++ b/library/cpp/yt/system/thread_name.h
diff --git a/library/cpp/yt/system/ya.make b/library/cpp/yt/system/ya.make
index 3f767f78f49..72e12e8a348 100644
--- a/library/cpp/yt/system/ya.make
+++ b/library/cpp/yt/system/ya.make
@@ -5,7 +5,9 @@ INCLUDE(${ARCADIA_ROOT}/library/cpp/yt/ya_cpp.make.inc)
 SRCS(
     env.cpp
     exit.cpp
+    process_id.cpp
     thread_id.cpp
+    thread_name.cpp
 )
 
 PEERDIR(
@@ -15,6 +17,10 @@ PEERDIR(
 
 END()
 
+RECURSE(
+    benchmarks
+)
+
 RECURSE_FOR_TESTS(
     unittests
 )