522cfbfda677a2b07a3e57bdc436933ac31ccf41
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Joseph Doherty
|
522cfbfda6 |
fix(localdb): flush the HLC anchor on deregister, bound the sync inbox (0.2.1)
Two adversarial-review findings against 0.2.0. Wire-compatible with 0.1.x/0.2.0 peers: no proto change, no lib_schema_version bump, no on-disk schema change. F4 — DeregisterReplicated destroyed the HLC clock's only durable crash anchor. Clock recovery on open is MAX(hlc_high_water, MAX(oplog.hlc), MAX(row_version.hlc), last_seen_hlc); hlc_high_water is written only by Dispose, and deregistration deletes the table's rows from BOTH ledgers. On a file whose registered tables have all been deregistered, an ungraceful exit (crash, SIGKILL, power loss) therefore left nothing above zero to recover from: the reopened clock restarts from wall-clock time and can re-issue stamps below HLCs peers already hold, which LWW discards as older. The current clock is now flushed inside the deregistration transaction, before the prune, as a separate command so it cannot inflate the "did anything change?" answer. FlushHlcHighWater takes an optional transaction (SQLite rejects a command with no transaction on a connection that has one) and is now MAX-guarded — every writer of that column is flushing a monotonic clock, so a lower value can only be stale. F3 — the sync inbox was Channel.CreateUnbounded, making inbound memory a function of what the PEER sends. The bad case is not hypothetical: a snapshot streams every row of every registered table, needs_snapshot is per-database, and both sides of a pair that enables replication late baseline unconditionally — so a mutual full snapshot of a large dataset (an event log near its storage cap) was buffered whole, in memory, on both nodes at once. The inbox is now bounded at 64 messages with wait-mode backpressure, so the ceiling is ~64 x MaxBatchBytes — a configured number rather than a peer-controlled one — and a full inbox stops draining the gRPC stream, which closes the peer's flow-control window. The pump writes with WriteAsync, not TryWrite: on a full bounded channel TryWrite drops the message, and every inbound message is protocol state. Bounding alone would have deadlocked the very case it was aimed at. The snapshot a node owes was streamed from INSIDE the handshake, before the receive loop existed, so neither side drained its inbox while sending; a mutual snapshot converged only because the inbox was unbounded. The snapshot now runs as the first phase of the outbound task, started alongside the receive loop, so the inbox is drained for the whole life of the session. Interleaving needed no new safety: snapshot rows and deltas both merge through LWW and the snapshot's watermark write is MAX-guarded. Ordering within the outbound lane is preserved (the pump still starts only after the snapshot, from its as-of seq). Per-table needs_snapshot (the finding's optional third part) is NOT done: the flag lives in __localdb_peer_state and narrowing it needs an on-disk schema change, which this fix deliberately avoids. Documented as a follow-up, and the cost is now written down where it is decided — RegisterReplicated(baselineExistingRows) remarks and the README both say that baselining one table re-streams every registered table in both directions, and that high-volume tables should be trimmed before enabling replication late. 188/188 tests pass (was 184). New coverage: the deregistered clock anchor surviving an ungraceful reopen (a second instance over the same file with no Dispose — fails without the flush, recovering at 0), the flush not counting as work done, bounded-inbox backpressure under a slow reader (the pump stalls at the bound and loses nothing, proven by contiguous in-order delivery), and a mutual snapshot over a capacity-1 transport converging — that last one deadlocks and times out against the old handshake-phase snapshot ordering. Version 0.2.1. |
||
|
Joseph Doherty
|
9377fa1678 |
perf(localdb): dereg API, O(1) backlog depth, byte-budget batching, prepare-once apply (0.2.0)
WP3.3 of the ScadaBridge architectural-review remediation. Six changes, all wire-compatible with 0.1.x peers — no proto change, no lib_schema_version bump. The on-disk bookkeeping schema goes to v2 and is upgraded in place when a file is opened. DeregisterReplicated(table). There was no way to stop capturing once a file had been registered, so a node that does not replicate kept paying three triggers inside every write transaction, appending to an oplog nothing would ever drain. Drops the triggers and prunes the table's oplog + row-version rows in one transaction; idempotent, and deliberately usable on a table this process never registered, since stale triggers left by an EARLIER build are the primary case. Deregistration must be symmetric across a pair, and already is enforced: the handshake compares registered-table digests fail-closed, so a node that drops a table its peer still replicates stops syncing loudly rather than diverging. RegisterReplicated(table, baselineExistingRows: true). Closes the late-opt-in hole: the snapshot streamer pages from __localdb_row_version, which only triggers populate, so a node that started replicating LATE never sent its pre-existing rows — silently, forever. Seeding alone would not have fixed it, because a node whose oplog starts at seq 1 presents no gap at the handshake and is told it needs no snapshot; the seed therefore also flags needs_snapshot, and the existing resync machinery carries the rows. The floor is HLC 0 stamped with this node's id: every genuine HLC is a UTC millisecond shifted left 16 bits and so is strictly greater, which makes a baselined row lose every comparison against a real remote write and win only where the peer has no version of that key at all. Two nodes baselining the same key both hold 0 and resolve by the node-id tie-break — convergent, and nothing better is available for two rows of unordered antiquity. Opt-in, so the documented CDC contract is unchanged unless asked. Pruning the ledger on deregister is safe precisely because re-registration can re-baseline; the two APIs are a pair. O(1) backlog depth. GetOplogDepth ran COUNT(*) over the entire unacked backlog, once a minute from maintenance AND on every metrics scrape — a full scan of a million rows exactly when a node is already behind. Now MAX(seq) - MAX(last_acked_seq, MIN(seq) - 1), floored at 0: three b-tree seeks, measured at 58 VM steps against 100,020 for the count on a 50k backlog. Exact, because seq is AUTOINCREMENT and every prune path removes a prefix (ack pruning by watermark, the row cap by count, the age cap by an HLC cutoff that rises with seq). The MIN clamp is what keeps it exact after a cap prune has moved the horizon above a stalled watermark — without it the caps would re-flag on every tick forever. Byte-budget batching. MaxBatchSize was the only cap on an outbound message, which made it a proxy for a byte budget that consumers had to tune against their widest row: ScadaBridge pinned it to 16 because 70 KB config_json times the 500 default is ~35 MB against a 4 MB gRPC limit. MaxBatchBytes (default 2 MB) now bounds delta and snapshot messages by summed serialized size, with the row count demoted to a secondary cap. A row larger than the budget is sent alone rather than stalling the stream on it. Since more, smaller messages would otherwise mean more apply transactions and more acks, inbound delta batches and acks ALREADY queued are coalesced — one transaction and one ack per burst, bounded by the sender's own caps, and a no-op under a trickle. LwwApplier prepares each statement once. It went through the generic transaction façade, which builds a fresh SqliteCommand per call: three SQL parses per applied row, on the path that drains a reconnect backlog. Statements are now held for the batch and rebound. Dropped __localdb_oplog_hlc. It indexed (table_name, pk_json, hlc), which no query has ever used — every oplog access path is by seq or a full hlc-predicate delete — while costing a b-tree insert inside every captured consumer write. 184/184 tests pass (was 149). New coverage: schema upgrade drops the index and preserves identity/watermarks/data, deregistration idempotency and trigger absence and cross-process cleanup, baseline seeding at the floor with LWW safety proved through the real applier, depth asserted equal to the counting query across every prune scenario plus a constant-time assertion over a 200k backlog, byte-budget chunking including the one-giant-row edge on both the delta and snapshot paths, prepare-once statement counting, and coalescing that must not swallow the message that ends its run. Version 0.2.0. |
||
|
Joseph Doherty
|
df2518611f | fix(localdb): capture-layer review fixes — BLOB fail-closed, identifier escaping, master lock, PK-change tombstone, digest canon | ||
|
Joseph Doherty
|
f4063f811f | feat(localdb): opt-in table registration + capture triggers (oplog + row_version, applying guard) |