Timer optimize by ankitkpandey1 · Pull Request #1 · ankitkpandey1/uringcore

ankitkpandey1 · 2026-01-02T17:17:40Z

No description provided.

…tions, and fixed type signatures.

…bprocess)

…legate execution.

…ing.

…python 3.14 deprecation warnings, and pass code quality checks

…optimization

…ve ENOMEM error

…es API

…ing()/uncancel() methods

- subprocess.py: Add returncode property and _wait() async method for asyncio compatibility - transport.py: Add _recv_pending flag to prevent duplicate recv submissions - loop.py: Fix accept/connect flows, increase default buffer count to 512 - benchmark_suite.py: Add Plotly HTML chart generation - BENCHMARK.md: Add comprehensive benchmark analysis - test_fastapi.py, test_starlette.py: Refactor to use httpx.AsyncClient + ASGITransport - conftest.py: Add global pytest configuration with buffer limits - task.rs: Add _loop getter for anyio compatibility - pyproject.toml: Add httpx, plotly, kaleido to dev dependencies - buf_ring.rs, fixed_fd.rs: New modules for provided buffer ring and registered FDs - Various verify tests for multishot accept, pbuf_ring, registered FDs Test Results: 85 passed, 11 failed (known limitations), 3 skipped

- scheduler.rs: Replace Mutex<VecDeque> with crossbeam-channel for lock-free push/drain - lib.rs: Merge Ring lock acquisitions (submit + drain_completions in single lock) - loop.py: Skip epoll.poll when ready tasks exist (fast path for gather/sleep) Results: - gather(100): 454µs → 387µs (~15% improvement) - sleep_conc_100: 577µs → 469µs (~19% improvement) Note: Remaining ~2-3x gap vs uvloop is inherent PyO3/Python call overhead, not Rust-side locking.

- ARCHITECTURE.md: Update Native Task Scheduling section for Phase 3+10 - Document lock-free MPSC scheduler (crossbeam-channel) - Add merged Ring lock and epoll skip optimizations - Update SOTA table with new optimizations - Revise Future Work section - README.md: Update project status and benchmarks - Change phase to Phase 10 (Lock-Free Scheduler) - Update key features with lock-free scheduler - Refresh benchmark numbers (sleep 2.8x, semaphore 3.2x, wait_for 2.3x)

ARCHITECTURE.md: - Added 'Performance Bottleneck Analysis (Phase 11)' section - Documented PyO3 call overhead as root cause (~500-1000ns × 8-12 calls per step) - Listed attempted optimizations that didn't improve performance - Noted architecture implication: need 90%+ Rust migration to match uvloop README.md: - Updated project status to Phase 11 - Split benchmarks into 'Single-Task Latency' (wins) and 'High-Concurrency' (loses) - Added honest performance comparison with explanation

- Removed _enter_task/_leave_task from run_step hot path (major win) - Implemented direct scheduler push (avoiding call_soon overhead) - Inlined wakeup caching in run_step logic - Moved run_step to internal impl block to handle Rust types - Result: sleep_conc_100 29% faster (419µs), gather(100) 8% faster (380µs)

- Fixed TimerHandle cancellation crash in Rust scheduler (src/lib.rs) - Updated README.md: Phase 15 status, latest benchmark results (2.9x sleep(0)) - Updated ARCHITECTURE.md: Added Phase 14 stress testing documentation - cargo fmt && clippy: All checks pass - Core tests: 27/27 PASS - Benchmarks: uringcore 4.26µs sleep(0) vs uvloop 12.53µs

Copilot

Pull request overview

This PR implements a comprehensive timer optimization for the uringcore event loop, introducing native Rust-based timer scheduling and modernizing the asyncio integration patterns. The changes include migrating from Python-based timer heaps to lock-free Rust schedulers, implementing native futures and tasks, and updating tests to use modern factory patterns instead of deprecated event loop policies.

Key Changes:

Native timer implementation using Rust BinaryHeap with lock-free scheduling
Replacement of Python event loop policy pattern with modern new_event_loop() factory
Introduction of native UringTask, UringFuture, and UringHandle types for performance
Buffer configuration via environment variables (URINGCORE_BUFFER_COUNT, URINGCORE_BUFFER_SIZE)

Reviewed changes

Copilot reviewed 51 out of 52 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/timer.rs	New native timer heap implementation using BinaryHeap for efficient timer scheduling
src/task.rs	Native UringTask implementation with optimized coroutine stepping
src/scheduler.rs	Lock-free task scheduler using crossbeam-channel MPSC
src/ring.rs	Extended io_uring operations with timer support and multishot operations
src/lib.rs	Integration of new modules and updated buffer initialization
python/uringcore/loop.py	Refactored event loop to use Rust-native scheduling and timer management
python/uringcore/init.py	Added new_event_loop factory function for Python 3.11+ compatibility
tests/*	Updated all tests to use modern event loop factory pattern
benchmarks/benchmark_suite.py	Expanded benchmark suite with additional test cases

Comments suppressed due to low confidence (2)

python/uringcore/subprocess.py:1

The broad exception handler on line 241 catches all exceptions but treats them all as if wait_for failed. This could mask unexpected errors. Consider being more specific about which exceptions warrant the manual timeout fallback.

"""Subprocess transport and protocol support for uringcore."""

python/uringcore/loop.py:1

This code snippet appears to have a syntax error - set_description is being called with incorrect syntax. The string "timeout" should not be passed as a positional argument in this manner. This line will fail at runtime.

"""UringEventLoop: Pure io_uring asyncio event loop.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-02T17:18:55Z

+        if self._paused:
+            self._paused = False
+            self._loop._core.resume_reading(self._fd)
+        # Always rearm receive when resuming or starting


The comment on line 82 states "Always rearm receive when resuming or starting" but _rearm_recv() is only called when self._paused was previously True (line 79-80). Consider clarifying the comment or adjusting the logic to match the description.

Suggested change

# Always rearm receive when resuming or starting

# Ensure receive is (re)armed when reading is enabled

Copilot · 2026-01-02T17:18:56Z

+             self._loop.remove_reader(self._pidfd)
+             try:
+                 os.close(self._pidfd)
+             except OSError:
+                 pass
+             self._pidfd = None
+
+        # Process has exited, wait() should return immediately
+        try:
+             # WNOHANG shouldn't be needed if pidfd signaled, but safer
+             # Actually for standard Popen, just wait() is fine as it reaps.
+             returncode = self._proc.wait()
+             self._process_exited(returncode)
+        except Exception:
+             # Should not happen
+             pass


The indentation appears inconsistent here with an extra space before self._loop.remove_reader. This should align with the surrounding code at the same indentation level.

Suggested change

self._loop.remove_reader(self._pidfd)

try:

os.close(self._pidfd)

except OSError:

pass

self._pidfd = None

# Process has exited, wait() should return immediately

try:

# WNOHANG shouldn't be needed if pidfd signaled, but safer

# Actually for standard Popen, just wait() is fine as it reaps.

returncode = self._proc.wait()

self._process_exited(returncode)

except Exception:

# Should not happen

pass

self._loop.remove_reader(self._pidfd)

try:

os.close(self._pidfd)

except OSError:

pass

self._pidfd = None

# Process has exited, wait() should return immediately

try:

# WNOHANG shouldn't be needed if pidfd signaled, but safer

# Actually for standard Popen, just wait() is fine as it reaps.

returncode = self._proc.wait()

self._process_exited(returncode)

except Exception:

# Should not happen

pass

Copilot · 2026-01-02T17:18:56Z

        """Test simple subprocess execution."""
-        proc = await asyncio.create_subprocess_exec(
-            sys.executable, "-c", "print('hello')",
-            stdout=asyncio.subprocess.PIPE,
-            stderr=asyncio.subprocess.PIPE,
-        )
-
-        stdout, stderr = await proc.communicate()
-
-        assert proc.returncode == 0
-        assert b"hello" in stdout
+        async def check():


The parameter name loop is inconsistent with the fixture name event_loop. While this works via the loop fixture (line 141-142), using event_loop directly would be clearer and avoid the additional indirection.

Copilot · 2026-01-02T17:18:56Z

+             from uringcore import new_event_loop
+             return new_event_loop(buffer_count=buffer_count, buffer_size=buffer_size)


The import statement is inside the function definition. While this works, consider moving the import to the top of the function scope (line 361) alongside the other imports for better readability and consistency.

ankitkpandey1 added 26 commits January 1, 2026 14:12

Restored native I/O with high-perf defaults, added TimerHeap optimiza…

81dff5d

…tions, and fixed type signatures.

Fix remaining mypy type errors in auxiliary modules (policy/server/su…

c7add2c

…bprocess)

Phase 2: Implemented Scheduler in Rust (VecDeque), updated loop to de…

db398fa

…legate execution.

Phase 3: Implemented UringHandle in Rust to optimize callback schedul…

09bc024

…ing.

Phase 5: Rust Native Future implementation with direct Task integration

7738302

refactor: resolve all borrow checker issues in Ring using Mutex, fix …

a7fcfe0

…python 3.14 deprecation warnings, and pass code quality checks

feat: Implement native timer and lazy I/O submission for performance …

59d1fab

…optimization

fix: Implement Drop for Ring to prevent locked memory leaks and impro…

11d6807

…ve ENOMEM error

Fix memory leak via Ring Drop and correct cancellation propagation

91a3220

Update docs and optimize task creation path

67220ac

Optimize gather: batch drain scheduler and prioritize UringTask downcast

36b0955

SOTA 2025: asyncio caching, native timers, multishot recv, capabiliti…

476df1f

…es API

SOTA: Registered FD table and zero-copy send APIs

7e29a6d

Update ARCHITECTURE.md: Mark Registered FDs and SEND_ZC as implemented

68b3b2b

Update README with SOTA 2025 features and benchmarks

2252a6c

Add real-world stress test (WIP)

bac94ca

Fix CancelledError detection in _step using Python isinstance

a901ece

Fix TypeError in cancellation: use throw(type, value) and add cancell…

adc108c

…ing()/uncancel() methods

Sync getaddrinfo, fix TypeError in cancellation, add stress test

f031e33

Standardize author email to ankitkpandey1@gmail.com

bdb9042

ankitkpandey1 requested a review from Copilot January 2, 2026 17:17

Copilot AI reviewed Jan 2, 2026

View reviewed changes

ankitkpandey1 merged commit e8dcf73 into main Jan 2, 2026
6 of 7 checks passed

Navigation Menu

Timer optimize - #1

Timer optimize#1
ankitkpandey1 merged 26 commits into
mainfrom
timer-optimize

ankitkpandey1 commented Jan 2, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 2, 2026

Uh oh!

Copilot AI Jan 2, 2026

Uh oh!

Copilot AI Jan 2, 2026

Uh oh!

Copilot AI Jan 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	# Always rearm receive when resuming or starting
	# Ensure receive is (re)armed when reading is enabled

		from uringcore import new_event_loop
		return new_event_loop(buffer_count=buffer_count, buffer_size=buffer_size)

Conversation

ankitkpandey1 commented Jan 2, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants