Dovetail - Sync / Async wrapper for Python.

A lightweight helper for bridging sync and async code, built on Asyncio & ThreadPoolExecutor.

Table of Contents

• Why Dovetail
• Install
• Quick Start
• API Summary
• Detailed Guides
• Development
• License

Why Dovetail?

Asyncio is powerful, but it has a cascading problem: the moment you want to await something, that function has to be async, which means everything calling it has to be async, and so on until your entire codebase has been rewritten around it. If you're adding parallelism to an existing project, or just want to speed up one specific bottleneck, that cost is hard to justify.

Dovetail's advantage isn't raw performance — asyncio.gather will always be at least as fast, since Dovetail is built on top of it. The advantage of Dovetail is its surgical approach: you can parallelise one function in an otherwise synchronous codebase without touching anything else.

# Existing sync code, completely unchanged around this one call
def process_report():
    data = dvt.task.map_blocking(fetch_url, urls)  # parallel, but sync caller
    return summarise(data)

With raw asyncio you can't do this — fetch_url being async forces process_report to be async, which forces everything above it to be async. Dovetail breaks that chain. If you're starting a greenfield project and are comfortable with async/await throughout, you probably don't need it. Dovetail is for the developers who are stuck between "I want parallelism" and "I don't want to rewrite everything."

Install

pip install pydovetail

Note: The install name is pydovetail, but the import name is dovetail.

from dovetail import Dovetail

Quick Start

Preferred usage: create and manage a Dovetail instance with a context manager.

Sync (recommended for most users):

from dovetail import Dovetail

with Dovetail(max_workers=8) as dvt:
  # Sync caller -> sync function (blocks)
  result = dvt.task.run_blocking(fetch_data)

  # Sync caller -> sync function in threadpool (blocks)
  result = dvt.task.to_thread_blocking(fetch_data, "target.json")

  # Sync caller -> batch map in threadpool (blocks, bounded concurrency)
  results = dvt.task.map_blocking(fetch_data, ["a.json", "b.json"], max_concurrency=4)

  # run an async coroutine from sync code
  value = dvt.task.run_blocking(fetch_data_async())

Async (use async with in async applications):

async with Dovetail() as dvt:
  # schedule coroutines and await
  tasks = [dvt.task.schedule(fetch_async(i)) for i in items]
  results = await asyncio.gather(*tasks)

Manual / library patterns (caller-managed ownership):

# Manual creation: by default Dovetail will attempt to shutdown on
# interpreter exit and on SIGINT/SIGTERM. You can opt out by passing
# `shutdown_on_exit=False` when constructing the instance.
dvt = Dovetail()  # shutdown_on_exit enabled by default
try:
  results = dvt.task.map_blocking(fetch, items)
finally:
  # explicit shutdown is still allowed and idempotent
  dvt.shutdown()

# Library-author pattern (prefer caller-managed instance when possible)
def process(items, dvt: Optional[Dovetail] = None):
  if dvt is None:
    with Dovetail() as _dvt:
      return _dvt.task.map_blocking(worker, items)
  return dvt.task.map_blocking(worker, items)

Note: Dovetail will, by default, register a best-effort shutdown handler that calls shutdown() on interpreter exit and when SIGINT or SIGTERM are received. This is opt-outable via shutdown_on_exit=False. Explicitly calling dvt.shutdown() is still supported and idempotent.

Sync context manager (recommended for synchronous code):

with Dovetail(max_workers=8) as dvt:
    results = dvt.task.map_blocking(fetch, items)
# threadpool shut down on block exit

Async context manager (recommended for async code):

async def main():
    async with Dovetail() as dvt:
        tasks = [dvt.task.schedule(coro(i)) for i in items]
        await asyncio.gather(*tasks)
# __aexit__ runs shutdown off the event loop to avoid blocking

Implementation note: Dovetail supports both __enter__/__exit__ and __aenter__/__aexit__. The async exit runs shutdown() in a background executor to avoid blocking the running loop. shutdown() is idempotent and safe to call multiple times.

API Summary

Task execution

• dvt.task.to_thread(func, *args, **kwargs) — await a sync callable in Dovetail's threadpool from async code; supports default timeout/retry settings.
• dvt.task.to_thread_blocking(func, *args, **kwargs) — synchronous wrapper for to_thread when you are not already in an event loop.
• dvt.task.schedule(coro_or_callable, *args, **kwargs) — create and return an asyncio.Task from a coroutine object, async function, or sync callable.
• dvt.task.run_blocking(func_or_coro, *args, **kwargs) — run async or sync work from synchronous code. Do not call from inside a running event loop.
• dvt.task.map_blocking(func, items, max_concurrency=None, return_exceptions=False) — run an ordered parallel map over items using the threadpool.

Events

• dvt.events.on_queued(...) — fires when a task is queued.
• dvt.events.on_start(...) — fires when a task starts executing.
• dvt.events.on_end(...) — fires when a task resolves (including after retries).
• dvt.events.on_error(...) — fires when a task raises an exception.
• dvt.events.on_retry(...) — fires on each retry attempt.
• dvt.events.on_cancel(...) — fires when a task is cancelled.
• dvt.events.stats() — returns cumulative counters: queued, started, done, error, retries, throttled.

Registry (auto-registration)

• dovetail now exposes a lightweight registry to track active Dovetail instances. By default Dovetail(..., auto_register=True) will register itself with the package registry so applications can perform coordinated shutdown.
• Public helpers:
  • from dovetail import register, unregister, list_active, shutdown_all, set_app_shutdown_hook
  • register(dvt) — explicitly register an instance (optional; instances auto-register by default).
  • unregister(dvt) — remove from the registry (useful if you manage shutdown yourself).
  • list_active() — return a list of active (live) instances.
  • shutdown_all(wait=True) — best-effort shutdown of all registered instances (useful for app shutdown hooks).
  • set_app_shutdown_hook(fn) — provide an application-level hook that the registry will call during process exit.

Notes:

• The registry uses weakrefs and logs a warning if a Dovetail is garbage-collected without an explicit shutdown() call.
• To opt-out of auto-registration when creating a Dovetail, pass auto_register=False to the constructor.

All dvt.events.on_* methods accept these optional scope parameters:

• function_target — only receive events for a specific callable.
• instance_target — only receive events for a specific execution id (available as id in the event payload).
• allow_reentry=False — prevents a listener from triggering itself recursively. Set to True to allow reentry, bounded by max_chain_depth.
• max_chain_depth=5 — maximum active callback depth when reentry is enabled.

Detailed Guides

Full, expanded guides live in the repository under docs/detailed_guide.md.

Development

Run tests locally:

python -m pytest -q

Contributions welcome — open an issue or PR.

License

MIT. See CHANGELOG.md for release notes.