dohertj2/ScadaBridge
1
0
Fork 0
Code Issues 2 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

test(localdb): two-node site-pair convergence over a real gRPC transport

Browse Source 
Task 10 of the LocalDb Phase 1 adoption plan. Two ScadaBridge site nodes
replicating the consolidated site database over loopback Kestrel h2c, through
the REAL fail-closed auth interceptor - not a stand-in.

This is the test that answers the question Phase 1 exists to answer. Every piece
upstream of it - schema helpers, DI wiring, the interceptor - can be
individually green while the pair still fails to converge.

It initializes both databases through SiteLocalDbSetup.OnReady rather than a
hand-written schema, so the tables, primary keys and registration ORDER under
test are the ones the host actually runs. A local schema would prove only that
the test agrees with itself.

Scenarios: A->B, B->A (bidirectional even though only A dials, which is what
makes failover safe in either direction), LWW convergence on a contended
operation, the event union across both nodes, and a peer-offline-then-rejoin
catch-up.

The event-union scenario is the one that pins the GUID id change: under the old
autoincrement scheme both nodes independently mint id=1,2,3..., and
last-writer-wins on the primary key would silently DESTROY one node's events
rather than merge them. Asserting the union count is what catches that.

The whole suite passes in ~0.6s, which is fast enough to be suspicious of a
vacuous pass, so it was verified red-first the hard way: deliberately mismatching
the two nodes' ApiKey turns all 5 red with 2m30s of timeouts. The tests really do
run through the interceptor and the real transport.

Node B's database survives its host teardown (registered as a pre-constructed
instance, which MS.DI does not dispose), so the rejoin scenario is a genuine
rejoin rather than a fresh node.

Offline - no docker, no external services.

Verified: build 0 warnings; IntegrationTests 80/80.

Claude-Session: https://claude.ai/code/session_01BL2Vu1ESDQ9SCN4gVKkdts
This commit is contained in:
Joseph Doherty
2026-07-19 09:42:21 -04:00
parent 98b94771ae
commit 3c395d794a
1 changed files with 348 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ZB.MOM.WW.ScadaBridge.IntegrationTests/LocalDbSitePairConvergenceTests.cs
+348
View File
@@ -0,0 +1,348 @@
using System.Net;
using Microsoft.AspNetCore.Builder;
using Microsoft.AspNetCore.Hosting;
using Microsoft.AspNetCore.Hosting.Server;
using Microsoft.AspNetCore.Hosting.Server.Features;
using Microsoft.AspNetCore.Server.Kestrel.Core;
using Microsoft.Extensions.Configuration;
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Hosting;
using ZB.MOM.WW.LocalDb;
using ZB.MOM.WW.LocalDb.Replication;
using ZB.MOM.WW.ScadaBridge.Host;
namespace ZB.MOM.WW.ScadaBridge.IntegrationTests;
/// <summary>
/// Serializes the site-pair convergence tests against each other: each one stands up a real
/// Kestrel listener plus two SQLite files, and running them concurrently under CI
/// contention is a flakiness risk.
/// </summary>
[CollectionDefinition("LocalDbSitePairConvergence")]
public sealed class LocalDbSitePairConvergenceCollection;
/// <summary>
/// Two ScadaBridge site nodes replicating the consolidated site database over a REAL
/// loopback gRPC transport, through the REAL fail-closed auth interceptor.
/// </summary>
/// <remarks>
/// <para>
/// This is the test that answers the question Phase 1 exists to answer: does a site node
/// pair actually stop losing operation-tracking and site-event state? Everything upstream
/// of it — schema helpers, DI wiring, the interceptor — can be individually green while the
/// pair still fails to converge.
/// </para>
/// <para>
/// It uses <see cref="SiteLocalDbSetup.OnReady"/>, not a hand-written schema, so the tables,
/// their primary keys, and the registration ORDER under test are the ones the host actually
/// runs. A separate schema here would prove only that the test agrees with itself.
/// </para>
/// <para>
/// Offline: no docker, no external services. Loopback Kestrel with h2c.
/// </para>
/// </remarks>
[Collection("LocalDbSitePairConvergence")]
public sealed class LocalDbSitePairConvergenceTests : IAsyncLifetime
{
private const string SharedApiKey = "site-pair-convergence-key";
private static readonly TimeSpan ConvergeTimeout = TimeSpan.FromSeconds(30);
private readonly string _pathA = Path.Combine(Path.GetTempPath(), $"sitepairA-{Guid.NewGuid():N}.db");
private readonly string _pathB = Path.Combine(Path.GetTempPath(), $"sitepairB-{Guid.NewGuid():N}.db");
// The databases are owned by the fixture, in their own providers, and registered into the
// hosts as pre-constructed instances. MS.DI does not dispose instances it did not create,
// so tearing a host down (the offline-peer scenario) leaves the databases intact and
// writable — which is exactly what lets node A accumulate writes while B is down.
private ServiceProvider _dbProviderA = null!;
private ServiceProvider _dbProviderB = null!;
private IHost? _serverHost; // node B — passive
private IHost? _initiatorHost; // node A — dials the peer
static LocalDbSitePairConvergenceTests() =>
// Grpc.Net.Client dials the loopback server over HTTP/2 cleartext.
AppContext.SetSwitch("System.Net.Http.SocketsHttpHandler.Http2UnencryptedSupport", true);
private ILocalDb A => _dbProviderA.GetRequiredService<ILocalDb>();
private ILocalDb B => _dbProviderB.GetRequiredService<ILocalDb>();
public async Task InitializeAsync()
{
_dbProviderA = BuildDatabaseProvider(_pathA, "node-a");
_dbProviderB = BuildDatabaseProvider(_pathB, "node-b");
// Force construction (and therefore OnReady) before anything replicates.
_ = A;
_ = B;
await StartPassiveAsync();
await StartInitiatorAsync();
}
public async Task DisposeAsync()
{
await StopHostAsync(_initiatorHost);
await StopHostAsync(_serverHost);
await _dbProviderA.DisposeAsync();
await _dbProviderB.DisposeAsync();
Microsoft.Data.Sqlite.SqliteConnection.ClearAllPools();
foreach (var path in new[] { _pathA, _pathB })
{
foreach (var suffix in new[] { "", "-wal", "-shm" })
{
try { File.Delete(path + suffix); } catch { /* best effort */ }
}
}
}
// ---- fixture internals ------------------------------------------------------------
/// <summary>
/// A provider owning one consolidated site database, initialized through the host's own
/// <see cref="SiteLocalDbSetup.OnReady"/> — same schema, same registration order.
/// </summary>
private static ServiceProvider BuildDatabaseProvider(string path, string nodeName)
{
var config = new ConfigurationBuilder()
.AddInMemoryCollection(new Dictionary<string, string?>
{
["LocalDb:Path"] = path,
["ScadaBridge:Node:NodeName"] = nodeName,
// Point the legacy migrator at paths that do not exist, so it no-ops rather
// than picking up stray files from the test working directory.
["ScadaBridge:OperationTracking:ConnectionString"] =
$"Data Source={Path.Combine(Path.GetTempPath(), $"absent-{Guid.NewGuid():N}.db")}",
["ScadaBridge:SiteEventLog:DatabasePath"] =
Path.Combine(Path.GetTempPath(), $"absent-{Guid.NewGuid():N}.db"),
})
.Build();
return new ServiceCollection()
.AddZbLocalDb(config, db => SiteLocalDbSetup.OnReady(db, config))
.BuildServiceProvider();
}
private static IConfiguration ReplicationConfig(string? peerAddress)
{
var values = new Dictionary<string, string?>
{
// Tight flush + bounded reconnect backoff so convergence is observable well
// inside the poll deadline. The 60 s production default would let the doubling
// backoff overrun it after a peer outage.
["LocalDb:Replication:FlushInterval"] = "00:00:00.050",
["LocalDb:Replication:ReconnectBackoffMax"] = "00:00:02",
// Both nodes share one key — the interceptor is fail-closed, so a mismatch here
// turns every scenario below red (verified by deliberately breaking it).
["LocalDb:Replication:ApiKey"] = SharedApiKey,
};
if (peerAddress is not null)
values["LocalDb:Replication:PeerAddress"] = peerAddress;
return new ConfigurationBuilder().AddInMemoryCollection(values).Build();
}
private async Task StartPassiveAsync()
{
var config = ReplicationConfig(peerAddress: null);
_serverHost = await new HostBuilder()
.ConfigureWebHost(web =>
{
web.UseKestrel(o =>
o.Listen(IPAddress.Loopback, 0, listen => listen.Protocols = HttpProtocols.Http2));
web.ConfigureServices(services =>
{
services.AddLogging();
services.AddRouting();
// The REAL interceptor, not a stand-in. If it rejected legitimate peer
// traffic, every scenario below would fail — which is the point.
services.AddGrpc(o => o.Interceptors.Add<LocalDbSyncAuthInterceptor>());
services.AddSingleton(B);
services.AddZbLocalDbReplication(config);
});
web.Configure(app =>
{
app.UseRouting();
app.UseEndpoints(e => e.MapZbLocalDbSync());
});
})
.StartAsync();
}
private async Task StartInitiatorAsync()
{
var config = ReplicationConfig(PassiveAddress());
_initiatorHost = await new HostBuilder()
.ConfigureServices(services =>
{
services.AddLogging();
services.AddSingleton(A);
services.AddZbLocalDbReplication(config);
})
.StartAsync();
}
private string PassiveAddress()
=> _serverHost!.Services.GetRequiredService<IServer>()
.Features.Get<IServerAddressesFeature>()!.Addresses.Single();
private static async Task StopHostAsync(IHost? host)
{
if (host is null) return;
try { await host.StopAsync(TimeSpan.FromSeconds(5)); } catch { /* teardown */ }
host.Dispose();
}
// ---- data helpers -----------------------------------------------------------------
private static Task WriteTrackingAsync(ILocalDb db, string id, string status, string target)
=> db.ExecuteAsync(
"""
INSERT INTO OperationTracking (
TrackedOperationId, Kind, TargetSummary, Status, RetryCount,
CreatedAtUtc, UpdatedAtUtc)
VALUES (@id, 'ApiCallCached', @target, @status, 0, @now, @now)
ON CONFLICT(TrackedOperationId) DO UPDATE SET
Status = excluded.Status,
TargetSummary = excluded.TargetSummary,
UpdatedAtUtc = excluded.UpdatedAtUtc;
""",
new { id, status, target, now = DateTime.UtcNow.ToString("o") });
private static Task WriteEventAsync(ILocalDb db, string id, string message)
=> db.ExecuteAsync(
"""
INSERT INTO site_events (id, timestamp, event_type, severity, source, message)
VALUES (@id, @ts, 'script', 'Info', 'convergence-test', @message);
""",
new { id, ts = DateTimeOffset.UtcNow.ToString("o"), message });
private static async Task<string?> ReadTrackingStatusAsync(ILocalDb db, string id)
{
var rows = await db.QueryAsync(
"SELECT Status FROM OperationTracking WHERE TrackedOperationId = @id",
static r => r.GetString(0), new { id });
return rows.Count == 0 ? null : rows[0];
}
private static Task<IReadOnlyList<string>> ReadEventIdsAsync(ILocalDb db)
=> db.QueryAsync("SELECT id FROM site_events ORDER BY id", static r => r.GetString(0));
/// <summary>Polls <paramref name="condition"/> until true or the deadline passes.</summary>
private static async Task WaitUntilAsync(Func<Task<bool>> condition, string because)
{
var deadline = DateTime.UtcNow + ConvergeTimeout;
while (DateTime.UtcNow < deadline)
{
if (await condition()) return;
await Task.Delay(50);
}
Assert.Fail($"Timed out after {ConvergeTimeout.TotalSeconds:0}s waiting for: {because}");
}
// ---- scenarios --------------------------------------------------------------------
[Fact]
public async Task TrackingRow_WrittenOnA_BecomesReadableOnB()
{
await WriteTrackingAsync(A, "op-a-1", "Submitted", "ERP.GetOrder");
await WaitUntilAsync(
async () => await ReadTrackingStatusAsync(B, "op-a-1") == "Submitted",
"the tracking row written on node A to appear on node B");
}
[Fact]
public async Task TrackingRow_WrittenOnB_BecomesReadableOnA()
{
// Replication is bidirectional even though only A dials: proving the passive node's
// writes flow back is what makes a failover in EITHER direction safe.
await WriteTrackingAsync(B, "op-b-1", "Submitted", "ERP.GetOrder");
await WaitUntilAsync(
async () => await ReadTrackingStatusAsync(A, "op-b-1") == "Submitted",
"the tracking row written on node B to appear on node A");
}
[Fact]
public async Task SameOperation_UpdatedOnBothNodes_ConvergesToOneWinner()
{
// Last-writer-wins on the primary key. The specific winner is not asserted — that is
// the HLC's business — but the two nodes MUST agree, and must agree on a value one of
// them actually wrote rather than a merge of both.
await WriteTrackingAsync(A, "op-conflict", "Submitted", "ERP.GetOrder");
await WaitUntilAsync(
async () => await ReadTrackingStatusAsync(B, "op-conflict") is not null,
"the conflicting row to exist on both nodes before it is updated");
await WriteTrackingAsync(A, "op-conflict", "Delivered", "ERP.GetOrder");
await WriteTrackingAsync(B, "op-conflict", "Parked", "ERP.GetOrder");
await WaitUntilAsync(
async () =>
{
var a = await ReadTrackingStatusAsync(A, "op-conflict");
var b = await ReadTrackingStatusAsync(B, "op-conflict");
return a is not null && a == b;
},
"both nodes to converge on one status for the contended operation");
var winner = await ReadTrackingStatusAsync(A, "op-conflict");
Assert.Contains(winner, new[] { "Delivered", "Parked" });
}
[Fact]
public async Task EventsLoggedOnBothNodes_ConvergeToTheUnion_WithNoIdCollisions()
{
// The reason site_events moved to GUID ids. Under the old autoincrement scheme both
// nodes would independently mint id=1, id=2, ... and last-writer-wins on the primary
// key would silently DESTROY one node's events instead of merging them. The union
// count is the assertion that catches that.
var idsFromA = Enumerable.Range(0, 5).Select(_ => Guid.NewGuid().ToString("N")).ToList();
var idsFromB = Enumerable.Range(0, 5).Select(_ => Guid.NewGuid().ToString("N")).ToList();
foreach (var id in idsFromA) await WriteEventAsync(A, id, "from A");
foreach (var id in idsFromB) await WriteEventAsync(B, id, "from B");
var expected = idsFromA.Concat(idsFromB).OrderBy(x => x, StringComparer.Ordinal).ToList();
await WaitUntilAsync(
async () => (await ReadEventIdsAsync(A)).SequenceEqual(expected)
&& (await ReadEventIdsAsync(B)).SequenceEqual(expected),
"both nodes to hold the union of all 10 events");
}
[Fact]
public async Task PeerOffline_ThenRejoins_CatchesUpOnEverythingItMissed()
{
// The failover case that motivates Phase 1: one node is down while the other keeps
// working, and nothing written during the outage may be lost.
await StopHostAsync(_serverHost);
_serverHost = null;
var idsDuringOutage = Enumerable.Range(0, 5).Select(_ => Guid.NewGuid().ToString("N")).ToList();
foreach (var id in idsDuringOutage) await WriteEventAsync(A, id, "written while B was down");
await WriteTrackingAsync(A, "op-during-outage", "Delivered", "ERP.GetOrder");
// Node B's database survived the host teardown (pre-constructed instance), so this is
// a genuine rejoin rather than a fresh node. It comes back on a NEW loopback port;
// the initiator's channel factory re-reads the peer address on each reconnect.
await StartPassiveAsync();
await StopHostAsync(_initiatorHost);
await StartInitiatorAsync();
await WaitUntilAsync(
async () =>
{
var events = await ReadEventIdsAsync(B);
return idsDuringOutage.All(events.Contains)
&& await ReadTrackingStatusAsync(B, "op-during-outage") == "Delivered";
},
"node B to catch up on everything written while it was offline");
}
}