feat(localdb): LWW applier (hlc+node_id tie-break, applying guard, dead-letter, atomic watermark)

Claude-Session: https://claude.ai/code/session_01BL2Vu1ESDQ9SCN4gVKkdts
This commit is contained in:
Joseph Doherty
2026-07-17 22:03:24 -04:00
parent 3cf8cabaf9
commit d7cbdce65c
5 changed files with 476 additions and 1 deletions
@@ -0,0 +1,308 @@
using Microsoft.Data.Sqlite;
using ZB.MOM.WW.LocalDb.Internal;
using ZB.MOM.WW.LocalDb.Replication;
using ZB.MOM.WW.LocalDb.Replication.Internal;
namespace ZB.MOM.WW.LocalDb.Tests;
public sealed class LwwApplierTests : IDisposable
{
private const string Pk1 = "{\"id\":1}";
private readonly string _path = Path.Combine(Path.GetTempPath(), Guid.NewGuid() + ".db");
public void Dispose()
{
SqliteConnection.ClearAllPools();
if (File.Exists(_path))
File.Delete(_path);
}
private async Task<SqliteLocalDb> NewOrdersDb()
{
var db = new SqliteLocalDb(new LocalDbOptions { Path = _path });
await db.ExecuteAsync("CREATE TABLE orders (id INTEGER PRIMARY KEY, sku TEXT, qty INTEGER)");
db.RegisterReplicated("orders");
return db;
}
private static async Task<long> RowVersionHlc(SqliteLocalDb db, string pkJson)
{
var r = await db.QueryAsync(
"SELECT hlc FROM __localdb_row_version WHERE table_name = 'orders' AND pk_json = @pk",
x => x.GetInt64(0), new { pk = pkJson });
return r[0];
}
private static async Task<(string? Sku, long? Qty)?> ReadOrder(SqliteLocalDb db, long id)
{
var r = await db.QueryAsync(
"SELECT sku, qty FROM orders WHERE id = @id",
x => (x.IsDBNull(0) ? null : x.GetString(0), (long?)(x.IsDBNull(1) ? null : x.GetInt64(1))),
new { id });
return r.Count == 0 ? null : r[0];
}
private static async Task<long> OplogCount(SqliteLocalDb db)
{
var r = await db.QueryAsync("SELECT COUNT(*) FROM __localdb_oplog", x => x.GetInt64(0));
return r[0];
}
private static OplogEntryRecord Entry(
long seq, string pkJson, string? rowJson, long hlc, string nodeId, bool tombstone = false) =>
new(seq, "orders", pkJson, rowJson, hlc, nodeId, tombstone);
[Fact]
public async Task Apply_NewerRemote_Wins()
{
using var db = await NewOrdersDb();
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var localHlc = await RowVersionHlc(db, Pk1);
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(1, Pk1, "{\"id\":1,\"sku\":\"REMOTE\",\"qty\":99}", localHlc + 1000, "remote-node")], default);
Assert.Equal(1, result.Applied);
var row = await ReadOrder(db, 1);
Assert.Equal(("REMOTE", (long?)99), row);
var rv = await db.QueryAsync(
"SELECT hlc, node_id FROM __localdb_row_version WHERE pk_json = @pk",
x => (Hlc: x.GetInt64(0), Node: x.GetString(1)), new { pk = Pk1 });
Assert.Equal(localHlc + 1000, rv[0].Hlc);
Assert.Equal("remote-node", rv[0].Node);
}
[Fact]
public async Task Apply_OlderRemote_Discarded_StillCounted()
{
using var db = await NewOrdersDb();
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var localHlc = await RowVersionHlc(db, Pk1);
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(7, Pk1, "{\"id\":1,\"sku\":\"OLD\",\"qty\":1}", localHlc - 1000, "remote-node")], default);
Assert.Equal(0, result.Applied);
Assert.Equal(1, result.Discarded);
Assert.Equal(7, result.AppliedThruSeq);
Assert.Equal(("LOCAL", (long?)5), await ReadOrder(db, 1));
var peer = await db.QueryAsync(
"SELECT last_applied_remote_seq FROM __localdb_peer_state WHERE id = 1", x => x.GetInt64(0));
Assert.Equal(7, peer[0]);
}
[Fact]
public async Task Apply_EqualHlc_NodeIdTieBreak_Deterministic()
{
// Greater remote node_id wins the tie.
using (var db = await NewOrdersDb())
{
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var localHlc = await RowVersionHlc(db, Pk1);
var greater = db.NodeId + "z";
Assert.True(string.CompareOrdinal(greater, db.NodeId) > 0);
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(1, Pk1, "{\"id\":1,\"sku\":\"WINS\",\"qty\":9}", localHlc, greater)], default);
Assert.Equal(1, result.Applied);
Assert.Equal(("WINS", (long?)9), await ReadOrder(db, 1));
}
SqliteConnection.ClearAllPools();
File.Delete(_path);
// Smaller remote node_id loses the tie.
using (var db = await NewOrdersDb())
{
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var localHlc = await RowVersionHlc(db, Pk1);
var smaller = "!" + db.NodeId;
Assert.True(string.CompareOrdinal(smaller, db.NodeId) < 0);
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(1, Pk1, "{\"id\":1,\"sku\":\"LOSES\",\"qty\":9}", localHlc, smaller)], default);
Assert.Equal(0, result.Applied);
Assert.Equal(1, result.Discarded);
Assert.Equal(("LOCAL", (long?)5), await ReadOrder(db, 1));
}
}
[Fact]
public async Task Apply_NewRow_Inserts()
{
using var db = await NewOrdersDb();
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(3, Pk1, "{\"id\":1,\"sku\":\"NEW\",\"qty\":42}", 5_000_000, "remote-node")], default);
Assert.Equal(1, result.Applied);
Assert.Equal(("NEW", (long?)42), await ReadOrder(db, 1));
}
[Fact]
public async Task Apply_Tombstone_DeletesRow()
{
using var db = await NewOrdersDb();
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var localHlc = await RowVersionHlc(db, Pk1);
var result = await new LwwApplier(db).ApplyBatchAsync(
[Entry(4, Pk1, null, localHlc + 1000, "remote-node", tombstone: true)], default);
Assert.Equal(1, result.Applied);
Assert.Null(await ReadOrder(db, 1));
var rv = await db.QueryAsync(
"SELECT is_tombstone, hlc FROM __localdb_row_version WHERE pk_json = @pk",
x => (Tomb: x.GetInt64(0), Hlc: x.GetInt64(1)), new { pk = Pk1 });
Assert.Equal(1, rv[0].Tomb);
Assert.Equal(localHlc + 1000, rv[0].Hlc);
}
[Fact]
public async Task Apply_DoesNotFireCaptureTriggers()
{
using var db = await NewOrdersDb();
await db.ExecuteAsync("INSERT INTO orders (id, sku, qty) VALUES (1, 'LOCAL', 5)");
var before = await OplogCount(db);
await new LwwApplier(db).ApplyBatchAsync(
[
Entry(1, Pk1, "{\"id\":1,\"sku\":\"R\",\"qty\":9}", 9_000_000, "remote-node"),
Entry(2, "{\"id\":2}", "{\"id\":2,\"sku\":\"R2\",\"qty\":7}", 9_000_001, "remote-node"),
], default);
Assert.Equal(before, await OplogCount(db));
}
[Fact]
public async Task Apply_Batch_Atomic_WatermarkSameTxn()
{
using var db = await NewOrdersDb();
var result = await new LwwApplier(db).ApplyBatchAsync(
[
Entry(10, Pk1, "{\"id\":1,\"sku\":\"A\",\"qty\":1}", 1_000_000, "remote-node"),
Entry(11, "{\"id\":2}", "{\"id\":2,\"sku\":\"B\",\"qty\":2}", 1_000_001, "remote-node"),
], default);
Assert.Equal(2, result.Applied);
Assert.Equal(11, result.AppliedThruSeq);
// Data and the watermark are both visible => they committed in one transaction.
Assert.Equal(("A", (long?)1), await ReadOrder(db, 1));
Assert.Equal(("B", (long?)2), await ReadOrder(db, 2));
var peer = await db.QueryAsync(
"SELECT last_applied_remote_seq, last_seen_hlc FROM __localdb_peer_state WHERE id = 1",
x => (Seq: x.GetInt64(0), Hlc: x.GetInt64(1)));
Assert.Equal(11, peer[0].Seq);
Assert.Equal(1_000_001, peer[0].Hlc);
}
[Fact]
public async Task Apply_UnregisteredTable_Throws_NothingApplied()
{
using var db = await NewOrdersDb();
var batch = new OplogEntryRecord[]
{
Entry(20, Pk1, "{\"id\":1,\"sku\":\"A\",\"qty\":1}", 1_000_000, "remote-node"),
new(21, "widgets", "{\"id\":9}", "{\"id\":9}", 1_000_001, "remote-node", false),
};
await Assert.ThrowsAsync<LocalDbSchemaMismatchException>(
() => new LwwApplier(db).ApplyBatchAsync(batch, default));
// The valid entry that preceded the violation must not have been applied.
Assert.Null(await ReadOrder(db, 1));
var peer = await db.QueryAsync(
"SELECT last_applied_remote_seq FROM __localdb_peer_state WHERE id = 1", x => x.GetInt64(0));
Assert.Equal(0, peer[0]);
}
[Fact]
public async Task Apply_PoisonEntry_DeadLettered_BatchContinues()
{
using var db = await NewOrdersDb();
var result = await new LwwApplier(db).ApplyBatchAsync(
[
Entry(30, Pk1, "NOT VALID JSON", 1_000_000, "remote-node"),
Entry(31, "{\"id\":2}", "{\"id\":2,\"sku\":\"OK\",\"qty\":8}", 1_000_001, "remote-node"),
], default);
Assert.Equal(1, result.DeadLettered);
Assert.Equal(1, result.Applied);
Assert.Equal(("OK", (long?)8), await ReadOrder(db, 2));
Assert.Null(await ReadOrder(db, 1));
var dead = await db.QueryAsync(
"SELECT pk_json, error FROM __localdb_dead_letter",
x => (Pk: x.GetString(0), Err: x.GetString(1)));
Assert.Single(dead);
Assert.Equal(Pk1, dead[0].Pk);
Assert.NotEmpty(dead[0].Err);
}
[Fact]
public async Task Apply_IsIdempotent()
{
using var db = await NewOrdersDb();
var batch = new OplogEntryRecord[]
{
Entry(40, Pk1, "{\"id\":1,\"sku\":\"X\",\"qty\":3}", 7_000_000, "remote-node"),
new(41, "orders", "{\"id\":2}", "{\"id\":2,\"sku\":\"Y\",\"qty\":4}", 7_000_001, "remote-node", false),
};
var applier = new LwwApplier(db);
var first = await applier.ApplyBatchAsync(batch, default);
Assert.Equal(2, first.Applied);
var second = await applier.ApplyBatchAsync(batch, default);
Assert.Equal(0, second.Applied);
Assert.Equal(2, second.Discarded);
Assert.Equal(("X", (long?)3), await ReadOrder(db, 1));
Assert.Equal(("Y", (long?)4), await ReadOrder(db, 2));
var rows = await db.QueryAsync("SELECT COUNT(*) FROM orders", x => x.GetInt64(0));
Assert.Equal(2, rows[0]);
}
[Fact]
public async Task Apply_ObservesRemoteHlc()
{
using var db = await NewOrdersDb();
var maxHlc = (DateTimeOffset.UtcNow.AddDays(365).ToUnixTimeMilliseconds() << 16);
await new LwwApplier(db).ApplyBatchAsync(
[Entry(1, Pk1, "{\"id\":1,\"sku\":\"R\",\"qty\":1}", maxHlc, "remote-node")], default);
Assert.True(db.Clock.Next() > maxHlc);
}
[Fact]
public async Task Apply_AdvancesPeerState()
{
using var db = await NewOrdersDb();
await new LwwApplier(db).ApplyBatchAsync(
[
Entry(50, Pk1, "{\"id\":1,\"sku\":\"A\",\"qty\":1}", 3_000_000, "remote-node"),
Entry(52, "{\"id\":2}", "{\"id\":2,\"sku\":\"B\",\"qty\":2}", 3_000_050, "remote-node"),
], default);
var peer = await db.QueryAsync(
"SELECT last_applied_remote_seq, last_seen_hlc, last_sync_utc FROM __localdb_peer_state WHERE id = 1",
x => (Seq: x.GetInt64(0), Hlc: x.GetInt64(1), Utc: x.IsDBNull(2) ? null : x.GetString(2)));
Assert.Equal(52, peer[0].Seq);
Assert.Equal(3_000_050, peer[0].Hlc);
Assert.NotNull(peer[0].Utc);
Assert.EndsWith("Z", peer[0].Utc);
}
}