fix(reconcile): heal all concurrently-missing nodes — return existing pending token instead of omitting
This commit is contained in:
+14
@@ -157,6 +157,20 @@ public interface IDeploymentManagerRepository
|
||||
/// <returns>The pending deployment, or null if not found.</returns>
|
||||
Task<PendingDeployment?> GetPendingDeploymentByIdAsync(string deploymentId, CancellationToken cancellationToken = default);
|
||||
/// <summary>
|
||||
/// Gets the pending deployment staged for a given instance (there is at most one per instance
|
||||
/// by design — <see cref="AddPendingDeploymentAsync"/> supersedes and
|
||||
/// <see cref="StagePendingIfAbsentAsync"/> inserts only if absent). Used by startup reconcile:
|
||||
/// when <see cref="StagePendingIfAbsentAsync"/> reports a row already exists (a concurrent
|
||||
/// reconcile from the other node, or an in-flight deploy), the handler reads that existing row
|
||||
/// so a second concurrently-missing node can fetch the same pending config with its (multi-use,
|
||||
/// TTL-bound) token instead of being omitted from the gap. Defensive against >1 row: returns
|
||||
/// the most recently created one.
|
||||
/// </summary>
|
||||
/// <param name="instanceId">The instance whose pending deployment to read.</param>
|
||||
/// <param name="cancellationToken">A cancellation token that can be used to cancel the operation.</param>
|
||||
/// <returns>The pending deployment for the instance, or null if none is staged.</returns>
|
||||
Task<PendingDeployment?> GetPendingDeploymentByInstanceIdAsync(int instanceId, CancellationToken cancellationToken = default);
|
||||
/// <summary>
|
||||
/// Deletes a pending deployment by its deployment ID. No-op if not found. Does NOT
|
||||
/// call <see cref="SaveChangesAsync"/> — the caller commits, mirroring the other
|
||||
/// Add/Delete repository methods.
|
||||
|
||||
@@ -112,11 +112,12 @@ public class ReconcileService
|
||||
// Stage with the snapshot's DeploymentId as the deploymentId so the gap item's
|
||||
// DeploymentId + token point the node at the right pending row to fetch.
|
||||
//
|
||||
// Reconcile staging is safe without a DB uniqueness guard: a gap arises only
|
||||
// from one-node-down-during-a-successful-deploy, so at most one node ever
|
||||
// reconciles a given instance (if BOTH were down the deploy failed and no
|
||||
// snapshot exists, so it is never in the expected set). Deploy-time
|
||||
// supersession serializes via the per-instance operation lock.
|
||||
// StagePendingIfAbsent is insert-if-absent: if BOTH site nodes are concurrently
|
||||
// missing the same instance (e.g. fresh container start / cleared SQLite after a
|
||||
// successful deploy), both attempt to stage here. The first succeeds (true); the
|
||||
// second gets false and is handled in the !staged branch below — it returns the
|
||||
// existing pending row's token so it heals in the same round, rather than being
|
||||
// omitted. Deploy-time supersession serializes via the per-instance operation lock.
|
||||
var staged = await _deploymentRepository.StagePendingIfAbsentAsync(
|
||||
exp.InstanceId, snapshot.DeploymentId, exp.RevisionHash,
|
||||
snapshot.ConfigurationJson, token, now, expiresAt, cancellationToken)
|
||||
@@ -124,12 +125,35 @@ public class ReconcileService
|
||||
|
||||
if (!staged)
|
||||
{
|
||||
// A pending row already exists — an in-flight deploy is mid-flight and its
|
||||
// replication will deliver this instance to the node shortly. Omit it from
|
||||
// the gap (reconcile is best-effort and re-runs).
|
||||
_logger.LogDebug(
|
||||
"Reconcile: pending row already exists for instance {Instance} (in-flight deploy); omitting from gap",
|
||||
exp.InstanceUniqueName);
|
||||
// A pending row already exists for this instance — either a CONCURRENT reconcile
|
||||
// from the other site node (both nodes' SQLite empty after a fresh/cleared deploy,
|
||||
// both reconciling at startup) or an in-flight deploy. Do NOT omit the item: if we
|
||||
// did, the second concurrently-missing node would get 0 fetched and stay unhealed
|
||||
// until a later restart. Instead, read the EXISTING pending row and emit a gap item
|
||||
// carrying ITS DeploymentId/RevisionHash/Token. The fetch token is multi-use within
|
||||
// its TTL, so both nodes fetch the same pending config and heal in the same round.
|
||||
// (If the existing row is from an in-flight deploy its config is newer than the
|
||||
// snapshot — fetching it is still correct; the site's guarded write handles ordering.)
|
||||
var existing = await _deploymentRepository
|
||||
.GetPendingDeploymentByInstanceIdAsync(exp.InstanceId, cancellationToken)
|
||||
.ConfigureAwait(false);
|
||||
if (existing != null)
|
||||
{
|
||||
_logger.LogDebug(
|
||||
"Reconcile: pending row already exists for instance {Instance} (concurrent reconcile or in-flight deploy); returning existing token so this node heals too",
|
||||
exp.InstanceUniqueName);
|
||||
gap.Add(new ReconcileGapItem(
|
||||
exp.InstanceUniqueName, existing.DeploymentId, existing.RevisionHash,
|
||||
exp.IsEnabled, existing.Token));
|
||||
}
|
||||
else
|
||||
{
|
||||
// Raced away: the pending row was purged between the stage attempt and this
|
||||
// read. Omit it — reconcile is best-effort and the node retries next round.
|
||||
_logger.LogDebug(
|
||||
"Reconcile: pending row for instance {Instance} disappeared between stage and read (purged race); omitting from gap",
|
||||
exp.InstanceUniqueName);
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
|
||||
+13
@@ -226,6 +226,19 @@ public class DeploymentManagerRepository : IDeploymentManagerRepository
|
||||
.FirstOrDefaultAsync(p => p.DeploymentId == deploymentId, cancellationToken);
|
||||
}
|
||||
|
||||
/// <inheritdoc />
|
||||
public Task<PendingDeployment?> GetPendingDeploymentByInstanceIdAsync(int instanceId, CancellationToken cancellationToken = default)
|
||||
{
|
||||
// At most one pending row per instance by design (supersession + stage-if-absent),
|
||||
// but order deterministically and take the most recent so a hypothetical duplicate
|
||||
// never makes the read non-deterministic — mirrors GetCurrentDeploymentStatusAsync.
|
||||
return _dbContext.Set<PendingDeployment>()
|
||||
.Where(p => p.InstanceId == instanceId)
|
||||
.OrderByDescending(p => p.CreatedAtUtc)
|
||||
.ThenByDescending(p => p.Id)
|
||||
.FirstOrDefaultAsync(cancellationToken);
|
||||
}
|
||||
|
||||
/// <inheritdoc />
|
||||
public async Task DeletePendingDeploymentByIdAsync(string deploymentId, CancellationToken cancellationToken = default)
|
||||
{
|
||||
|
||||
@@ -117,8 +117,12 @@ public class ReconcileServiceTests
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Reconcile_StagePendingReturnsFalse_OmitsThatGapItem()
|
||||
public async Task Reconcile_StagePendingReturnsFalse_ExistingRowPresent_ReturnsExistingTokenInGap()
|
||||
{
|
||||
// Live-found bug: two site nodes concurrently missing the SAME instance both reconcile at
|
||||
// startup. The first stages a pending row; the second's StagePendingIfAbsent returns false.
|
||||
// The second node must STILL heal — the handler reads the existing pending row and returns
|
||||
// ITS deploymentId/revHash/token (the fetch token is multi-use within its TTL).
|
||||
SiteResolves();
|
||||
var b = Expected(2, "inst-B", "rev2", "dep-B");
|
||||
var c = Expected(3, "inst-C", "rev3", "dep-C");
|
||||
@@ -126,7 +130,8 @@ public class ReconcileServiceTests
|
||||
SnapshotFor(b);
|
||||
SnapshotFor(c);
|
||||
|
||||
// Both missing locally, but C already has an in-flight pending row.
|
||||
// B stages fresh (true). C already has a pending row (a concurrent reconcile from the
|
||||
// other node, or an in-flight deploy, staged it first) → stage returns false.
|
||||
StageReturns(true);
|
||||
_deploymentRepo.StagePendingIfAbsentAsync(
|
||||
3, Arg.Any<string>(), Arg.Any<string>(), Arg.Any<string>(),
|
||||
@@ -134,6 +139,52 @@ public class ReconcileServiceTests
|
||||
Arg.Any<CancellationToken>())
|
||||
.Returns(false);
|
||||
|
||||
// The existing pending row for inst-C carries the FIRST node's deploymentId + token.
|
||||
var nowUtc = DateTimeOffset.UtcNow;
|
||||
_deploymentRepo.GetPendingDeploymentByInstanceIdAsync(3, Arg.Any<CancellationToken>())
|
||||
.Returns(new PendingDeployment(
|
||||
"dep-C-existing", 3, "rev3-existing", "{\"cfg\":\"inst-C\"}",
|
||||
"tok-C-existing", nowUtc, nowUtc.AddMinutes(5)));
|
||||
|
||||
var response = await CreateService().ReconcileAsync(Request(/* empty local inventory */));
|
||||
|
||||
Assert.Equal(2, response.Gap.Count);
|
||||
|
||||
var gapB = Assert.Single(response.Gap, g => g.InstanceUniqueName == "inst-B");
|
||||
Assert.Equal("dep-B", gapB.DeploymentId);
|
||||
|
||||
// inst-C IS included, carrying the EXISTING row's deploymentId/revHash/token — NOT the
|
||||
// snapshot's (snapshot.DeploymentId would be "dep-C").
|
||||
var gapC = Assert.Single(response.Gap, g => g.InstanceUniqueName == "inst-C");
|
||||
Assert.Equal("dep-C-existing", gapC.DeploymentId);
|
||||
Assert.Equal("rev3-existing", gapC.RevisionHash);
|
||||
Assert.Equal("tok-C-existing", gapC.FetchToken);
|
||||
Assert.True(gapC.IsEnabled);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Reconcile_StagePendingReturnsFalse_NoExistingRow_OmitsThatGapItem()
|
||||
{
|
||||
// Fallback path: stage returns false, but the pending row raced away (was purged between
|
||||
// the stage attempt and the read). Omit the item — the node retries on the next reconcile.
|
||||
SiteResolves();
|
||||
var b = Expected(2, "inst-B", "rev2", "dep-B");
|
||||
var c = Expected(3, "inst-C", "rev3", "dep-C");
|
||||
ExpectedSet(b, c);
|
||||
SnapshotFor(b);
|
||||
SnapshotFor(c);
|
||||
|
||||
StageReturns(true);
|
||||
_deploymentRepo.StagePendingIfAbsentAsync(
|
||||
3, Arg.Any<string>(), Arg.Any<string>(), Arg.Any<string>(),
|
||||
Arg.Any<string>(), Arg.Any<DateTimeOffset>(), Arg.Any<DateTimeOffset>(),
|
||||
Arg.Any<CancellationToken>())
|
||||
.Returns(false);
|
||||
|
||||
// The pending row was purged between the stage attempt and the read.
|
||||
_deploymentRepo.GetPendingDeploymentByInstanceIdAsync(3, Arg.Any<CancellationToken>())
|
||||
.Returns((PendingDeployment?)null);
|
||||
|
||||
var response = await CreateService().ReconcileAsync(Request(/* empty local inventory */));
|
||||
|
||||
var gap = Assert.Single(response.Gap);
|
||||
|
||||
+30
@@ -159,6 +159,36 @@ public class PendingDeploymentRepositoryTests : IDisposable
|
||||
Assert.Null(await _repository.GetPendingDeploymentByIdAsync("does-not-exist"));
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task GetPendingDeploymentByInstanceId_ReturnsRow_WhenPresent()
|
||||
{
|
||||
// Startup-reconcile read path: when StagePendingIfAbsent reports a row already exists
|
||||
// (concurrent reconcile from the other node, or an in-flight deploy), the handler reads
|
||||
// the existing row by instance id to hand the second node the same fetch token.
|
||||
var instanceId = await SeedInstanceAsync("Inst9");
|
||||
await _repository.AddPendingDeploymentAsync(NewPending("dep9", instanceId, "{\"v\":9}"));
|
||||
await _repository.SaveChangesAsync();
|
||||
|
||||
var row = await _repository.GetPendingDeploymentByInstanceIdAsync(instanceId);
|
||||
|
||||
Assert.NotNull(row);
|
||||
Assert.Equal("dep9", row!.DeploymentId);
|
||||
Assert.Equal(instanceId, row.InstanceId);
|
||||
Assert.Equal("tok-dep9", row.Token);
|
||||
Assert.Equal("{\"v\":9}", row.ConfigurationJson);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task GetPendingDeploymentByInstanceId_ReturnsNull_WhenAbsent()
|
||||
{
|
||||
// No pending row staged for this instance → null (the reconcile fallback/omit path).
|
||||
var instanceId = await SeedInstanceAsync("Inst10");
|
||||
|
||||
var row = await _repository.GetPendingDeploymentByInstanceIdAsync(instanceId);
|
||||
|
||||
Assert.Null(row);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task AddPendingDeployment_MultiplePriorRowsSameInstance_AllSuperseded()
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user