Persist authored action idempotency across restarts and workers #202

Merged
timmy merged 1 commits from timmy/201-durable-idempotency into main 2026-08-07 14:38:07 +00:00
7 changed files with 417 additions and 80 deletions

1
.gitignore vendored
View File

@ -2,3 +2,4 @@ __pycache__/
*.py[cod]
.pytest_cache/
.release-engine/
.stackchain-state/

View File

@ -23,8 +23,11 @@ Pull-request replies and mobile My Work issue and PR comments use Gitea's
issue-comment API; mobile issue capture requires issue
creation and assignment permission. Issue capture and authored mobile actions (issue
comments, pull-request comments, notification replies, and reviews) persist per-draft
idempotency keys, so retrying after a timeout or reload replays a confirmed result instead
of posting duplicate content. Direct API callers should preserve the `Idempotency-Key`
idempotency keys, so retrying after a timeout, reload, process restart, or handoff to
another worker replays a confirmed result instead of posting duplicate content. Results
are coordinated through a bounded SQLite ledger. Set `STACKCHAIN_STATE_DIR` to a
persistent, writable service directory (or set `STACKCHAIN_IDEMPOTENCY_DB` to an explicit
SQLite path); the local default is `.stackchain-state/idempotency.sqlite3`. Direct API callers should preserve the `Idempotency-Key`
header with the unchanged route and payload until a `201` response is confirmed. Closing an assigned issue, native Comment,
Approve, and Request changes reviews, and assigned-PR merge require repository
write permission. Native Comment, Approve, and Request changes reviews support

114
src/idempotency.py Normal file
View File

@ -0,0 +1,114 @@
import json
import sqlite3
import time
from dataclasses import dataclass
from pathlib import Path
from typing import Any, Callable
@dataclass(frozen=True)
class Reservation:
state: str
response: Any = None
class IdempotencyLedger:
"""Small durable ledger for replaying successful Gitea mutations."""
def __init__(
self,
path: str | Path,
*,
ttl_seconds: float,
max_entries: int,
clock: Callable[[], float] = time.time,
) -> None:
self.path = Path(path)
self.ttl_seconds = ttl_seconds
self.max_entries = max_entries
self.clock = clock
self.path.parent.mkdir(parents=True, exist_ok=True)
self._initialize()
def _connect(self) -> sqlite3.Connection:
connection = sqlite3.connect(self.path, timeout=5)
connection.execute("PRAGMA busy_timeout = 5000")
return connection
def _initialize(self) -> None:
with self._connect() as connection:
connection.execute(
"""
CREATE TABLE IF NOT EXISTS idempotency_operations (
key TEXT PRIMARY KEY,
fingerprint TEXT NOT NULL,
status TEXT NOT NULL CHECK(status IN ('pending', 'completed')),
response_json TEXT,
created_at REAL NOT NULL,
completed_at REAL
)
"""
)
connection.execute(
"CREATE INDEX IF NOT EXISTS idempotency_completed_at_idx "
"ON idempotency_operations(completed_at)"
)
@staticmethod
def _fingerprint(value: tuple[Any, ...]) -> str:
return json.dumps(value, separators=(",", ":"), sort_keys=True)
def reserve(self, key: str, fingerprint: tuple[Any, ...]) -> Reservation:
encoded = self._fingerprint(fingerprint)
now = self.clock()
with self._connect() as connection:
connection.execute("BEGIN IMMEDIATE")
connection.execute(
"DELETE FROM idempotency_operations "
"WHERE status = 'completed' AND completed_at <= ?",
(now - self.ttl_seconds,),
)
row = connection.execute(
"SELECT fingerprint, status, response_json FROM idempotency_operations "
"WHERE key = ?",
(key,),
).fetchone()
if row is not None:
if row[0] != encoded:
return Reservation("conflict")
if row[1] == "completed":
return Reservation("completed", json.loads(row[2]))
return Reservation("pending")
count = connection.execute(
"SELECT COUNT(*) FROM idempotency_operations"
).fetchone()[0]
if count >= self.max_entries:
completed = connection.execute(
"SELECT key FROM idempotency_operations "
"WHERE status = 'completed' ORDER BY completed_at, rowid LIMIT 1"
).fetchone()
if completed is None:
return Reservation("busy")
connection.execute(
"DELETE FROM idempotency_operations WHERE key = ?",
(completed[0],),
)
connection.execute(
"INSERT INTO idempotency_operations "
"(key, fingerprint, status, created_at) VALUES (?, ?, 'pending', ?)",
(key, encoded, now),
)
return Reservation("reserved")
def clear(self) -> None:
with self._connect() as connection:
connection.execute("DELETE FROM idempotency_operations")
def complete(self, key: str, response: Any) -> None:
encoded = json.dumps(response, separators=(",", ":"), sort_keys=True)
with self._connect() as connection:
connection.execute(
"UPDATE idempotency_operations SET status = 'completed', "
"response_json = ?, completed_at = ? WHERE key = ?",
(encoded, self.clock(), key),
)

View File

@ -1,5 +1,6 @@
import asyncio
import math
import os
import time
from collections.abc import Awaitable, Coroutine
from contextlib import asynccontextmanager
@ -24,6 +25,7 @@ from src.gitea_proxy import (
pull_review_detail,
repos,
)
from src.idempotency import IdempotencyLedger
from src.models import Issue, PullRequest, Repo, User
from src.suggestion_engine import compute
from src.views import router as frontend_router
@ -90,12 +92,18 @@ _live_section_retry_at: dict[str, float | None] = {
}
_live_snapshot_refreshing_sections: set[str] = set()
_read_notification_ids: set[int] = set()
_issue_creation_operations: dict[
str, tuple[tuple[Any, ...], asyncio.Task, float]
] = {}
_authored_action_operations: dict[
str, tuple[tuple[Any, ...], asyncio.Task, float]
] = {}
_state_dir = Path(os.getenv("STACKCHAIN_STATE_DIR", ".stackchain-state"))
_idempotency_ledger = IdempotencyLedger(
os.getenv("STACKCHAIN_IDEMPOTENCY_DB", str(_state_dir / "idempotency.sqlite3")),
ttl_seconds=AUTHORED_ACTION_IDEMPOTENCY_TTL_SECONDS,
max_entries=(
AUTHORED_ACTION_IDEMPOTENCY_MAX_ENTRIES
+ ISSUE_CREATION_IDEMPOTENCY_MAX_ENTRIES
),
)
_available_issue_snapshot_task: asyncio.Task | None = None
_available_issue_snapshot_value: list[dict] | None = None
_available_issue_snapshot_created_at: float | None = None
@ -236,39 +244,52 @@ async def _run_idempotent_authored_action(
if not idempotency_key:
return await asyncio.wait_for(operation, timeout=timeout)
now = time.monotonic()
expired = [
key for key, (_, task, created_at) in _authored_action_operations.items()
if task.done() and now - created_at >= AUTHORED_ACTION_IDEMPOTENCY_TTL_SECONDS
]
for key in expired:
_authored_action_operations.pop(key, None)
reservation = _idempotency_ledger.reserve(idempotency_key, fingerprint)
if reservation.state == "conflict":
operation.close()
raise HTTPException(status_code=409, detail="Idempotency key already used")
if reservation.state == "completed":
operation.close()
return reservation.response
if reservation.state == "busy":
operation.close()
raise HTTPException(
status_code=503,
detail="Authored action queue is busy; please retry",
headers={"Retry-After": "1"},
)
existing = _authored_action_operations.get(idempotency_key)
if existing is not None:
if reservation.state == "pending":
operation.close()
if existing[0] != fingerprint:
raise HTTPException(status_code=409, detail="Idempotency key already used")
if existing is None or existing[0] != fingerprint:
raise HTTPException(
status_code=503,
detail=(
"This action may still be completing; verify its result before retrying"
),
headers={"Retry-After": "5"},
)
task = existing[1]
else:
while len(_authored_action_operations) >= AUTHORED_ACTION_IDEMPOTENCY_MAX_ENTRIES:
completed = [
key for key, (_, task, _) in _authored_action_operations.items()
if task.done()
]
if not completed:
operation.close()
raise HTTPException(
status_code=503,
detail="Authored action queue is busy; please retry",
headers={"Retry-After": "1"},
)
oldest = min(
completed, key=lambda key: _authored_action_operations[key][2]
)
_authored_action_operations.pop(oldest)
task = asyncio.create_task(operation)
_authored_action_operations[idempotency_key] = (fingerprint, task, now)
async def persist_result():
result = await operation
_idempotency_ledger.complete(idempotency_key, result)
return result
task = asyncio.create_task(persist_result())
_authored_action_operations[idempotency_key] = (
fingerprint,
task,
time.monotonic(),
)
def discard_completed(completed: asyncio.Task) -> None:
existing_operation = _authored_action_operations.get(idempotency_key)
if existing_operation is not None and existing_operation[1] is completed:
_authored_action_operations.pop(idempotency_key, None)
task.add_done_callback(discard_completed)
return await asyncio.wait_for(asyncio.shield(task), timeout=timeout)
@ -1269,48 +1290,19 @@ async def create_assigned_issue(
repository, creation.title, creation.body, login, creation.label_ids
)
operation = create_issue()
if idempotency_key:
now = time.monotonic()
expired = [
key for key, (_, task, created_at) in _issue_creation_operations.items()
if task.done()
and now - created_at >= ISSUE_CREATION_IDEMPOTENCY_TTL_SECONDS
]
for key in expired:
_issue_creation_operations.pop(key, None)
fingerprint = (
repository, creation.title, creation.body, tuple(creation.label_ids)
)
existing = _issue_creation_operations.get(idempotency_key)
if existing is not None:
operation.close()
if existing[0] != fingerprint:
raise HTTPException(status_code=409, detail="Idempotency key already used")
task = existing[1]
else:
while len(_issue_creation_operations) >= ISSUE_CREATION_IDEMPOTENCY_MAX_ENTRIES:
completed = [
key for key, (_, task, _) in _issue_creation_operations.items()
if task.done()
]
if not completed:
operation.close()
raise HTTPException(
status_code=503,
detail="Issue creation is busy; please retry",
headers={"Retry-After": "1"},
)
oldest = min(
completed, key=lambda key: _issue_creation_operations[key][2]
)
_issue_creation_operations.pop(oldest)
task = asyncio.create_task(operation)
_issue_creation_operations[idempotency_key] = (fingerprint, task, now)
operation = asyncio.shield(task)
try:
result = await asyncio.wait_for(operation, timeout=ISSUE_ACTION_TIMEOUT_SECONDS)
result = await _run_idempotent_authored_action(
create_issue(),
idempotency_key=idempotency_key,
fingerprint=(
"issue-create",
repository,
creation.title,
creation.body,
tuple(creation.label_ids),
),
timeout=ISSUE_ACTION_TIMEOUT_SECONDS,
)
except HTTPException:
raise
except Exception:

View File

@ -4,13 +4,16 @@ import httpx
import pytest
from src import main
from src.idempotency import IdempotencyLedger
@pytest.fixture(autouse=True)
def clear_authored_operations():
main._authored_action_operations.clear()
main._idempotency_ledger.clear()
yield
main._authored_action_operations.clear()
main._idempotency_ledger.clear()
@pytest.mark.anyio
@ -140,3 +143,88 @@ async def test_review_replays_same_key_and_rejects_changed_head(monkeypatch):
assert [created.status_code, replayed.status_code, conflict.status_code] == [201, 201, 409]
assert calls == [("abc", "approve", "Good")]
@pytest.mark.anyio
async def test_completed_comment_replays_after_ledger_reconstruction(monkeypatch, tmp_path):
calls = 0
async def assigned(_repository, _number):
return True
async def comment(_repository, _number, body):
nonlocal calls
calls += 1
return {"id": 201, "body": body}
database = tmp_path / "actions.sqlite3"
monkeypatch.setattr(
main,
"_idempotency_ledger",
IdempotencyLedger(database, ttl_seconds=600, max_entries=256),
raising=False,
)
monkeypatch.setattr(main.gitea_proxy, "is_assigned_issue", assigned)
monkeypatch.setattr(main.gitea_proxy, "comment_on_issue", comment)
transport = httpx.ASGITransport(app=main.app)
headers = {"Idempotency-Key": "restart-comment-201"}
async with httpx.AsyncClient(transport=transport, base_url="http://test") as client:
created = await client.post(
"/api/v1/repos/stackchain/api/issues/7/comments",
json={"body": "Persist me"},
headers=headers,
)
main._authored_action_operations.clear()
monkeypatch.setattr(
main,
"_idempotency_ledger",
IdempotencyLedger(database, ttl_seconds=600, max_entries=256),
)
replayed = await client.post(
"/api/v1/repos/stackchain/api/issues/7/comments",
json={"body": "Persist me"},
headers=headers,
)
assert [created.status_code, replayed.status_code] == [201, 201]
assert replayed.json() == {"id": 201, "body": "Persist me"}
assert calls == 1
@pytest.mark.anyio
async def test_orphaned_pending_comment_fails_closed_without_upstream_retry(
monkeypatch, tmp_path
):
calls = 0
async def assigned(_repository, _number):
return True
async def comment(_repository, _number, body):
nonlocal calls
calls += 1
return {"id": 202, "body": body}
ledger = IdempotencyLedger(
tmp_path / "orphan.sqlite3", ttl_seconds=600, max_entries=256
)
fingerprint = ("issue-comment", "stackchain/api", 7, "Verify first")
assert ledger.reserve("orphan-comment-201", fingerprint).state == "reserved"
monkeypatch.setattr(main, "_idempotency_ledger", ledger)
monkeypatch.setattr(main.gitea_proxy, "is_assigned_issue", assigned)
monkeypatch.setattr(main.gitea_proxy, "comment_on_issue", comment)
transport = httpx.ASGITransport(app=main.app)
async with httpx.AsyncClient(transport=transport, base_url="http://test") as client:
response = await client.post(
"/api/v1/repos/stackchain/api/issues/7/comments",
json={"body": "Verify first"},
headers={"Idempotency-Key": "orphan-comment-201"},
)
assert response.status_code == 503
assert response.json()["detail"] == (
"This action may still be completing; verify its result before retrying"
)
assert calls == 0

View File

@ -0,0 +1,73 @@
import sqlite3
from concurrent.futures import ThreadPoolExecutor
from threading import Barrier
from src.idempotency import IdempotencyLedger
def test_completed_result_replays_after_ledger_reconstruction(tmp_path):
database = tmp_path / "idempotency.sqlite3"
first = IdempotencyLedger(database, ttl_seconds=600, max_entries=256)
reservation = first.reserve("create-201", ("issue-create", "stackchain/api", "Ship it"))
assert reservation.state == "reserved"
first.complete("create-201", {"number": 42, "title": "Ship it"})
reconstructed = IdempotencyLedger(database, ttl_seconds=600, max_entries=256)
replay = reconstructed.reserve(
"create-201", ("issue-create", "stackchain/api", "Ship it")
)
assert replay.state == "completed"
assert replay.response == {"number": 42, "title": "Ship it"}
assert sqlite3.connect(database).execute(
"SELECT status FROM idempotency_operations WHERE key = ?", ("create-201",)
).fetchone() == ("completed",)
def test_reservation_expires_completed_rows_and_enforces_capacity(tmp_path):
now = [1_000.0]
ledger = IdempotencyLedger(
tmp_path / "bounded.sqlite3",
ttl_seconds=10,
max_entries=2,
clock=lambda: now[0],
)
assert ledger.reserve("old", ("comment", "old")).state == "reserved"
ledger.complete("old", {"id": 1})
now[0] += 11
assert ledger.reserve("new-1", ("comment", "new-1")).state == "reserved"
ledger.complete("new-1", {"id": 2})
assert ledger.reserve("new-2", ("comment", "new-2")).state == "reserved"
ledger.complete("new-2", {"id": 3})
replacement = ledger.reserve("new-3", ("comment", "new-3"))
assert replacement.state == "reserved"
with sqlite3.connect(ledger.path) as connection:
keys = connection.execute(
"SELECT key FROM idempotency_operations ORDER BY key"
).fetchall()
indexes = connection.execute(
"PRAGMA index_list(idempotency_operations)"
).fetchall()
assert keys == [("new-2",), ("new-3",)]
assert any(index[1] == "idempotency_completed_at_idx" for index in indexes)
def test_independent_connections_atomically_reserve_one_operation(tmp_path):
database = tmp_path / "workers.sqlite3"
ledgers = [
IdempotencyLedger(database, ttl_seconds=600, max_entries=256)
for _ in range(2)
]
barrier = Barrier(2)
def reserve(ledger):
barrier.wait()
return ledger.reserve("worker-key-201", ("review", "stackchain/api", 7)).state
with ThreadPoolExecutor(max_workers=2) as executor:
states = list(executor.map(reserve, ledgers))
assert sorted(states) == ["pending", "reserved"]

View File

@ -5,6 +5,7 @@ import httpx
import pytest
from src import gitea_proxy, main
from src.idempotency import IdempotencyLedger
@pytest.mark.anyio
@ -226,9 +227,11 @@ async def test_edit_assigned_issue_reports_revision_conflict_without_mutation(mo
@pytest.fixture(autouse=True)
def clear_issue_creation_operations():
main._issue_creation_operations.clear()
main._authored_action_operations.clear()
main._idempotency_ledger.clear()
yield
main._issue_creation_operations.clear()
main._authored_action_operations.clear()
main._idempotency_ledger.clear()
@pytest.mark.anyio
@ -331,6 +334,61 @@ async def test_create_issue_replays_one_upstream_result_for_concurrent_idempoten
assert calls == 1
@pytest.mark.anyio
async def test_completed_issue_creation_replays_after_ledger_reconstruction(monkeypatch, tmp_path):
calls = 0
async def user():
return {"login": "timmy"}
async def available_repos():
return [{"full_name": "stackchain/api"}]
async def create(repository, title, body, assignee, label_ids):
nonlocal calls
calls += 1
return {
"id": 201, "number": 201, "title": title, "state": "open",
"repository": repository, "labels": [], "assignees": [assignee],
"updated_at": "2026-08-07T15:00:00Z",
"url": "https://forge.example/stackchain/api/issues/201",
}
database = tmp_path / "issues.sqlite3"
monkeypatch.setattr(
main,
"_idempotency_ledger",
IdempotencyLedger(database, ttl_seconds=600, max_entries=256),
)
monkeypatch.setattr(main.gitea_proxy, "current_user", user)
monkeypatch.setattr(main.gitea_proxy, "repos", available_repos)
monkeypatch.setattr(main.gitea_proxy, "create_issue", create)
transport = httpx.ASGITransport(app=main.app)
headers = {"Idempotency-Key": "restart-create-201"}
async with httpx.AsyncClient(transport=transport, base_url="http://test") as client:
created = await client.post(
"/api/v1/repos/stackchain/api/issues",
json={"title": "Durable capture"},
headers=headers,
)
main._authored_action_operations.clear()
monkeypatch.setattr(
main,
"_idempotency_ledger",
IdempotencyLedger(database, ttl_seconds=600, max_entries=256),
)
replayed = await client.post(
"/api/v1/repos/stackchain/api/issues",
json={"title": "Durable capture"},
headers=headers,
)
assert [created.status_code, replayed.status_code] == [201, 201]
assert replayed.json()["number"] == 201
assert calls == 1
@pytest.mark.anyio
async def test_create_issue_rejects_changed_payload_for_an_existing_idempotency_key(monkeypatch):
calls = []
@ -412,7 +470,9 @@ async def test_create_issue_retry_recovers_result_after_the_first_request_times_
@pytest.mark.anyio
async def test_create_issue_idempotency_registry_evicts_oldest_entry_at_size_limit(monkeypatch):
async def test_create_issue_durable_ledger_evicts_completed_entry_at_size_limit(
monkeypatch, tmp_path
):
async def user():
return {"login": "timmy"}
@ -427,7 +487,10 @@ async def test_create_issue_idempotency_registry_evicts_oldest_entry_at_size_lim
"url": "https://forge.example/stackchain/api/issues/17",
}
monkeypatch.setattr(main, "ISSUE_CREATION_IDEMPOTENCY_MAX_ENTRIES", 2)
ledger = IdempotencyLedger(
tmp_path / "bounded-issues.sqlite3", ttl_seconds=600, max_entries=2
)
monkeypatch.setattr(main, "_idempotency_ledger", ledger)
monkeypatch.setattr(main.gitea_proxy, "current_user", user)
monkeypatch.setattr(main.gitea_proxy, "repos", available_repos)
monkeypatch.setattr(main.gitea_proxy, "create_issue", create)
@ -441,8 +504,11 @@ async def test_create_issue_idempotency_registry_evicts_oldest_entry_at_size_lim
)
assert response.status_code == 201
assert len(main._issue_creation_operations) == 2
assert "capture-177-bounded-0" not in main._issue_creation_operations
replay = ledger.reserve(
"capture-177-bounded-0",
("issue-create", "stackchain/api", "Capture 0", "", ()),
)
assert replay.state == "reserved"
@pytest.mark.anyio