Merge pull request #22 from modern-python/feat/v0.4-retry-and-budget

feat: Retry middleware + Finagle RetryBudget (0.4.0 slice 1)

Author: lesnik512

.gitignore

@@ -11,6 +11,7 @@ __pycache__/*
 .env
 .pytest_cache
 .ruff_cache
+.hypothesis/
 .coverage
 htmlcov/
 coverage.xml

planning/deferred-work.md

@@ -4,6 +4,10 @@ Items raised in reviews that are real but not actionable now.
 
 ## Open
 
+### Retry + streaming bodies (Epic 4 interaction)
+
+- **`Retry` re-invokes `next(request)` with the same `httpx2.Request` on each attempt.** Safe for in-memory bytes/JSON bodies; unsafe for streaming/async-iterable bodies (consumed iterator can't replay). When Epic 4 ships `AsyncClient.stream` (`4-3`), Retry needs to refuse to retry streamed-body requests (or document that callers supply a body factory). Spec: `planning/specs/2026-06-05-retry-and-retry-budget-design.md` §"Open questions".
+
 ### Decoder-side
 
 - **`_get_adapter` `lru_cache` is module-global, not per-decoder instance** — keyed by `model` only; two `PydanticDecoder()` instances with different configurations (none today) would share adapters, and the cache survives across tests unless explicitly cleared. Revisit if/when a configurable `PydanticDecoder(mode=..., strict=...)` lands. (`src/httpware/decoders/pydantic.py:12-14`)

planning/engineering.md

@@ -121,7 +121,9 @@ Post-pivot, the roadmap has three categories. Topic slugs in `planning/specs/` a
 
 ### Surviving (land in subsequent PRs)
 
-- **Epic 3 — Resilience:** `3-1` per-attempt timeout, `3-2` retry, `3-3` `RetryBudget`, `3-4` `RetryBudget` middleware integration, `3-5` `Bulkhead`, `3-6` extension-slot docs.
+- **Epic 3 — Resilience:**
+  - **Shipped in v0.4 slice 1:** `Retry` middleware + Finagle-style `RetryBudget` token bucket + `attempt_timeout=` parameter (folded-in 3-1). See [`planning/specs/2026-06-05-retry-and-retry-budget-design.md`](specs/2026-06-05-retry-and-retry-budget-design.md) and [`planning/plans/2026-06-05-retry-and-retry-budget-plan.md`](plans/2026-06-05-retry-and-retry-budget-plan.md).
+  - **Remaining:** `3-5` `Bulkhead`, `3-6` extension-slot docs.
 - **Epic 4 — Streaming:** `4-3` `AsyncClient.stream` context manager (forwards to `httpx2.AsyncClient.stream`; no `StreamResponse` type).
 - **Epic 5 — Observability:** `5-1` Layer 1 middleware hooks, `5-2` wire into resilience middlewares, `5-4` OpenTelemetry middleware (`otel` extra), `5-5` logging policy CI grep.
 - **Epic 6 — Ship v1.0:** `6-2` docs site (`mkdocs`), `6-3` benchmarks, `6-5` release flow (Trusted Publishers + Sigstore).

planning/plans/2026-06-05-retry-and-retry-budget-plan.md

@@ -63,12 +63,14 @@ mkdir -p src/httpware/middleware/resilience
 
 Then create each file with the contents below. Use the Write tool, not bash heredocs.
 
-`src/httpware/middleware/resilience/__init__.py`:
+`src/httpware/middleware/resilience/__init__.py` (docstring-only — re-exports defer to Task 7 so intermediate tasks can `import httpware.middleware.resilience.budget` without tripping an import-time `ImportError` from this `__init__.py`):
 ```python
-"""Resilience primitives: Retry middleware and RetryBudget token bucket."""
+"""Resilience primitives: Retry middleware and RetryBudget token bucket.
 
-from httpware.middleware.resilience.budget import RetryBudget
-from httpware.middleware.resilience.retry import Retry
+Re-exports land in Task 7 once both classes exist; until then this file is
+docstring-only so that importing ``httpware.middleware.resilience.budget``
+during the intermediate tasks does not trip an import-time ``ImportError``.
+"""
 ```
 
 `src/httpware/middleware/resilience/budget.py`:
@@ -137,7 +139,11 @@ Expected: FAIL (`ImportError: cannot import name 'NetworkError'`).
 Edit `src/httpware/errors.py`. Add a new class immediately after the existing `class TransportError`:
 ```python
 class NetworkError(TransportError):
-    """Transient network-layer failure (connect/read/write/pool). Safe to retry."""
+    """Transient network-layer failure (connect/read/write/close). Safe to retry.
+
+Pool-acquisition timeouts are NOT under this class; they raise ``TimeoutError``
+via ``httpx2.PoolTimeout`` (a ``TimeoutException`` subclass).
+"""
 ```
 
 Run: `uv run pytest tests/test_errors.py::test_network_error_is_transport_error -v`
@@ -218,7 +224,7 @@ Becomes:
 
 The `httpx2.NetworkError` branch must come BEFORE `httpx2.HTTPError` (HTTPError is the broader base). `httpx2.NetworkError` is httpx's documented base for `ConnectError`, `ReadError`, `WriteError`, `PoolTimeout` — if `httpx2`'s symbol name differs (e.g., `httpx2.exceptions.NetworkError`), use whichever import path mirrors the existing `httpx2.ConnectError` import in `tests/test_error_mapping_terminal.py` (which works via top-level `httpx2`).
 
-If `httpx2.NetworkError` does not exist, fall back to enumerating the transient subset explicitly: `except (httpx2.ConnectError, httpx2.ReadError, httpx2.WriteError, httpx2.PoolTimeout) as exc:`. The plan author has confirmed `httpx2.ConnectError` and `httpx2.ReadTimeout` already work in the existing tests; the enumeration fallback is safe.
+If `httpx2.NetworkError` does not exist, fall back to enumerating the transient subset explicitly: `except (httpx2.ConnectError, httpx2.ReadError, httpx2.WriteError, httpx2.CloseError) as exc:`. (`PoolTimeout` is NOT in this list — it inherits from `httpx2.TimeoutException` and is already caught by the timeout branch above.) The plan author has confirmed `httpx2.ConnectError` and `httpx2.ReadTimeout` already work in the existing tests; the enumeration fallback is safe.
 
 - [ ] **Step 6: Run the new terminal-mapping test**
 
@@ -265,7 +271,7 @@ git add src/httpware/errors.py src/httpware/client.py tests/test_errors.py tests
 git commit -m "feat(errors): add NetworkError(TransportError) for transient httpx2 failures
 
 Refines _terminal so httpx2.NetworkError-family exceptions (ConnectError, ReadError,
-WriteError, PoolTimeout) map to httpware.NetworkError. InvalidURL and CookieConflict
+WriteError, CloseError) map to httpware.NetworkError. InvalidURL and CookieConflict
 stay bare TransportError. Prerequisite for the Retry middleware so it can retry
 transient failures without retrying typos."
 ```
@@ -1029,15 +1035,30 @@ class Retry:
 Run: `uv run pytest tests/test_retry.py -v`
 Expected: all PASS.
 
-- [ ] **Step 4: Lint**
+- [ ] **Step 4: Wire `Retry` + `RetryBudget` into `resilience/__init__.py`**
 
-Run: `uv run ruff check src/httpware/middleware/resilience/retry.py tests/test_retry.py && uv run ty check src/httpware/middleware/resilience/retry.py`
+Now that both classes exist, replace `src/httpware/middleware/resilience/__init__.py` with:
+```python
+"""Resilience primitives: Retry middleware and RetryBudget token bucket."""
+
+from httpware.middleware.resilience.budget import RetryBudget
+from httpware.middleware.resilience.retry import Retry
+
+
+__all__ = ["Retry", "RetryBudget"]
+```
+
+The `__all__` is required to silence ruff F401 ("imported but unused") and matches the pattern used by `httpware/__init__.py` and `httpware/decoders/__init__.py`.
+
+- [ ] **Step 5: Lint**
+
+Run: `uv run ruff check src/httpware/middleware/resilience/ tests/test_retry.py && uv run ty check src/httpware/middleware/resilience/`
 Expected: clean. If ruff flags `Callable` / `Awaitable` import paths, adjust per existing project pattern (see `middleware/__init__.py` which uses `from collections.abc import Awaitable, Callable`).
 
-- [ ] **Step 5: Stage and commit**
+- [ ] **Step 6: Stage and commit**
 
 ```bash
-git add src/httpware/middleware/resilience/retry.py tests/test_retry.py
+git add src/httpware/middleware/resilience/retry.py src/httpware/middleware/resilience/__init__.py tests/test_retry.py
 git commit -m "feat(resilience): Retry middleware — status-code retry + exhaustion
 
 Covers: happy path, 503-then-200, max_attempts exhaustion with PEP-678 note,

planning/specs/2026-06-05-retry-and-retry-budget-design.md

@@ -28,7 +28,7 @@ The current `AsyncClient._terminal` maps every non-timeout `httpx2.HTTPError` (i
 ```python
 # In src/httpware/errors.py:
 class NetworkError(TransportError):
-    """Transient network-layer failure (connect / read / write / pool). Safe to retry."""
+    """Transient network-layer failure (connect / read / write / close). Safe to retry."""
 ```
 
 And refines the terminal mapping so that `httpx2`'s transient-network exception family (`httpx2.NetworkError` per httpx convention, or whichever symbols httpx2 exposes for the same hierarchy) raises `httpware.NetworkError` rather than the broader `TransportError`. `InvalidURL` and `CookieConflict` continue to raise `TransportError` directly so they are NOT retried. Existing tests catching `TransportError` keep working (`NetworkError` is a subclass).

src/httpware/__init__.py

@@ -10,8 +10,10 @@
     ConflictError,
     ForbiddenError,
     InternalServerError,
+    NetworkError,
     NotFoundError,
     RateLimitedError,
+    RetryBudgetExhaustedError,
     ServerStatusError,
     ServiceUnavailableError,
     StatusError,
@@ -21,6 +23,7 @@
     UnprocessableEntityError,
 )
 from httpware.middleware import Middleware, Next, after_response, before_request, on_error
+from httpware.middleware.resilience import Retry, RetryBudget
 
 
 __all__ = [
@@ -33,10 +36,14 @@
     "ForbiddenError",
     "InternalServerError",
     "Middleware",
+    "NetworkError",
     "Next",
     "NotFoundError",
     "RateLimitedError",
     "ResponseDecoder",
+    "Retry",
+    "RetryBudget",
+    "RetryBudgetExhaustedError",
     "ServerStatusError",
     "ServiceUnavailableError",
     "StatusError",

src/httpware/client.py

@@ -11,6 +11,7 @@
 from httpware.errors import (
     STATUS_TO_EXCEPTION,
     ClientStatusError,
+    NetworkError,
     ServerStatusError,
     TimeoutError,  # noqa: A004
     TransportError,
@@ -110,6 +111,8 @@ async def _terminal(self, request: httpx2.Request) -> httpx2.Response:
             raise TimeoutError(str(exc)) from exc
         except (httpx2.InvalidURL, httpx2.CookieConflict) as exc:
             raise TransportError(str(exc)) from exc
+        except httpx2.NetworkError as exc:
+            raise NetworkError(str(exc)) from exc
         except httpx2.HTTPError as exc:
             raise TransportError(str(exc)) from exc
         except RuntimeError as exc:

src/httpware/errors.py

@@ -40,6 +40,14 @@ class TransportError(ClientError):
     """Connection / network / protocol failure raised before a response was received."""
 
 
+class NetworkError(TransportError):
+    """Transient network-layer failure (connect/read/write/close). Safe to retry.
+
+    Pool-acquisition timeouts are NOT under this class; they raise ``TimeoutError``
+    via ``httpx2.PoolTimeout`` (a ``TimeoutException`` subclass).
+    """
+
+
 class TimeoutError(ClientError, builtins.TimeoutError):  # noqa: A001
     """Client-side timeout (connect / read / write / pool).
 
@@ -136,3 +144,42 @@ class ServiceUnavailableError(ServerStatusError):
     500: InternalServerError,
     503: ServiceUnavailableError,
 }
+
+
+def _reconstruct_budget_exhausted(
+    cls: "type[RetryBudgetExhaustedError]",
+    last_response: httpx2.Response | None,
+    last_exception: BaseException | None,
+    attempts: int,
+) -> "RetryBudgetExhaustedError":
+    return cls(last_response=last_response, last_exception=last_exception, attempts=attempts)
+
+
+class RetryBudgetExhaustedError(ClientError):
+    """Raised when a retry was needed but the RetryBudget refused to permit it.
+
+    Carries the last response and/or exception observed before the budget refused,
+    plus the number of attempts already completed.
+    """
+
+    last_response: httpx2.Response | None
+    last_exception: BaseException | None
+    attempts: int
+
+    def __init__(
+        self,
+        *,
+        last_response: httpx2.Response | None,
+        last_exception: BaseException | None,
+        attempts: int,
+    ) -> None:
+        self.last_response = last_response
+        self.last_exception = last_exception
+        self.attempts = attempts
+        super().__init__(f"retry budget exhausted after {attempts} attempt(s)")
+
+    def __reduce__(self) -> tuple[Any, ...]:
+        return (
+            _reconstruct_budget_exhausted,
+            (type(self), self.last_response, self.last_exception, self.attempts),
+        )

src/httpware/middleware/resilience/__init__.py

@@ -0,0 +1,7 @@
+"""Resilience primitives: Retry middleware and RetryBudget token bucket."""
+
+from httpware.middleware.resilience.budget import RetryBudget
+from httpware.middleware.resilience.retry import Retry
+
+
+__all__ = ["Retry", "RetryBudget"]

src/httpware/middleware/resilience/_backoff.py

@@ -0,0 +1,26 @@
+"""Full-jitter exponential backoff helper (private)."""
+
+import random
+from collections.abc import Callable
+
+
+def full_jitter_delay(
+    attempt_index: int,
+    *,
+    base_delay: float,
+    max_delay: float,
+    _random_uniform: Callable[[float, float], float] = random.uniform,
+) -> float:
+    """Return a backoff delay using AWS's "full jitter" formulation.
+
+    sleep = uniform(0, min(max_delay, base_delay * 2.0 ** attempt_index))
+
+    `attempt_index` is 0 for the first retry, 1 for the second, etc.
+
+    Uses ``2.0 **`` (float exponentiation) rather than ``2 **`` so that
+    ``attempt_index >= 1024`` saturates to ``math.inf`` and ``min`` clamps to
+    ``max_delay`` — ``2 ** 1024`` would raise ``OverflowError`` during the
+    int→float conversion.
+    """
+    ceiling = min(max_delay, base_delay * (2.0**attempt_index))
+    return _random_uniform(0.0, ceiling)