Review

HA birch-sap ledger: multi-replica API on NFS SQLite + WAL

Sapline co-op records gallons pulled from ridgeline taps during sugaring season. Ops wants the FastAPI ledger HA-ready for the March surge without standing up Postgres. This PR points all replicas at a single SQLite file on the NFS PVC, turns on WAL + busy_timeout for 'concurrent readers across pods', and flips check_same_thread=False so uvicorn workers can share the connection factory.

SQLiteTier 5sqlitewalnfsmulti-instance

Click a line to flag it, pick one or more labels, then submit. If the change looks correct, approve it.

sapline/db.py+11-3

1	1		import sqlite3
2	2		from contextlib import contextmanager
3	3		from pathlib import Path
4	4
5	5		from sapline.config import settings
6	6
7		-	# Single-node: local disk under the API pod.
8		-	DB_PATH = Path("/var/lib/sapline/ledger.db")
	7	+	# Shared ledger for every replica — mount is RWX NFS (EFS) so pods see one file.
	8	+	DB_PATH = Path("/mnt/nfs/sapline/ledger.db")
9	9
10	10
11	11		def connect() -> sqlite3.Connection:
12		-	conn = sqlite3.connect(DB_PATH)
	12	+	# check_same_thread=False: uvicorn workers reuse this factory across threads.
	13	+	conn = sqlite3.connect(
	14	+	DB_PATH,
	15	+	check_same_thread=False,
	16	+	timeout=30.0, # busy_timeout via connect timeout
	17	+	)
13	18		conn.row_factory = sqlite3.Row
	19	+	# WAL lets multiple pods read while one writes — no more SQLITE_BUSY storms.
	20	+	conn.execute("PRAGMA journal_mode=WAL")
	21	+	conn.execute("PRAGMA synchronous=NORMAL")
14	22		return conn
15	23
16	24
17	25		@contextmanager
18	26		def session():
19	27		conn = connect()
20	28		try:
21	29		yield conn
22	30		conn.commit()
23	31		except Exception:
24	32		conn.rollback()
25	33		raise
26	34		finally:
27	35		conn.close()