Files
DodoSSH/tests/DodoSSH.Client.Sync.Tests/ConflictMatrixTests.cs
jaap-jan e3fd3e1728 Sync and authenticate with SSH keys on the client
Completes the client half of SSH keys: they sync alongside hosts, appear in
their own list, and can be selected to authenticate a connection instead of
typing a password.

The reconciler and the repository were Host-typed throughout, so the choice was
to generalise them or to keep a second copy per item type. Generalised, because
ItemReconciler's whole premise is that the pull and the push paths must answer
the same collision the same way — two copies would drift the first time one of
them was fixed. What is genuinely per-type now arrives through
IItemKind<TSecret>: the cipher, the merge, the plaintext columns, and the noun
to use when telling a person what happened to their item. Generic where the
server's IItemKind is not, and for the reason that reverses there — the client
needs the concrete type, because it merges field by field.

The pull filter is derived from the same registry that builds the reconcilers.
That is the specific failure being designed out: an item type that encrypts,
merges and lists perfectly and is never once requested from the server, so it
works on the machine that made it and exists nowhere else.

No client cache migration. The item table's primary key and the outbox's unique
index already carry the entity type, and AadResourceTypes already mapped SshKey
— so a host and a key may share an id and never see each other's rows, which
SshKeySyncTests now arranges deliberately.

A key hands the server nothing in plaintext. There is a public_key_fingerprint
column and it would be accepted; leaving it null is deliberate. A fingerprint is
not secret but it is a stable identifier for a key pair, so filling it would let
an operator tell which of their users hold the same key and correlate one across
vaults, for a column nothing reads. The design allows itself one plaintext
concession — the relay address, which the relay cannot work without — and this
is not that.

A key is chosen per connection rather than bound to a host, which works the way
ssh -i does. Binding one needs a field on HostSecret and therefore a payload
schema bump, which makes every host written afterwards read-only on an older
build; worth doing deliberately rather than as a side effect of adding keys.

Three things this found, all of them by being falsified rather than by review:

- Making the reconciler generic silently turned a record comparison into
  reference equality, because == on a type parameter is not value equality. The
  effect would have been a conflict recorded on every pass for an unacknowledged
  create that had in fact landed. Sabotaging the fix left all 73 tests passing —
  nothing covered that branch — so ConflictMatrixTests now has
  AnUnacknowledgedCreateThatDidLand_IsDroppedQuietly, which fails without it.

- A test asserting that a blank passphrase reaches SSH.NET as null was vacuous:
  it exercised the editor, not the credential path, and passed with the guard
  deleted. Resolved by making SshKeySecret.Passphrase normalise an empty string
  to null, so there is one spelling of one state — which also keeps two clients
  from producing different payload bytes for an identical key. That exposed a
  wider gap: SshKeySecret, its codec and its merge had no direct unit tests at
  all. They have 25 now.

- The reason first given for that normalisation was false. It claimed SSH.NET
  rejects a passphrase supplied for an unprotected key; measured against a real
  sshd it ignores it and authenticates anyway. Corrected everywhere it was
  stated and recorded in docs/platform-flags.md. The same test file also closes
  a real hole: SshPrivateKeyCredential had never been exercised against a
  server, because the existing key test builds SSH.NET's auth method directly
  and bypasses the path a vault-held key actually takes.

Only one editor may be open at a time. Both sit in the same 340-pixel column as
Auto rows and their heights together exceed it at the window's minimum size, so
two open editors put the lower one's Save and Cancel past the bottom edge — the
same failure this window already shipped once with the setup screens. Expressed
as a state rule because that is the only form of it this repository can check:
nothing here loads a .axaml. The refusal keeps what was typed, since in the key
editor that is a pasted private key the user may have nowhere else.

The end-to-end slice now carries a key as well as a host, so both item types go
through the real API, the real PostgreSQL and the real crypto in one pass — the
three hand-kept mappings between enums that do not line up are the reason that
is worth doing rather than trusting the unit suites.

735 tests green, including the container-backed SSH and end-to-end suites. Zero
warnings, dotnet format clean.
2026-07-29 20:27:23 +02:00

555 lines
22 KiB
C#

using DodoSSH.Client.Storage;
using static DodoSSH.Client.Sync.Tests.SyncHarness;
namespace DodoSSH.Client.Sync.Tests;
/// <summary>
/// Two machines, one vault, every way they can disagree.
/// </summary>
/// <remarks>
/// The highest-value suite in the product, because this is the only place data can be lost. Every case
/// asserts two things: that the two devices converge on the same state, and that whatever the merge had
/// to override is recorded rather than gone. A merge that quietly drops the password someone just typed
/// is worse than one that refuses to merge at all.
/// </remarks>
public sealed class ConflictMatrixTests : IAsyncLifetime
{
private SyncHarness harness = null!;
/// <inheritdoc />
public async ValueTask InitializeAsync() => harness = await CreateAsync();
/// <inheritdoc />
public ValueTask DisposeAsync()
{
harness.Dispose();
return ValueTask.CompletedTask;
}
// ---- The uncontested paths ----
[Fact]
public async Task ACreatedHost_ReachesTheOtherMachine()
{
var entityId = await harness.First.CreateAsync(Host("prod-db", notes: "primary"));
await harness.SettleAsync();
var seen = await harness.Second.FindAsync(entityId);
seen.Secret.Label.ShouldBe("prod-db");
seen.Secret.Notes.ShouldBe("primary");
seen.HasUnsyncedChanges.ShouldBeFalse();
harness.Server.RowCount.ShouldBe(1);
}
[Fact]
public async Task AHostCreatedOffline_IsVisibleLocallyBeforeAnySync()
{
// The reason the outbox exists. A host typed in on a plane has to be usable on that plane.
var entityId = await harness.First.CreateAsync(Host("prod-db"));
var local = await harness.First.FindAsync(entityId);
local.Secret.Label.ShouldBe("prod-db");
local.HasUnsyncedChanges.ShouldBeTrue();
harness.Server.RowCount.ShouldBe(0);
}
[Fact]
public async Task ALocalOnlyEdit_IsPushed()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.UpdateAsync(entityId, Host("prod-db", notes: "rotate quarterly"));
await harness.SettleAsync();
(await harness.Second.FindAsync(entityId)).Secret.Notes.ShouldBe("rotate quarterly");
(await harness.First.ConflictsAsync()).ShouldBeEmpty();
}
[Fact]
public async Task ARemoteOnlyEdit_IsPulledWithoutAConflict()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.Second.UpdateAsync(entityId, Host("prod-db", username: "postgres"));
await harness.SettleAsync();
(await harness.First.FindAsync(entityId)).Secret.Username.ShouldBe("postgres");
(await harness.First.ConflictsAsync()).ShouldBeEmpty();
}
// ---- Both edited ----
[Fact]
public async Task BothEditedDifferentFields_BothSurvive()
{
// The payoff for a field-level merge. Last-writer-wins would lose one of these.
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.UpdateAsync(entityId, Host("prod-db", notes: "from the laptop"));
await harness.Second.UpdateAsync(entityId, Host("prod-db", username: "postgres"));
await harness.SettleAsync();
var merged = (await harness.First.FindAsync(entityId)).Secret;
merged.Notes.ShouldBe("from the laptop");
merged.Username.ShouldBe("postgres");
(await harness.Second.FindAsync(entityId)).Secret.ShouldBe(merged);
(await harness.First.ConflictsAsync()).ShouldBeEmpty();
}
[Fact]
public async Task BothAddedADifferentDirective_BothSurvive()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.UpdateAsync(
entityId, Host("prod-db", options: [("Compression", "yes")]));
await harness.Second.UpdateAsync(
entityId, Host("prod-db", options: [("ServerAliveInterval", "30")]));
await harness.SettleAsync();
var merged = (await harness.First.FindAsync(entityId)).Secret;
merged.Options.Count.ShouldBe(2);
merged.Options.TryGetValue("Compression", out _).ShouldBeTrue();
merged.Options.TryGetValue("ServerAliveInterval", out _).ShouldBeTrue();
}
[Fact]
public async Task BothEditedTheSameField_OneValueWinsAndTheOtherIsRecorded()
{
// A genuine clash. Whichever side loses, its value has to be retrievable — that is the entire
// justification for resolving automatically instead of blocking.
var entityId = await harness.First.CreateAsync(Host("prod-db", notes: "original"));
await harness.SettleAsync();
await harness.First.UpdateAsync(entityId, Host("prod-db", notes: "from the laptop"));
await harness.Second.UpdateAsync(entityId, Host("prod-db", notes: "from the desktop"));
await harness.SettleAsync();
var first = (await harness.First.FindAsync(entityId)).Secret;
var second = (await harness.Second.FindAsync(entityId)).Secret;
first.ShouldBe(second);
var winner = first.Notes.ShouldNotBeNull();
var lost = string.Equals(winner, "from the laptop", StringComparison.Ordinal)
? "from the desktop"
: "from the laptop";
// One of the two, and the same one on both machines. Which is not the point; that the other is
// retrievable is.
new[] { "from the laptop", "from the desktop" }
.Contains(winner, StringComparer.Ordinal)
.ShouldBeTrue();
(await ConflictsAcrossDevicesAsync())
.ShouldContain(kind => kind == ConflictKind.FieldOverridden);
(await DiscardedValuesAsync())
.ShouldContain(
detail => detail.Contains(lost, StringComparison.Ordinal),
"the overridden value must be recoverable from the conflict log");
}
[Fact]
public async Task BothMadeTheSameEdit_IsNotAConflict()
{
// Two people fixing the same typo must not be asked to arbitrate.
var entityId = await harness.First.CreateAsync(Host("prod-db", hostname: "db.internl"));
await harness.SettleAsync();
await harness.First.UpdateAsync(entityId, Host("prod-db", hostname: "db.internal"));
await harness.Second.UpdateAsync(entityId, Host("prod-db", hostname: "db.internal"));
await harness.SettleAsync();
(await harness.First.FindAsync(entityId)).Secret.Hostname.ShouldBe("db.internal");
(await ConflictsAcrossDevicesAsync()).ShouldBeEmpty();
}
// ---- Deletes ----
[Fact]
public async Task ADeletedHost_DisappearsEverywhere()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.DeleteAsync(entityId);
await harness.SettleAsync();
(await harness.First.ListAsync()).Items.ShouldBeEmpty();
(await harness.Second.ListAsync()).Items.ShouldBeEmpty();
harness.Server.RowCount.ShouldBe(0);
}
[Fact]
public async Task BothDeleted_IsNotAConflict()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.DeleteAsync(entityId);
await harness.Second.DeleteAsync(entityId);
await harness.SettleAsync();
(await harness.First.ListAsync()).Items.ShouldBeEmpty();
(await harness.Second.ListAsync()).Items.ShouldBeEmpty();
(await ConflictsAcrossDevicesAsync()).ShouldBeEmpty();
}
[Fact]
public async Task DeletedElsewhereWhileEditedHere_TheLocalWorkSurvivesUnderANewName()
{
// The case where naive handling loses data outright. The tombstone has to stand — arguing with it
// conflicts for ever — so the edit is preserved as a separate host instead of being dropped.
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
// The laptop syncs first, so its delete is what reaches the server; the desktop's edit is the
// one that has to be rescued. Which side loses is decided by who gets there first, and both
// orderings are covered — see EditedElsewhereWhileDeletedHere for the mirror image.
await harness.First.DeleteAsync(entityId);
await harness.Second.UpdateAsync(
entityId, Host("prod-db", notes: "credentials rotated, do not delete"));
await harness.SettleAsync();
var listing = await harness.First.ListAsync();
var restored = listing.Items.ShouldHaveSingleItem();
restored.EntityId.ShouldNotBe(entityId);
restored.Secret.Label.ShouldBe("prod-db (restored)");
restored.Secret.Notes.ShouldBe("credentials rotated, do not delete");
(await ConflictsAcrossDevicesAsync())
.ShouldContain(kind => kind == ConflictKind.RemoteDeleteResurrected);
// And the other machine sees it too, so the rescue is not local-only.
(await harness.Second.ListAsync()).Items.ShouldHaveSingleItem()
.EntityId.ShouldBe(restored.EntityId);
}
[Fact]
public async Task ReplayingAPulledDeletion_DoesNotDuplicateTheRescuedCopy()
{
// Applying a pulled change is at-least-once — the cursor is saved after the page is applied — so
// a whole page can arrive twice. This checks the replay is harmless end to end; the deterministic
// id it relies on is pinned by ResurrectionIdTests.
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.DeleteAsync(entityId);
await harness.Second.UpdateAsync(entityId, Host("prod-db", notes: "keep me"));
await harness.First.SyncAsync();
// Rewind the desktop's cursor, so the deletion arrives a second time and it tries to rescue the
// same content twice.
await harness.Second.SyncAsync();
await harness.Second.SyncState.ResetAsync(VaultId, TestContext.Current.CancellationToken);
await harness.Second.SyncAsync();
await harness.SettleAsync();
(await harness.First.ListAsync()).Items.Count.ShouldBe(1);
harness.Server.RowCount.ShouldBe(1);
}
[Fact]
public async Task EditedElsewhereWhileDeletedHere_TheDeleteIsAbandonedAndReported()
{
// The mirror image, and resolved the same way round: an edit outlives a removal. Re-deleting
// costs a click; a discarded edit may be the only copy of something.
var entityId = await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
await harness.First.UpdateAsync(entityId, Host("prod-db", notes: "still needed"));
await harness.Second.DeleteAsync(entityId);
await harness.SettleAsync();
var survivor = await harness.First.FindAsync(entityId);
survivor.Secret.Notes.ShouldBe("still needed");
(await ConflictsAcrossDevicesAsync())
.ShouldContain(kind => kind == ConflictKind.LocalDeleteOverridden);
}
// ---- Ordering, retries and idempotence ----
[Fact]
public async Task OfflineChanges_ArePushedInTheOrderTheyWereMade()
{
var first = await harness.First.CreateAsync(Host("a-bastion"));
var second = await harness.First.CreateAsync(Host("b-database"));
var third = await harness.First.CreateAsync(Host("c-cache"));
await harness.SettleAsync();
var order = (await harness.Second.ListAsync()).Items
.OrderBy(host => host.Version)
.ThenBy(host => host.Secret.Label, StringComparer.Ordinal)
.Select(host => host.EntityId)
.ToArray();
order.ShouldBe([first, second, third], ignoreOrder: false);
}
[Fact]
public async Task ASecondSyncPass_ChangesNothing()
{
// Idempotence, which is what makes re-reading a client's own writes a safe way to avoid the
// cursor-gap hazard in the push response.
await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
var before = await harness.First.HostsSortedAsync();
var pushesBefore = harness.Server.PushCount;
var report = await harness.First.SyncAsync();
report.Pushed.ShouldBe(0);
harness.Server.PushCount.ShouldBe(pushesBefore);
(await harness.First.HostsSortedAsync()).ShouldBe(before);
}
[Fact]
public async Task AnEditWhileAnEarlierPushIsUnacknowledged_KeepsTheNewerValueWithoutDuplicating()
{
// A create that reached the server and whose answer did not come back, followed by another edit.
// The newer value has to win and there must be exactly one host afterwards. Two mechanisms keep
// that true: the pull sees the server's row and re-bases the queued edit onto it, and a coalesced
// row carries a fresh operation id so the server cannot answer Duplicate — "already applied" —
// for an operation whose contents have since changed. The second is pinned directly by
// OutboxStoreTests.ACoalescedEdit_GetsAFreshOperationId; here they are exercised together.
var entityId = await harness.First.CreateAsync(Host("prod-db", notes: "first"));
// The create lands on the server, but the acknowledgement never reaches the laptop.
await PushBehindTheEnginesBackAsync(entityId);
await harness.First.UpdateAsync(entityId, Host("prod-db", notes: "second"));
await harness.SettleAsync();
harness.Server.RowCount.ShouldBe(1);
(await harness.First.FindAsync(entityId)).Secret.Notes.ShouldBe("second");
(await harness.Second.FindAsync(entityId)).Secret.Notes.ShouldBe("second");
}
[Fact]
public async Task AnUnacknowledgedCreateThatDidLand_IsDroppedQuietly()
{
// The same lost acknowledgement as above, but with no subsequent edit — which is the common case,
// since a timeout is far more likely than a timeout followed by a change. The queued create meets
// the server's own copy of itself, and the only correct answer is to stop trying to send it. In
// particular this must not be reported as a conflict: there is nothing for a person to decide, and a
// vault that produced a conflict notice every time a push timed out would train people to ignore
// them.
//
// This is the one path that compares two decrypted items for equality, and the comparison has to go
// through EqualityComparer rather than ==, because the reconciler is generic over the secret type
// and == on a type parameter is reference equality.
var entityId = await harness.First.CreateAsync(Host("prod-db", notes: "first"));
await PushBehindTheEnginesBackAsync(entityId);
await harness.SettleAsync();
(await ConflictsAcrossDevicesAsync()).ShouldBeEmpty(
"an item that came back exactly as it was sent is not something to arbitrate");
// And the operation is gone rather than still being offered.
(await harness.First.Outbox.TakeAsync(VaultId, 100, TestContext.Current.CancellationToken))
.ShouldBeEmpty();
harness.Server.RowCount.ShouldBe(1);
(await harness.First.FindAsync(entityId)).HasUnsyncedChanges.ShouldBeFalse();
}
[Fact]
public async Task AConcurrentWriteDuringAPush_IsNotSkipped()
{
// The cursor-gap hazard. The push response carries a cursor sitting after this client's own
// changes; adopting it would skip anything another client committed at a lower sequence in the
// window between this client's pull and its push. The engine keeps its own cursor instead.
var mine = await harness.First.CreateAsync(Host("mine"));
var theirs = Guid.CreateVersion7();
harness.Server.OnPush = () => harness.Server.ExternalUpsert(theirs, ForeignPayload(), null);
await harness.First.SyncAsync();
var ids = (await harness.First.Items
.ListAsync(VaultId, Contracts.SyncEntityType.Host, false, TestContext.Current.CancellationToken))
.Select(item => item.EntityId)
.ToArray();
ids.ShouldContain(mine);
ids.ShouldContain(theirs, "a change committed during the push was skipped");
}
[Fact]
public async Task AForbiddenWrite_IsParkedRatherThanRetriedForever()
{
var entityId = await harness.First.CreateAsync(Host("prod-db"));
harness.Server.DenyWrites = true;
var report = await harness.First.SyncAsync();
report.Parked.ShouldBe(1);
(await harness.First.Outbox.ListParkedAsync(VaultId, TestContext.Current.CancellationToken))
.ShouldHaveSingleItem().EntityId.ShouldBe(entityId);
// A parked operation is not retried, so a second pass sends nothing.
var pushes = harness.Server.PushCount;
await harness.First.SyncAsync();
harness.Server.PushCount.ShouldBe(pushes);
(await harness.First.ConflictsAsync()).ShouldContain(c => c.Kind == ConflictKind.Rejected);
}
[Fact]
public async Task AParkedChange_IsStillWhatTheUserSees()
{
// Hiding it because the server refused would show the old values and look like the edit was lost.
var entityId = await harness.First.CreateAsync(Host("prod-db", notes: "mine"));
harness.Server.DenyWrites = true;
await harness.First.SyncAsync();
var host = await harness.First.FindAsync(entityId);
host.Secret.Notes.ShouldBe("mine");
host.IsBlocked.ShouldBeTrue();
host.HasUnsyncedChanges.ShouldBeTrue();
}
[Fact]
public async Task ARekeyedVault_IsReportedRatherThanShowingAnEmptyList()
{
// After a rekey this client's grant is stale, so items pulled meanwhile cannot be read. Silently
// showing nothing would look exactly like an empty vault.
await harness.First.CreateAsync(Host("prod-db"));
await harness.SettleAsync();
harness.Server.KeyGeneration = 2;
var report = await harness.First.SyncAsync();
report.RekeyRequired.ShouldBeTrue();
report.ServerKeyGeneration.ShouldBe(2u);
report.NeedsAttention.ShouldBeTrue();
}
// ---- The overall property ----
[Fact]
public async Task AfterAnInterleavedSession_BothMachinesAgree()
{
// Convergence, over a fixed script that exercises creates, edits, a delete and a resurrection at
// once. Two devices that ended up with different host lists would be the worst possible outcome
// for a synced vault, and no single-case test rules it out.
var shared = await harness.First.CreateAsync(Host("a-shared"));
var doomed = await harness.First.CreateAsync(Host("b-doomed"));
await harness.SettleAsync();
await harness.First.UpdateAsync(shared, Host("a-shared", notes: "laptop note"));
await harness.Second.UpdateAsync(shared, Host("a-shared", username: "desktop-user"));
await harness.First.DeleteAsync(doomed);
await harness.Second.UpdateAsync(doomed, Host("b-doomed", notes: "still wanted"));
await harness.First.CreateAsync(Host("c-laptop-only"));
await harness.Second.CreateAsync(Host("d-desktop-only"));
await harness.SettleAsync();
await harness.SettleAsync();
var first = await harness.First.HostsSortedAsync();
var second = await harness.Second.HostsSortedAsync();
first.ShouldBe(second);
first.Count.ShouldBe(4);
first.Select(host => host.Label).ShouldBe(
["a-shared", "b-doomed (restored)", "c-laptop-only", "d-desktop-only"]);
// The merged host kept both sides' contributions.
var merged = first.Single(host => string.Equals(host.Label, "a-shared", StringComparison.Ordinal));
merged.Notes.ShouldBe("laptop note");
merged.Username.ShouldBe("desktop-user");
// And nothing is still queued anywhere.
(await harness.First.Outbox.TakeAsync(VaultId, 100, TestContext.Current.CancellationToken))
.ShouldBeEmpty();
(await harness.Second.Outbox.TakeAsync(VaultId, 100, TestContext.Current.CancellationToken))
.ShouldBeEmpty();
}
// ---- Helpers ----
private static Contracts.EncryptedPayload ForeignPayload() =>
// Deliberately not decryptable: this stands in for another user's item, and the point of the test
// is only that the change is not skipped. Pulling never decrypts, so nothing here needs to open.
new([1, 2, 3], [4, 5], Guid.CreateVersion7(), 1, Crypto.CryptoSpec.CurrentAadVersion);
/// <summary>
/// Sends a device's queued operation straight to the server, leaving the outbox row in place.
/// </summary>
/// <remarks>
/// Reproduces the one situation the engine cannot reach on its own: a push that the server applied
/// and whose answer never came back. That window is where an idempotency key either saves the newer
/// edit or destroys it.
/// </remarks>
private async Task PushBehindTheEnginesBackAsync(Guid entityId)
{
var pending = await harness.First.Outbox.FindAsync(
VaultId, Contracts.SyncEntityType.Host, entityId, TestContext.Current.CancellationToken);
pending.ShouldNotBeNull();
await harness.Server.SyncPushAsync(
VaultId,
new Contracts.SyncPushRequest(
[
new Contracts.SyncPushOperation(
pending.OperationId,
pending.EntityType,
pending.EntityId,
pending.Operation,
pending.ExpectedVersion,
pending.Payload,
pending.Fields),
]),
TestContext.Current.CancellationToken);
}
private async Task<IReadOnlyList<ConflictKind>> ConflictsAcrossDevicesAsync()
{
var first = await harness.First.ConflictsAsync();
var second = await harness.Second.ConflictsAsync();
return [.. first.Concat(second).Select(conflict => conflict.Kind)];
}
private async Task<IReadOnlyList<string>> DiscardedValuesAsync()
{
var first = await harness.First.ConflictsAsync();
var second = await harness.Second.ConflictsAsync();
return
[
.. first.Concat(second)
.Select(conflict => System.Text.Encoding.UTF8.GetString(conflict.Detail)),
];
}
}