LocalDb adoption Phase 1 + 2: consolidate the site database, delete the bespoke replicators #23

Merged
dohertj2 merged 55 commits from feat/localdb-phase2 into main 2026-07-20 06:06:08 -04:00
Showing only changes of commit c56bf4ae65 - Show all commits
@@ -0,0 +1,177 @@
using ZB.MOM.WW.LocalDb;
namespace ZB.MOM.WW.ScadaBridge.IntegrationTests;
/// <summary>
/// Phase 2 convergence: deployed configuration across a real site pair, plus the resync
/// behaviour that the bespoke replicator needed a directional guard for.
/// </summary>
/// <remarks>
/// Companion to <see cref="LocalDbStoreAndForwardConvergenceTests"/>: same
/// specifications-before-deletion role, for the intents held by
/// <c>SiteReplicationActorTests</c> and <c>SfBufferResyncPredicateTests</c> rather than by
/// the store-and-forward <c>ReplicationService</c>.
/// </remarks>
[Collection("LocalDbSitePairConvergence")]
public sealed class LocalDbConfigConvergenceTests : LocalDbSitePairHarness
{
// ---- data helpers -----------------------------------------------------------------
private static Task DeployConfigAsync(ILocalDb db, string instance, string configJson, string deploymentId)
=> db.ExecuteAsync(
"""
INSERT INTO deployed_configurations (
instance_unique_name, config_json, deployment_id, revision_hash,
is_enabled, deployed_at)
VALUES (@instance, @configJson, @deploymentId, @deploymentId, 1, @now)
ON CONFLICT(instance_unique_name) DO UPDATE SET
config_json = excluded.config_json,
deployment_id = excluded.deployment_id,
revision_hash = excluded.revision_hash,
deployed_at = excluded.deployed_at;
""",
new { instance, configJson, deploymentId, now = DateTime.UtcNow.ToString("o") });
private static async Task<string?> ReadConfigAsync(ILocalDb db, string instance)
{
var rows = await db.QueryAsync(
"SELECT config_json FROM deployed_configurations WHERE instance_unique_name = @instance",
static r => r.GetString(0), new { instance });
return rows.Count == 0 ? null : rows[0];
}
private static Task EnqueueAsync(ILocalDb db, string id)
=> db.ExecuteAsync(
"""
INSERT INTO sf_messages (
id, category, target, payload_json, retry_count, max_retries,
retry_interval_ms, created_at, status)
VALUES (@id, 0, 'central', '{}', 0, 50, 30000, @now, 0);
""",
new { id, now = DateTime.UtcNow.ToString("o") });
private static async Task<bool> MessageExistsAsync(ILocalDb db, string id)
{
var rows = await db.QueryAsync(
"SELECT 1 FROM sf_messages WHERE id = @id", static r => r.GetInt32(0), new { id });
return rows.Count > 0;
}
// ---- scenarios --------------------------------------------------------------------
[Fact]
public async Task ANodeWithNewerLocalRows_KeepsThem_WhenAPeerSnapshotArrives()
{
// N1 Critical (was: SfBufferResyncPredicateTests). That test existed because the
// bespoke replicator's ReplaceAllAsync was a destructive DELETE-then-INSERT: a resync
// running in the wrong direction WIPED a live buffer, so the code needed a
// same-oldest-Up predicate to decide who had authority. The divergence case was a
// rolling restart leaving the delivering node oldest but not leader.
//
// Under LocalDb that failure mode is structurally impossible rather than guarded
// against: snapshot resync merges per row under LWW and never deletes. So this is
// deliberately NOT a directional-authority test — there is no active/standby
// asymmetry left to enforce. It asserts the property the guard was protecting: no
// node loses rows to a peer's snapshot.
//
// The setup is the wipe scenario made concrete. Each node holds a row the other has
// never seen, and they disagree about a third; then they resync.
await DeployConfigAsync(A, "shared-instance", """{"v":1}""", "dep-1");
await WaitUntilAsync(
async () => await ReadConfigAsync(B, "shared-instance") is not null,
"the shared row to exist on both nodes before they diverge");
await StopPassiveAsync();
// A keeps working while B is down: a brand-new buffered message and a newer version
// of the shared config.
await EnqueueAsync(A, "live-row-on-a");
await DeployConfigAsync(A, "shared-instance", """{"v":2}""", "dep-2");
// B is offline but its database is still writable — it also has local state the
// snapshot exchange must not destroy.
await EnqueueAsync(B, "live-row-on-b");
await RestartPairAsync();
await WaitUntilAsync(
async () =>
await MessageExistsAsync(A, "live-row-on-a") &&
await MessageExistsAsync(A, "live-row-on-b") &&
await MessageExistsAsync(B, "live-row-on-a") &&
await MessageExistsAsync(B, "live-row-on-b"),
"both nodes to hold the UNION of the rows each wrote while partitioned");
// And the contended row converges to the newer write rather than either node's
// snapshot flattening the other.
await WaitUntilAsync(
async () => await ReadConfigAsync(A, "shared-instance") == """{"v":2}"""
&& await ReadConfigAsync(B, "shared-instance") == """{"v":2}""",
"both nodes to converge on the newer config for the contended instance");
}
[Fact]
public async Task ConfigDeployedOnA_ConvergesToB_WithoutAnyFetch()
{
// Was: SiteReplicationActorTests #1-#5 (notify-and-fetch — the standby is told a
// deploy happened, then HTTP-fetches the config itself, with retries and a
// superseded-404 path). Under CDC the config row simply replicates, so that whole
// exchange is deleted at Task 15.
//
// The plan called for asserting a fetcher test double records zero invocations. That
// would be THEATRE here and is deliberately not done: this harness has no actor
// system, no central, and no IDeploymentConfigFetcher in the graph at all, so a
// double would record zero calls whether or not the fetch path still existed. An
// assertion that cannot fail is worse than none.
//
// What this test honestly proves is the positive half: the config reaches node B
// through replication ALONE, in a process where fetching is not merely unused but
// absent. The negative half — that no fetch path survives — is proved by Task 15
// deleting the code and the build still passing, which is a stronger check than any
// runtime counter.
//
// Scope note (D1): SiteReconciliationActor survives Phase 2 and legitimately fetches
// over HTTP at node startup when central reports gaps. "The standby never fetches,
// ever" would be a FALSE claim; the claim is about the deploy path only.
await DeployConfigAsync(A, "inst-deploy", """{"setpoint":42}""", "dep-100");
await WaitUntilAsync(
async () => await ReadConfigAsync(B, "inst-deploy") == """{"setpoint":42}""",
"the deployed config to reach node B over replication alone");
}
[Fact]
public async Task RedeployingAnInstance_ConvergesToTheNewRevision_OnBothNodes()
{
// The steady-state deploy loop, and the reason the old path needed a deployed_at
// guard: a standby that applied a STALE fetch would pin an instance to superseded
// config until the next deploy. Under LWW the newer write wins on the primary key,
// so the guard is not what protects this — but the outcome still has to hold.
await DeployConfigAsync(A, "inst-redeploy", """{"rev":"a"}""", "dep-a");
await WaitUntilAsync(
async () => await ReadConfigAsync(B, "inst-redeploy") == """{"rev":"a"}""",
"the first revision to reach node B");
await DeployConfigAsync(A, "inst-redeploy", """{"rev":"b"}""", "dep-b");
await WaitUntilAsync(
async () => await ReadConfigAsync(A, "inst-redeploy") == """{"rev":"b"}"""
&& await ReadConfigAsync(B, "inst-redeploy") == """{"rev":"b"}""",
"both nodes to converge on the redeployed revision");
}
[Fact]
public async Task ConfigDeployedOnB_ConvergesToA()
{
// Deliberately the reverse direction. The bespoke replicator was asymmetric by
// design — the active node pushed, the standby fetched — so "which node deployed it"
// was a meaningful question. Under CDC it is not, and this test is what says so.
// After a failover the surviving node must be able to deploy without waiting to be
// promoted to some authority role.
await DeployConfigAsync(B, "inst-from-b", """{"origin":"b"}""", "dep-b1");
await WaitUntilAsync(
async () => await ReadConfigAsync(A, "inst-from-b") == """{"origin":"b"}""",
"a config deployed on node B to reach node A");
}
}