distributed structured concurrency https://github.com/goodboy/tractor
Go to file
Gud Boi 80aa465cf6 Refine `subint_forkserver` nested-cancel hang diagnosis
Major rewrite of
`subint_forkserver_test_cancellation_leak_issue.md`
after empirical investigation revealed the earlier
"descendant-leak + missing tree-kill" diagnosis
conflated two unrelated symptoms:

1. **5-zombie leak holding `:1616`** — turned out to
   be a self-inflicted cleanup bug: `pkill`-ing a bg
   pytest task (SIGTERM/SIGKILL, no SIGINT) skipped
   the SC graceful cancel cascade entirely. Codified
   the real fix — SIGINT-first ladder w/ bounded
   wait before SIGKILL — in e5e2afb5 (`run-tests`
   SKILL) and
   `feedback_sc_graceful_cancel_first.md`.
2. **`test_nested_multierrors[subint_forkserver]`
   hangs indefinitely** — the actual backend bug,
   and it's a deadlock not a leak.

Deats,
- new diagnosis: all 5 procs are kernel-`S` in
  `do_epoll_wait`; pytest-main's trio-cache workers
  are in `os.waitpid` waiting for children that are
  themselves waiting on IPC that never arrives —
  graceful `Portal.cancel_actor` cascade never
  reaches its targets
- tree-structure evidence: asymmetric depth across
  two identical `run_in_actor` calls — child 1
  (3 threads) spawns both its grandchildren; child 2
  (1 thread) never completes its first nursery
  `run_in_actor`. Smells like a race on fork-
  inherited state landing differently per spawn
  ordering
- new hypothesis: `os.fork()` from a subactor
  inherits the ROOT parent's IPC listener FDs
  transitively. Grandchildren end up with three
  overlapping FD sets (own + direct-parent + root),
  so IPC routing becomes ambiguous. Predicts bug
  scales with fork depth — matches reality: single-
  level spawn works, multi-level hangs
- ruled out: `_ForkedProc.kill()` tree-kill (never
  reaches hard-kill path), `:1616` contention (fixed
  by `reg_addr` fixture wiring), GIL starvation
  (each subactor has its own OS process+GIL),
  child-side KBI absorption (`_trio_main` only
  catches KBI at `trio.run()` callsite, reached
  only on trio-loop exit)
- four fix directions ranked: (1) blanket post-fork
  `closerange()`, (2) `FD_CLOEXEC` + audit,
  (3) targeted FD cleanup via `actor.ipc_server`
  handle, (4) `os.posix_spawn` w/ `file_actions`.
  Vote: (3) — surgical, doesn't break the "no exec"
  design of `subint_forkserver`
- standalone repro added (`spawn_and_error(breadth=
  2, depth=1)` under `trio.fail_after(20)`)
- stopgap: skip `test_nested_multierrors` + multi-
  level-spawn tests under the backend via
  `@pytest.mark.skipon_spawn_backend(...)` until
  fix lands

Killing the "tree-kill descendants" fix-direction
section: it addressed a bug that didn't exist.

(this patch was generated in some part by [`claude-code`][claude-code-gh])
[claude-code-gh]: https://github.com/anthropics/claude-code

(cherry picked from commit 35da808905)
2026-06-29 19:20:21 -04:00
.claude Add `--shm` orphan sweep to `tractor-reap` 2026-06-17 17:39:44 -04:00
.github/workflows Fix review nits from PR #460 self-review 2026-06-28 13:29:01 -04:00
ai Refine `subint_forkserver` nested-cancel hang diagnosis 2026-06-29 19:20:21 -04:00
docs Show error/cancel teardowns in the `Context` diagram 2026-06-28 21:55:39 -04:00
examples Rework landing example onto the `Context` API 2026-06-28 13:29:02 -04:00
nooz Add news file 2023-05-15 09:35:59 -04:00
notes_to_self Rework landing example onto the `Context` API 2026-06-28 13:29:02 -04:00
scripts Fix uds `get_random` reaper-regex break (+2 nits) 2026-06-29 18:51:26 -04:00
tests Mv `test_subint_cancellation.py` to `tests/spawn/` subpkg 2026-06-29 19:20:21 -04:00
tractor Mv `test_subint_cancellation.py` to `tests/spawn/` subpkg 2026-06-29 19:20:21 -04:00
xontrib Hoist proc-title prefix to `_def_prefix` const 2026-06-24 16:20:50 -04:00
.gitignore Reorganize `.gitignore` by skill/purpose 2026-06-16 14:21:08 -04:00
LICENSE Re-license code base for distribution under AGPL 2021-12-14 23:33:27 -05:00
MANIFEST.in Include ./docs/README.rst in src dist 2022-07-11 14:25:26 -04:00
NEWS.rst Fix informal-RST in docstrings for clean autodoc 2026-06-28 13:29:01 -04:00
default.nix Trying to make full suite pass with uds 2025-07-08 12:57:28 -04:00
flake.lock Add a dev-overlay nix flake 2026-01-23 16:27:19 -05:00
flake.nix Add opt-in `docs` dev-shell with `d2` 2026-06-28 13:29:01 -04:00
mypy.ini Add mypy.ini lel 2020-01-21 15:28:12 -05:00
pyproject.toml Add `docs` dependency-group for sphinx 9 stack 2026-06-28 13:29:01 -04:00
ruff.toml Bump `ruff.toml` to target py313 2026-03-08 18:39:08 -04:00
uv.lock Relock `uv.lock` for the `docs` dependency-group 2026-06-28 13:29:02 -04:00

docs/README.rst

logo tractor: distributed structured concurrency

tractor is a structured concurrency (SC), multi-processing runtime built on trio.

Fundamentally, tractor provides parallelism via trio-"actors": independent Python processes (i.e. non-shared-memory threads) which can schedule trio tasks whilst maintaining end-to-end SC inside a distributed supervision tree.

Cross-process (and thus cross-host) SC is accomplished through the combined use of our,

  • "actor nurseries" which provide for spawning multiple, and possibly nested, Python processes each running a trio scheduled runtime - a call to trio.run(),
  • an "SC-transitive supervision protocol" enforced as an IPC-message-spec encapsulating all RPC-dialogs.

We believe the system adheres to the 3 axioms of an "actor model" but likely does not look like what you probably think an "actor model" looks like, and that's intentional.

Where do i start!?

New to trio and structured concurrency? Our docs collect the best starting points and then walk you straight into a hands-on quickstart:

https://goodboy.github.io/tractor/start/quickstart.html

Features

  • It's just a trio API!
  • Infinitely nesteable process trees running embedded trio tasks.
  • Swappable, OS-specific, process spawning via multiple backends.
  • Modular IPC stack, allowing for custom interchange formats (eg. as offered from msgspec), varied transport protocols (TCP, RUDP, QUIC, wireguard), and OS-env specific higher-perf primitives (UDS, shm-ring-buffers).
  • Optionally distributed: all IPC and RPC APIs work over multi-host transports the same as local.
  • Builtin high-level streaming API that enables your app to easily leverage the benefits of a "cheap or nasty" (un)protocol.
  • A "native UX" around a multi-process safe debugger REPL using pdbp (a fork & fix of pdb++)
  • "Infected asyncio" mode: support for starting an actor's runtime as a guest on the asyncio loop allowing us to provide stringent SC-style trio.Task-supervision around any asyncio.Task spawned via our tractor.to_asyncio APIs.
  • A very naive and still very much work-in-progress inter-actor discovery sys with plans to support multiple modern protocol approaches.
  • Various trio extension APIs via tractor.trionics such as,
    • task fan-out broadcasting,
    • multi-task-single-resource-caching and fan-out-to-multi __aenter__() APIs for @acm functions,
    • (WIP) a TaskMngr: one-cancels-one style nursery supervisor.

Status of main / infra

  • gh_actions
  • Documentation

Install

tractor is still in a alpha-near-beta-stage for many of its subsystems, however we are very close to having a stable lowlevel runtime and API.

As such, it's currently recommended that you clone and install the repo from source:

pip install git+git://github.com/goodboy/tractor.git

We use the very hip uv for project mgmt:

git clone https://github.com/goodboy/tractor.git
cd tractor
uv sync --dev
uv run python examples/rpc_bidir_streaming.py

Consider activating a virtual/project-env before starting to hack on the code base:

# you could use plain ol' venvs
# https://docs.astral.sh/uv/pip/environments/
uv venv tractor_py313 --python 3.13

# but @goodboy prefers the more explicit (and shell agnostic)
# https://docs.astral.sh/uv/configuration/environment/#uv_project_environment
UV_PROJECT_ENVIRONMENT="tractor_py313"

# hint hint, enter @goodboy's fave shell B)
uv run --dev xonsh

Alongside all this we ofc offer "releases" on PyPi:

pip install tractor

Just note that YMMV since the main git branch is often much further ahead then any latest release.

Hacking on the docs themselves? The build + live-preview one-liners (incl. nix-shell specifics) are collected in notes_to_self/howtodocs.md, and rendered as the "Building these docs" section of our dev-tips guide.

Example codez

We prefer to point you at the runnable scripts under examples/ - each is CI-run and literalinclude-d straight into the docs, so what you read there is what actually runs - rather than inline a pile of them here. The one-minute pitch: spawn a subactor per core, open a Context into each, then crash the root on purpose and watch the runtime reap the whole tree - zero zombies, guaranteed (if you can make a zombie child without a system signal, it is a bug).

See it run - plus the full tour (the flagship multi-process debugger, bidirectional streaming over a Context, cancellation, discovery, "infected asyncio", typed messaging and worker-pool / cluster patterns) - in the docs:

Under the hood

tractor is an attempt to pair trionic structured concurrency with distributed Python - think of it as trio -across-processes, or as an opinionated replacement for the stdlib's multiprocessing built on async primitives from the ground up. But really it is just trio: nurseries that spawn processes and cancel-able streaming IPC between them. If you can drive trio, you can drive tractor.

"But wait - don't 'actors' have mailboxes and messages and stuff?!" Well, we've got (well referenced) opinions on what an "actor model" actually is (tl;dr: the 3 axioms, not the cultural baggage) - that whole riff lives in our docs:

https://goodboy.github.io/tractor/explain/sc-distributed.html#hold-up-is-this-an-actor-model

What's on the TODO

The roadmap lives with our docs - see what the future holds for where tractor is headed.

Feel like saying hi?

This project is very much coupled to the ongoing development of trio (i.e. tractor gets most of its ideas from that brilliant community). If you want to help, have suggestions or just want to say hi, please feel free to reach us in our matrix channel. If matrix seems too hip, we're also mostly all in the the trio gitter channel!