feat(stream): retry a dropped push instead of ending the session (TASK 32·1)

A dead encoder used to run the whole stop path, so a one-second blip ended the broadcast. Now a dropped
push is relaunched against the same ingest while the broadcast is still alive.

- `Services/StreamRetryPolicy.cs` — immediate, +1 s, +2 s, +3 s. Minimum waits on purpose (creator
  ruling): YouTube's grace is short, so every second not pushing is grace spent.
- `Services/StreamRecovery.cs` — the loop, with the audio pipe recreated before every attempt (its
  server is single-instance and unusable after the old client dies, so a naive relaunch fails on a
  missing input). After a push comes up it asks the API whether the broadcast survived: "complete"
  ends recovery as `BroadcastComplete` instead of showing a live sign over a dead stream; an API
  hiccup counts as recovered rather than ending a healthy session.
- `FramePump.EncoderFailed` — a dead push is recoverable; a compositor fault still goes through
  `Failed` and ends the session, because retrying it would loop.
- `MainViewModel.Recovery.cs` — wires the steps, reports each episode into the left panel alert stack
  (one row per episode, updated in place, log range painted), and cancels on Stop.
- Alert stack polish: a rule and a gap now separate the stack from the scene layers (they read as one
  list before), and the licence indicator uses the creator's wording — `STEAM ID NOT PRESENT` /
  `Steam ID Limit: 30fps` / `Click for More Info`, becoming `STEAM ID PRESENT` / `60fps` with a close
  control once owned.

Tests: `StreamRecoveryTests` (schedule, recover on attempt 2 with the pipe restarted each time,
complete-broadcast honesty, unknown liveness, give-up, cancel, audio-restart fault),
`FramePumpTests` now pin that process death raises `EncoderFailed` and not `Failed`.
FAST gate: 538 passed (one documented pre-existing audio failure), 0 warnings.
This commit is contained in:
2026-11-05 12:09:38 -08:00
parent a80108dd96
commit 72cb3ad55a
16 changed files with 583 additions and 35 deletions
+3 -3
View File
@@ -99,10 +99,10 @@ public sealed class AboutHubTests
// The old red panel under the quality picker is gone (2026-11-05): the licence state
// lives in the left panel's alert stack now, as a pinned status indicator.
Assert.Null(window.FindName("EntitlementErrorPanel"));
var licenceRow = vm.AttentionAlerts.Single(alert => alert.Label == "STEAM LICENCE");
var licenceRow = vm.AttentionAlerts.Single(alert => alert.Label == "STEAM ID NOT PRESENT");
Assert.True(licenceRow.IsPinned);
Assert.Contains("Entitlement 30 fps", licenceRow.Text);
Assert.Contains("Maximum 60 fps", licenceRow.Text);
Assert.Contains("Steam ID Limit: 30fps", licenceRow.Text);
Assert.Contains("Click for More Info", licenceRow.Text);
var picker = (ComboBox)window.FindName("QualityPicker");
Assert.True(vm.CanChangeQuality);
+10 -9
View File
@@ -32,15 +32,15 @@ public sealed class AttentionAlertStackTests
vm.EntitlementOverride = new FakeEntitlement(false, "Steam licensing is not configured in this build.");
var row = Assert.Single(vm.AttentionAlerts);
Assert.Equal("STEAM LICENCE", row.Label);
Assert.Equal("STEAM ID NOT PRESENT", row.Label);
Assert.True(row.IsPinned);
Assert.True(row.ShowsPin);
Assert.False(row.CanDismiss);
Assert.Equal(AttentionAlertAction.About, row.Action);
// A status readout: what this build is entitled to, and the maximum.
Assert.Contains("Entitlement 30 fps", row.Text);
Assert.Contains("Maximum 60 fps", row.Text);
// Creator wording: the limit this build runs at, and where to find out why.
Assert.Contains("Steam ID Limit: 30fps", row.Text);
Assert.Contains("Click for More Info", row.Text);
Assert.DoesNotContain("unavailable", row.Text, StringComparison.OrdinalIgnoreCase);
});
}
@@ -56,18 +56,18 @@ public sealed class AttentionAlertStackTests
Assert.False(row.IsPinned);
Assert.False(row.ShowsPin);
Assert.True(row.CanDismiss);
Assert.Contains("Entitlement 60 fps", row.Text);
Assert.Contains("Steam ID Limit: 60fps", row.Text);
vm.DismissAlertCommand.Execute(row);
Assert.Empty(vm.AttentionAlerts);
// Permanent: a fresh view-model on the same profile must not bring it back…
var reloaded = new MainViewModel();
Assert.DoesNotContain(reloaded.AttentionAlerts, alert => alert.Label == "STEAM LICENCE");
Assert.DoesNotContain(reloaded.AttentionAlerts, alert => alert.Label.StartsWith("STEAM ID"));
// …but losing ownership restores it as a pinned indicator: a change of state, not a preference.
reloaded.EntitlementOverride = new FakeEntitlement(false, "Steam licensing is not configured in this build.");
var restored = reloaded.AttentionAlerts.First(alert => alert.Label == "STEAM LICENCE");
var restored = reloaded.AttentionAlerts.First(alert => alert.Label == "STEAM ID NOT PRESENT");
Assert.True(restored.IsPinned);
});
}
@@ -89,10 +89,11 @@ public sealed class AttentionAlertStackTests
{
WithVm(vm =>
{
vm.EntitlementOverride = new FakeEntitlement(false, "Steam licensing is not configured in this build.");
vm.PushAttentionAlertForTest("drop-1", "STREAM", "Stream dropped — reconnecting (attempt 1 of 4)", 10, 12);
vm.PushAttentionAlertForTest("drop-2", "STREAM", "Stream dropped — reconnecting (attempt 2 of 4)", 13, 15);
Assert.Equal(new[] { "STEAM LICENCE", "STREAM", "STREAM" }, vm.AttentionAlerts.Select(a => a.Label));
Assert.Equal(new[] { "STEAM ID NOT PRESENT", "STREAM", "STREAM" }, vm.AttentionAlerts.Select(a => a.Label));
// Newest event sits directly under the indicator.
Assert.Contains("attempt 2", vm.AttentionAlerts[1].Text);
Assert.Contains("attempt 1", vm.AttentionAlerts[2].Text);
@@ -104,7 +105,7 @@ public sealed class AttentionAlertStackTests
vm.DismissAlertCommand.Execute(vm.AttentionAlerts[1]);
Assert.Equal(2, vm.AttentionAlerts.Count);
Assert.Equal("STEAM LICENCE", vm.AttentionAlerts[0].Label);
Assert.Equal("STEAM ID NOT PRESENT", vm.AttentionAlerts[0].Label);
});
}
+9 -4
View File
@@ -298,18 +298,23 @@ public class FramePumpTests
}
[Fact]
public async Task ProcessDeath_StopsPump_AndRaisesFailed()
public async Task ProcessDeath_StopsPump_AndRaisesEncoderFailed()
{
var encoder = new FakeEncoder();
var failed = new TaskCompletionSource<string>(TaskCreationOptions.RunContinuationsAsynchronously);
var encoderFailed = new TaskCompletionSource<string>(TaskCreationOptions.RunContinuationsAsynchronously);
var fatal = false;
using var pump = NewPump(encoder);
pump.Failed += (_, m) => failed.TrySetResult(m);
// TASK 32·1: a dead push is recoverable — the broadcast may still be live — so it is a
// separate signal from a fatal pump fault, and the session gets to relaunch it.
pump.EncoderFailed += (_, m) => encoderFailed.TrySetResult(m);
pump.Failed += (_, _) => fatal = true;
await pump.StartAsync();
encoder.RaiseProcessFailed("FFmpeg exited with code 1");
var message = await failed.Task.WaitAsync(TimeSpan.FromSeconds(5));
var message = await encoderFailed.Task.WaitAsync(TimeSpan.FromSeconds(5));
Assert.Contains("1", message);
Assert.False(fatal, "an encoder death must not be reported as a fatal pump failure");
// The pump self-stops (fire-and-forget); poll the definitive teardown
// marker — the encoder's disposal — rather than the IsRunning flag, which
+162
View File
@@ -0,0 +1,162 @@
using System;
using System.Collections.Generic;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;
using Xunit;
using ytLive.Services;
namespace ytLive.Tests;
/// <summary>
/// TASK 32·1 — blip retry, exercised through <see cref="StreamRecovery"/> with fake steps: no ffmpeg,
/// no network, no waiting on the real schedule. The rules that matter: the first attempt is immediate,
/// the audio pipe is recreated before every push, a reconnect that lands on a COMPLETE broadcast is
/// reported honestly instead of pretending to be live, and giving up is a definite outcome, not a hang.
/// </summary>
public class StreamRecoveryTests
{
private static StreamRetryPolicy NoWaits(int attempts = 4)
=> new(Enumerable.Repeat(TimeSpan.Zero, attempts).ToArray());
[Fact]
public void DefaultPolicy_StartsImmediately_ThenOneTwoThreeSeconds()
{
var policy = StreamRetryPolicy.Default;
Assert.Equal(4, policy.Attempts);
Assert.Equal(TimeSpan.Zero, policy.WaitBefore(0));
Assert.Equal(TimeSpan.FromSeconds(1), policy.WaitBefore(1));
Assert.Equal(TimeSpan.FromSeconds(2), policy.WaitBefore(2));
Assert.Equal(TimeSpan.FromSeconds(3), policy.WaitBefore(3));
}
[Fact]
public async Task RecoversOnTheSecondAttempt_AndRestartsTheAudioPipeEachTime()
{
var pushes = 0;
var audioRestarts = 0;
var reported = new List<string>();
var recovery = new StreamRecovery(NoWaits(), new StreamRecoverySteps
{
RestartAudio = _ => { audioRestarts++; return Task.CompletedTask; },
StartPush = _ => Task.FromResult(++pushes > 1), // first start fails, second works
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Live),
Report = reported.Add,
Log = _ => { },
});
var result = await recovery.RunAsync();
Assert.Equal(RecoveryOutcome.Recovered, result.Outcome);
Assert.Equal(2, result.Attempts);
Assert.Equal(2, pushes);
Assert.Equal(2, audioRestarts); // the pipe is unusable after the old client died
Assert.Contains(reported, text => text.Contains("reconnecting (attempt 1"));
Assert.Contains(reported, text => text.StartsWith("Stream recovered"));
}
[Fact]
public async Task ABroadcastThatAlreadyEnded_IsReportedAsComplete_NotAsRecovered()
{
var reported = new List<string>();
var recovery = new StreamRecovery(NoWaits(), new StreamRecoverySteps
{
RestartAudio = _ => Task.CompletedTask,
StartPush = _ => Task.FromResult(true),
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Complete),
Report = reported.Add,
Log = _ => { },
});
var result = await recovery.RunAsync();
Assert.Equal(RecoveryOutcome.BroadcastComplete, result.Outcome);
Assert.Contains(reported, text => text.Contains("cannot be resumed"));
Assert.DoesNotContain(reported, text => text.StartsWith("Stream recovered"));
}
[Fact]
public async Task UnknownLiveness_CountsAsRecovered()
{
// If the API hiccups we cannot prove the broadcast is gone; treating that as failure would
// end sessions that are actually fine.
var recovery = new StreamRecovery(NoWaits(), new StreamRecoverySteps
{
RestartAudio = _ => Task.CompletedTask,
StartPush = _ => Task.FromResult(true),
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Unknown),
Report = _ => { },
Log = _ => { },
});
Assert.Equal(RecoveryOutcome.Recovered, (await recovery.RunAsync()).Outcome);
}
[Fact]
public async Task EveryAttemptFailing_GivesUpWithTheCount()
{
var reported = new List<string>();
var recovery = new StreamRecovery(NoWaits(3), new StreamRecoverySteps
{
RestartAudio = _ => Task.CompletedTask,
StartPush = _ => Task.FromResult(false),
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Live),
Report = reported.Add,
Log = _ => { },
});
var result = await recovery.RunAsync();
Assert.Equal(RecoveryOutcome.GaveUp, result.Outcome);
Assert.Equal(3, result.Attempts);
Assert.Contains(reported, text => text.Contains("off air"));
}
[Fact]
public async Task CancellingMidRecovery_StopsWithoutTouchingThePush()
{
using var cts = new CancellationTokenSource();
var pushes = 0;
var recovery = new StreamRecovery(StreamRetryPolicy.Default, new StreamRecoverySteps
{
RestartAudio = _ => Task.CompletedTask,
StartPush = _ => { pushes++; return Task.FromResult(false); },
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Live),
Report = _ => { },
Log = _ => { },
Delay = (_, ct) => { cts.Cancel(); ct.ThrowIfCancellationRequested(); return Task.CompletedTask; },
});
var result = await recovery.RunAsync(cts.Token);
Assert.Equal(RecoveryOutcome.Cancelled, result.Outcome);
Assert.True(pushes <= 1, $"cancelling must stop the loop, but the push was started {pushes} times");
}
[Fact]
public async Task AAudioRestartFault_IsAnAttemptFailure_NotACrash()
{
var attempts = 0;
var recovery = new StreamRecovery(NoWaits(2), new StreamRecoverySteps
{
RestartAudio = _ =>
{
attempts++;
throw new InvalidOperationException("pipe busy");
},
StartPush = _ => Task.FromResult(true),
CheckBroadcast = _ => Task.FromResult(BroadcastLiveness.Live),
Report = _ => { },
Log = _ => { },
});
var result = await recovery.RunAsync();
Assert.Equal(RecoveryOutcome.GaveUp, result.Outcome);
Assert.Equal(2, attempts);
}
}