lmxopcua/tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.Historian.QuestDb.IntegrationTests/LiveWriteReadRaceStressTests.cs

using System.Collections.Concurrent;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Driver.Historian.Sql.Common;

namespace ZB.MOM.WW.OtOpcUa.Driver.Historian.QuestDb.IntegrationTests;

/// <summary>
///     Live-gate 8 F1 (#680 follow-up): per-sample processed reads of a tag that is being written. The StartEnd and
///     ActualTime queries reference their classified rows more than once; QuestDB evaluates each reference with its own
///     table reader, so a row committed between two of them produced a bucket whose counts included a sample its
///     first/last (or extreme) join could not find — a NULL first-of-any-quality time threw <c>InvalidCastException</c>
///     and failed the node's whole read (~1 in 90k reads on the rig), and ActualTime returned a value-less Good. StartEnd
///     now reads the table once; ActualTime keeps its form and an inconsistent bucket evaluates to <c>BadNoData</c>. This
///     test interleaves a continuous single-row writer with tight-loop reads and asserts that no read throws and no
///     non-Bad result is value-less.
/// </summary>
/// <remarks>
///     Opt-in: runs only when <c>HISTORIAN_QUESTDB_STRESS_SECONDS</c> is set (and the fixture's connection is), so it never
///     slows the default suite. Category <c>SlowUnit</c> as well, for filter-based exclusion.
/// </remarks>
[Collection(QuestDbCollection.Name)]
[Trait("Category", "SlowUnit")]
public sealed class LiveWriteReadRaceStressTests(QuestDbFixture fx)
{
    /// <summary>Seconds to run the race loop for; absent ⇒ the test skips.</summary>
    public const string SecondsEnvVar = "HISTORIAN_QUESTDB_STRESS_SECONDS";

    private static readonly HistoryAggregateType[] Aggregates =
    [
        HistoryAggregateType.Start, HistoryAggregateType.End, HistoryAggregateType.Delta,
        HistoryAggregateType.MinimumActualTime, HistoryAggregateType.MaximumActualTime,
    ];

    [Fact]
    public async Task Per_sample_reads_of_a_tag_being_written_never_fault_and_never_return_a_valueless_non_bad_result()
    {
        if (!int.TryParse(Environment.GetEnvironmentVariable(SecondsEnvVar), out var seconds) || seconds <= 0)
            Assert.Skip($"{SecondsEnvVar} is not set; the live write/read race stress test is opt-in.");
        await using var b = await QuestDbConformanceBackend.CreateConcreteAsync(fx);
        const string tag = "Race";
        await b.NewCatalog().EnsureAsync(tag, 0, CancellationToken.None);
        var src = b.CreateDataSource();

        using var stop = new CancellationTokenSource(TimeSpan.FromSeconds(seconds));
        long written = 0, reads = 0;
        var faults = new ConcurrentBag<string>();

        var writer = Task.Run(async () =>
        {
            var i = 0;
            while (!stop.IsCancellationRequested)
            {
                // One committed row per statement (one WAL transaction each), every 7th Bad, never waiting for visibility.
                var status = ++i % 7 == 0 ? HistorianStatusCodes.Bad : 0u;
                await b.InsertRowsNoWaitAsync(tag, [(DateTime.UtcNow, HistorianValue.FromDouble(i), status, false)]);
                Interlocked.Increment(ref written);
            }
        });

        var readers = Enumerable.Range(0, 4).Select(_ => Task.Run(async () =>
        {
            while (!stop.IsCancellationRequested)
            {
                var now = DateTime.UtcNow;
                var start = new DateTime(now.Ticks / TimeSpan.TicksPerSecond / 10 * 10 * TimeSpan.TicksPerSecond, DateTimeKind.Utc)
                    .AddSeconds(-30);
                var end = start.AddSeconds(70);
                foreach (var agg in Aggregates)
                foreach (var reverse in new[] { false, true })
                {
                    try
                    {
                        var r = await src.ReadProcessedAsync(tag, reverse ? end : start, reverse ? start : end,
                            TimeSpan.FromSeconds(10), agg, CancellationToken.None);
                        foreach (var v in r.Samples)
                            if (!StatusBits.IsBad(v.StatusCode) && v.Value is null)
                                faults.Add($"{agg} {(reverse ? "rev" : "fwd")}: non-Bad 0x{v.StatusCode:X8} with no value at {v.SourceTimestampUtc:O}");
                    }
                    catch (Exception ex)
                    {
                        faults.Add($"{agg} {(reverse ? "rev" : "fwd")}: {ex.GetType().Name}: {ex.Message}");
                    }
                    Interlocked.Increment(ref reads);
                }
            }
        })).ToArray();

        await Task.WhenAll(readers.Append(writer));
        var summary = $"race stress: {seconds}s, {Interlocked.Read(ref written)} rows written, {Interlocked.Read(ref reads)} reads, {faults.Count} faults";
        var output = TestContext.Current.TestOutputHelper;
        output?.WriteLine(summary);
        foreach (var f in faults.Take(20)) output?.WriteLine("fault: " + f);
        written.ShouldBeGreaterThan(0);
        faults.ShouldBeEmpty(summary);
    }
}