Build reusable bot framework
Some checks failed
CI / test (push) Has been cancelled
CI / compose-smoke (push) Has been cancelled

This commit is contained in:
Chelsea Lee
2026-07-19 21:53:24 -05:00
parent 7ecc1107b2
commit fbdf33e894
66 changed files with 8428 additions and 0 deletions

357
core/outbox.py Normal file
View File

@@ -0,0 +1,357 @@
"""Durable, lease-based outbound message operations."""
from contextlib import contextmanager
from datetime import datetime, timezone
import uuid
from psycopg2.extras import Json
from core import postgres
DEFAULT_MAX_ATTEMPTS = 3
DEFAULT_LEASE_SECONDS = 300
DEFAULT_RETRY_SECONDS = 30
MAX_RETRY_SECONDS = 900
@contextmanager
def _using_cursor(cursor=None):
if cursor is not None:
yield cursor
return
with postgres.get_cursor() as owned_cursor:
yield owned_cursor
def _timestamp(value, field="timestamp"):
if isinstance(value, str):
value = datetime.fromisoformat(value.replace("Z", "+00:00"))
if not isinstance(value, datetime):
raise ValueError(f"{field} must be a datetime or ISO-8601 string")
if value.tzinfo is None:
value = value.replace(tzinfo=timezone.utc)
return value.astimezone(timezone.utc)
def _positive(value, field, maximum=None):
try:
value = int(value)
except (TypeError, ValueError) as error:
raise ValueError(f"{field} must be a whole number") from error
if value < 1 or (maximum is not None and value > maximum):
suffix = f" and at most {maximum}" if maximum is not None else ""
raise ValueError(f"{field} must be at least 1{suffix}")
return value
def _row(cursor):
record = cursor.fetchone()
return dict(record) if record else None
def enqueue_message(
user_uuid,
channel,
payload,
idempotency_key,
available_at=None,
max_attempts=DEFAULT_MAX_ATTEMPTS,
message_id=None,
cursor=None,
):
"""Queue a message, returning the existing row for a repeated key."""
if not user_uuid:
raise ValueError("user_uuid is required")
if not isinstance(channel, str) or not channel.strip():
raise ValueError("channel is required")
if not idempotency_key:
raise ValueError("idempotency_key is required")
if payload is None:
raise ValueError("payload is required")
max_attempts = _positive(max_attempts, "max_attempts", 100)
available_at = available_at or datetime.now(timezone.utc)
values = {
"id": str(message_id or uuid.uuid4()),
"user_uuid": user_uuid,
"channel": channel.strip(),
"payload": Json(payload),
"idempotency_key": idempotency_key,
"available_at": _timestamp(available_at, "available_at"),
"max_attempts": max_attempts,
}
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
INSERT INTO outbound_messages (
id, user_uuid, channel, payload, idempotency_key,
available_at, max_attempts
) VALUES (
%(id)s, %(user_uuid)s, %(channel)s, %(payload)s,
%(idempotency_key)s, %(available_at)s, %(max_attempts)s
)
ON CONFLICT (idempotency_key)
DO UPDATE SET idempotency_key = EXCLUDED.idempotency_key
WHERE outbound_messages.user_uuid = EXCLUDED.user_uuid
AND outbound_messages.channel = EXCLUDED.channel
RETURNING *
""",
values,
)
return _row(active_cursor)
def get_message(message_id, cursor=None):
"""Return one outbound message by UUID."""
try:
message_id = str(uuid.UUID(str(message_id)))
except (TypeError, ValueError, AttributeError):
return None
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"SELECT * FROM outbound_messages WHERE id = %s", (message_id,)
)
return _row(active_cursor)
def list_messages(user_uuid=None, status=None, channel=None, limit=100, cursor=None):
"""List outbound messages newest first with optional filters."""
limit = _positive(limit, "limit", 500)
clauses = []
params = []
if user_uuid is not None:
clauses.append("user_uuid = %s")
params.append(user_uuid)
if status is not None:
clauses.append("status = %s")
params.append(status)
if channel is not None:
clauses.append("channel = %s")
params.append(channel)
where = f"WHERE {' AND '.join(clauses)}" if clauses else ""
params.append(limit)
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
f"""
SELECT * FROM outbound_messages
{where}
ORDER BY created_at DESC
LIMIT %s
""",
params,
)
return [dict(record) for record in active_cursor.fetchall()]
def claim_messages(
worker_id,
channel=None,
limit=10,
lease_seconds=DEFAULT_LEASE_SECONDS,
cursor=None,
):
"""Atomically lease deliverable messages while skipping other workers."""
if not worker_id:
raise ValueError("worker_id is required")
limit = _positive(limit, "limit", 100)
lease_seconds = _positive(lease_seconds, "lease_seconds")
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
WITH exhausted AS (
SELECT id
FROM outbound_messages
WHERE status = 'delivering'
AND (lease_until IS NULL OR lease_until <= NOW())
AND attempts >= max_attempts
FOR UPDATE SKIP LOCKED
)
UPDATE outbound_messages AS message
SET status = 'failed',
lease_until = NULL,
leased_by = NULL,
last_error = COALESCE(last_error, 'lease expired'),
updated_at = NOW()
FROM exhausted
WHERE message.id = exhausted.id
"""
)
active_cursor.execute(
"""
WITH candidates AS (
SELECT id
FROM outbound_messages
WHERE available_at <= NOW()
AND attempts < max_attempts
AND (
status = 'pending'
OR (
status = 'delivering'
AND (lease_until IS NULL OR lease_until <= NOW())
)
)
AND (%(channel)s IS NULL OR channel = %(channel)s)
ORDER BY available_at, created_at
FOR UPDATE SKIP LOCKED
LIMIT %(limit)s
)
UPDATE outbound_messages AS message
SET status = 'delivering',
attempts = message.attempts + 1,
leased_by = %(worker_id)s,
lease_until = NOW() + (%(lease_seconds)s * INTERVAL '1 second'),
updated_at = NOW()
FROM candidates
WHERE message.id = candidates.id
RETURNING message.*
""",
{
"channel": channel,
"limit": limit,
"worker_id": worker_id,
"lease_seconds": lease_seconds,
},
)
return [dict(record) for record in active_cursor.fetchall()]
def renew_message_lease(
message_id, worker_id, lease_seconds=DEFAULT_LEASE_SECONDS, cursor=None
):
"""Extend a delivery lease owned by the requesting worker."""
lease_seconds = _positive(lease_seconds, "lease_seconds")
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
UPDATE outbound_messages
SET lease_until = NOW() + (%s * INTERVAL '1 second'), updated_at = NOW()
WHERE id = %s AND status = 'delivering' AND leased_by = %s
RETURNING *
""",
(lease_seconds, message_id, worker_id),
)
return _row(active_cursor)
def mark_delivered(
message_id, worker_id, external_message_id=None, cursor=None
):
"""Mark a message delivered when its lease is owned by the worker."""
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
UPDATE outbound_messages
SET status = 'delivered', delivered_at = NOW(), updated_at = NOW(),
leased_by = NULL, lease_until = NULL, last_error = NULL,
external_message_id = %s
WHERE id = %s AND status = 'delivering' AND leased_by = %s
RETURNING *
""",
(external_message_id, message_id, worker_id),
)
return _row(active_cursor)
def retry_message(
message_id,
worker_id,
error,
retry_seconds=DEFAULT_RETRY_SECONDS,
max_retry_seconds=MAX_RETRY_SECONDS,
cursor=None,
):
"""Fail or reschedule an owned delivery with exponential backoff."""
retry_seconds = _positive(retry_seconds, "retry_seconds")
max_retry_seconds = _positive(max_retry_seconds, "max_retry_seconds")
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
SELECT attempts, max_attempts
FROM outbound_messages
WHERE id = %s AND status = 'delivering' AND leased_by = %s
FOR UPDATE
""",
(message_id, worker_id),
)
current = active_cursor.fetchone()
if not current:
return None
exhausted = current["attempts"] >= current["max_attempts"]
delay = min(
max_retry_seconds,
retry_seconds * (2 ** min(max(current["attempts"] - 1, 0), 30)),
)
active_cursor.execute(
"""
UPDATE outbound_messages
SET status = %(status)s,
available_at = CASE
WHEN %(exhausted)s THEN available_at
ELSE NOW() + (%(delay)s * INTERVAL '1 second')
END,
leased_by = NULL,
lease_until = NULL,
last_error = %(error)s,
updated_at = NOW()
WHERE id = %(message_id)s
RETURNING *
""",
{
"status": "failed" if exhausted else "pending",
"exhausted": exhausted,
"delay": delay,
"error": str(error)[:4000],
"message_id": message_id,
},
)
return _row(active_cursor)
def cancel_message(message_id, user_uuid=None, cursor=None):
"""Cancel one undelivered message, optionally enforcing its owner."""
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
"""
UPDATE outbound_messages
SET status = 'cancelled', cancelled_at = NOW(), updated_at = NOW(),
leased_by = NULL, lease_until = NULL
WHERE id = %(message_id)s
AND status IN ('pending', 'delivering')
AND (%(user_uuid)s IS NULL OR user_uuid = %(user_uuid)s)
RETURNING *
""",
{"message_id": message_id, "user_uuid": user_uuid},
)
return _row(active_cursor)
def cancel_messages(
user_uuid=None, channel=None, idempotency_key=None, cursor=None
):
"""Cancel matching undelivered messages; at least one filter is required."""
filters = {
"user_uuid": user_uuid,
"channel": channel,
"idempotency_key": idempotency_key,
}
clauses = [f"{name} = %({name})s" for name, value in filters.items() if value is not None]
if not clauses:
raise ValueError("at least one cancellation filter is required")
with _using_cursor(cursor) as active_cursor:
active_cursor.execute(
f"""
UPDATE outbound_messages
SET status = 'cancelled', cancelled_at = NOW(), updated_at = NOW(),
leased_by = NULL, lease_until = NULL
WHERE status IN ('pending', 'delivering')
AND {' AND '.join(clauses)}
RETURNING *
""",
filters,
)
return [dict(record) for record in active_cursor.fetchall()]