missions/tests/Azaion.Missions.E2E.Tests/Helpers/MissionsContainerHelper.cs

using System.Diagnostics;
using System.Globalization;
using System.Net;

namespace Azaion.Missions.E2E.Helpers;

/// <summary>
/// Spawns standalone <c>azaion/missions:test</c> containers via <c>docker run</c>
/// (NOT compose) so startup-time behavior can be exercised independently of
/// the long-running compose stack. Used by NFT-SEC-12, NFT-SEC-13,
/// NFT-RES-05, NFT-RES-06 — each provides its own env override map and asserts
/// against the captured exit code + logs.
/// </summary>
/// <remarks>
/// Like <see cref="Fixtures.ComposeRestartFixture"/>, this helper is gated on
/// <c>COMPOSE_RESTART_ENABLED=1</c> and a docker CLI on PATH; tests using it
/// must <see cref="Xunit.Skip.IfNot(bool, string)"/> when the gate fails so
/// CI environments without Docker access skip with an explicit reason
/// instead of silently passing.
/// </remarks>
public static class MissionsContainerHelper
{
    public const string MissionsImageEnvVar = "MISSIONS_TEST_IMAGE";
    public const string DefaultMissionsImage = "azaion/missions:test";
    public const string NetworkEnvVar = "MISSIONS_TEST_NETWORK";
    public const string DefaultNetwork = "missions-e2e-net";

    public static bool Enabled =>
        Environment.GetEnvironmentVariable("COMPOSE_RESTART_ENABLED") == "1";

    public static string Image =>
        Environment.GetEnvironmentVariable(MissionsImageEnvVar) ?? DefaultMissionsImage;

    public static string Network =>
        Environment.GetEnvironmentVariable(NetworkEnvVar) ?? DefaultNetwork;

    /// <summary>
    /// Runs <c>docker run --rm --name &lt;name&gt; --network &lt;net&gt; &lt;env&gt; &lt;image&gt;</c>,
    /// waits for the container to exit (up to <paramref name="timeout"/>),
    /// and returns its exit code + combined logs. Forces removal of any
    /// stale container with the same name before starting (an earlier crash
    /// can leave a stopped container behind).
    /// </summary>
    public static RunResult RunUntilExit(
        string containerName,
        IReadOnlyDictionary<string, string> envOverrides,
        TimeSpan timeout)
    {
        ForceRemove(containerName);
        var args = BuildRunArgs(containerName, envOverrides);
        Run("docker", args, out var runStdout, out var runStderr);

        var deadline = DateTime.UtcNow + timeout;
        while (DateTime.UtcNow < deadline)
        {
            if (TryGetExitCode(containerName, out var exitCode))
            {
                var logs = ReadLogs(containerName);
                ForceRemove(containerName);
                return new RunResult(exitCode, logs, runStdout, runStderr);
            }
            Thread.Sleep(250);
        }

        var partialLogs = ReadLogs(containerName);
        ForceRemove(containerName);
        throw new TimeoutException(
            $"container '{containerName}' did not exit within {timeout.TotalSeconds:F0}s. " +
            $"Partial logs:\n{partialLogs}");
    }

    /// <summary>
    /// Captures <c>docker inspect --format '{{.State.StartedAt}}'</c> for a
    /// running container, returned as a stable ISO-8601 string. Used by
    /// NFT-RES-07 to assert the missions service did NOT restart during a
    /// JWKS rotation flow.
    /// </summary>
    public static string GetStartedAt(string containerName)
    {
        Run("docker",
            $"inspect --format '{{{{.State.StartedAt}}}}' {containerName}",
            out var stdout, out _);
        return stdout.Trim().Trim('\'');
    }

    /// <summary>
    /// Starts a missions container detached (<c>-d</c>) and polls its <c>/health</c>
    /// endpoint over the shared e2e network until it responds 200 (or
    /// <paramref name="readyTimeout"/> elapses). Used by tests that need a
    /// running SUT with non-default env (NFT-SEC-12 HTTP-not-HTTPS,
    /// NFT-SEC-13 CORS preflight) — the test then drives the container
    /// over the network and reads <c>docker logs</c> for log-line assertions.
    /// </summary>
    public static async Task<DetachedContainer> StartAndWaitForHealthAsync(
        string containerName,
        IReadOnlyDictionary<string, string> envOverrides,
        TimeSpan readyTimeout)
    {
        ForceRemove(containerName);
        var args = BuildRunArgs(containerName, envOverrides);
        Run("docker", args, out _, out _);

        using var http = new HttpClient { Timeout = TimeSpan.FromSeconds(2) };
        var healthUrl = new Uri($"http://{containerName}:8080/health");
        var deadline = DateTime.UtcNow + readyTimeout;
        while (DateTime.UtcNow < deadline)
        {
            try
            {
                using var resp = await http.GetAsync(healthUrl);
                if (resp.StatusCode == HttpStatusCode.OK)
                    return new DetachedContainer(containerName);
            }
            catch (HttpRequestException) { /* container not yet listening */ }
            catch (TaskCanceledException) { /* slow first response */ }
            await Task.Delay(500);
        }

        // Health never came up — capture logs for the failure message before
        // tearing down, so the test reporter shows why the harness gave up.
        var logs = ReadLogs(containerName);
        ForceRemove(containerName);
        throw new TimeoutException(
            $"container '{containerName}' did not become healthy within {readyTimeout.TotalSeconds:F0}s. " +
            $"Logs:\n{logs}");
    }

    public sealed class DetachedContainer : IDisposable
    {
        public string Name { get; }
        public DetachedContainer(string name) => Name = name;
        public string ReadLogs() => MissionsContainerHelper.ReadLogs(Name);
        public void Dispose() => ForceRemove(Name);
    }

    private static string BuildRunArgs(
        string containerName,
        IReadOnlyDictionary<string, string> envOverrides)
    {
        var sb = new System.Text.StringBuilder();
        sb.Append("run --rm -d ");
        sb.Append("--name ").Append(containerName).Append(' ');
        sb.Append("--network ").Append(Network).Append(' ');
        foreach (var (key, value) in envOverrides)
        {
            sb.Append("-e ").Append(key).Append('=').Append('"')
              .Append(value.Replace("\"", "\\\"", StringComparison.Ordinal))
              .Append("\" ");
        }
        sb.Append(Image);
        return sb.ToString();
    }

    private static bool TryGetExitCode(string containerName, out int exitCode)
    {
        // `docker inspect` succeeds while the container exists (running OR
        // exited). Once `--rm` removes it the inspect call fails — but we
        // already captured exitCode by then.
        var psi = new ProcessStartInfo("docker",
            $"inspect --format '{{{{.State.Running}}}} {{{{.State.ExitCode}}}}' {containerName}")
        {
            RedirectStandardOutput = true,
            RedirectStandardError = true,
            UseShellExecute = false
        };
        using var p = Process.Start(psi)
            ?? throw new InvalidOperationException("docker CLI not available");
        var stdout = p.StandardOutput.ReadToEnd();
        p.WaitForExit();
        if (p.ExitCode != 0)
        {
            // Container is gone (already removed); treat as "still in flight".
            exitCode = 0;
            return false;
        }
        var parts = stdout.Trim().Trim('\'').Split(' ', StringSplitOptions.RemoveEmptyEntries);
        if (parts.Length < 2 ||
            !bool.TryParse(parts[0], out var running) ||
            !int.TryParse(parts[1], NumberStyles.Integer, CultureInfo.InvariantCulture, out exitCode))
        {
            exitCode = 0;
            return false;
        }
        return !running;
    }

    internal static string ReadLogs(string containerName)
    {
        var psi = new ProcessStartInfo("docker", $"logs {containerName}")
        {
            RedirectStandardOutput = true,
            RedirectStandardError = true,
            UseShellExecute = false
        };
        using var p = Process.Start(psi);
        if (p is null) return string.Empty;
        var stdout = p.StandardOutput.ReadToEnd();
        var stderr = p.StandardError.ReadToEnd();
        p.WaitForExit();
        return stdout + stderr;
    }

    private static void ForceRemove(string containerName)
    {
        var psi = new ProcessStartInfo("docker", $"rm -f {containerName}")
        {
            RedirectStandardOutput = true,
            RedirectStandardError = true,
            UseShellExecute = false
        };
        try
        {
            using var p = Process.Start(psi);
            p?.WaitForExit();
        }
        catch (System.ComponentModel.Win32Exception)
        {
            // docker CLI absent — let the caller's Enabled check surface the issue.
            throw new InvalidOperationException(
                "docker CLI not available in test container; " +
                "MissionsContainerHelper requires docker access (set COMPOSE_RESTART_ENABLED=1 and mount the socket).");
        }
    }

    private static void Run(string file, string args, out string stdout, out string stderr)
    {
        var psi = new ProcessStartInfo(file, args)
        {
            RedirectStandardOutput = true,
            RedirectStandardError = true,
            UseShellExecute = false
        };
        using var p = Process.Start(psi)
            ?? throw new InvalidOperationException($"failed to launch `{file} {args}`");
        stdout = p.StandardOutput.ReadToEnd();
        stderr = p.StandardError.ReadToEnd();
        p.WaitForExit();
        if (p.ExitCode != 0)
            throw new InvalidOperationException(
                $"`{file} {args}` exited {p.ExitCode}.\nstdout: {stdout}\nstderr: {stderr}");
    }

    public sealed record RunResult(int ExitCode, string Logs, string RunStdout, string RunStderr);
}