Publish runtime metrics and byte-weighted progress via System.Diagnostics.Metrics

ARCHITECTURE.md

@@ -75,6 +75,14 @@ All external process execution uses [CliWrap](https://github.com/Tyrrrz/CliWrap)
 - See `MediaTool.cs` for base execution patterns
 - All tool execution supports cancellation via `Program.CancelToken()`
 
+### Runtime Metrics
+
+`Metrics.cs` owns a single `System.Diagnostics.Metrics.Meter` (`PlexCleaner.Process`) published for the whole process and read externally with `dotnet-counters` (no config flag, and instruments are inert until observed).
+
+- Hooks: `ProcessDriver.ProcessFiles` (the choke point every command and monitor cycle funnels through) drives the file/byte/in-flight instruments and the byte-weighted `progress.ratio`. `Process.ProcessFiles` records the per-`SidecarFile.StatesType` outcomes. `MediaTool` execution paths record `tool.duration`.
+- Progress is weighted by input bytes (summed once up front and credited at completion from the same map), not file count.
+- Run-scoped gauges (totals, in-flight, progress, ETA) reset per `ProcessFiles` call, while the counters stay cumulative for the process. All writers use `Interlocked`, so the parallel loop needs no lock, and observable-gauge callbacks only read. Tags are bounded (`state`, `tool`) - no filename tags.
+
 ### Sidecar File System
 
 Critical performance feature - DO NOT break compatibility:

Docker/Dockerfile

@@ -134,6 +134,9 @@ COPY --chmod=ug=rwx,o=rx ./Docker/InstallDebugTools.sh ./
 RUN ./InstallDebugTools.sh \
     && rm -rf ./InstallDebugTools.sh
 
+# Wrapper to read the PlexCleaner.Process metrics: "docker exec <container> counters"
+COPY --chmod=ug=rwx,o=rx ./Docker/counters.sh /usr/local/bin/counters
+
 # Copy version script
 COPY --chmod=ug=rwx,o=rx ./Docker/Version.sh /PlexCleaner/
 

Docker/counters.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+
+# Run dotnet-counters against the in-container PlexCleaner process.
+# The single-file tool needs a writable extract dir, and the app is PID 1 in the container.
+# Usage from the host:
+#   docker exec <container> counters             # live monitor of the PlexCleaner.Process meter
+#   docker exec <container> counters collect ... # any other dotnet-counters verb/args, passed through
+
+set -euo pipefail
+
+export DOTNET_BUNDLE_EXTRACT_BASE_DIR="${DOTNET_BUNDLE_EXTRACT_BASE_DIR:-/tmp}"
+
+if [[ $# -eq 0 ]]; then
+    set -- monitor -p 1 --counters PlexCleaner.Process
+fi
+
+exec /dotnet-tools/dotnet-counters "$@"

HISTORY.md

@@ -4,6 +4,13 @@ Utility to optimize media files for Direct Play in Plex, Emby, Jellyfin, etc.
 
 ## Release History
 
+- Version 3.22:
+  - Added always-on runtime metrics published via `System.Diagnostics.Metrics` under the `PlexCleaner.Process` meter, readable with `dotnet-counters` with no extra infrastructure.
+    - Overall progress is weighted by input bytes rather than file count, so a run mixing tiny and huge files reports the actual work completed. In this first version an in-flight file contributes no partial credit until it finishes.
+    - Instruments include the run file and byte totals, in-flight and active-thread counts, completed bytes, the weighted `progress.ratio` and `eta.seconds`, cumulative per-outcome counters (completed, modified, errors, verify-failed, and a per-`State`-flag tally), and the `file.duration` and per-tool `tool.duration` histograms. Metrics are aggregate only, with bounded `state` and `tool` tags and no filename tags.
+    - The run-scoped gauges reset at the start of every processing pass, so monitor mode and back-to-back commands each report their own run, while the counters stay cumulative for rate display.
+    - Instruments are inert until a listener observes them, so the feature is always on with no configuration flag and negligible idle overhead.
+    - The Docker image ships a `counters` wrapper, so reading the metrics is `docker exec <container> counters`.
 - Version 3.21:
   - Repair non-monotonic DTS muxer warnings losslessly instead of failing repair permanently.
     - `ffmpeg -f null` can exit `0` yet emit `Application provided invalid, non monotonically increasing dts to muxer` for files that may decode and play correctly.

PlexCleaner/FfProbeTool.cs

@@ -1,3 +1,4 @@
+using System.Diagnostics;
 using System.Runtime.CompilerServices;
 using System.Text;
 using System.Text.Json;
@@ -79,6 +80,7 @@ public bool GetPackets(
         )
         {
             int processId = -1;
+            long startTimestamp = Stopwatch.GetTimestamp();
             try
             {
                 // Pipe target to deserialize JSON packets
@@ -194,6 +196,13 @@ public bool GetPackets(
             {
                 return (false, string.Empty);
             }
+            finally
+            {
+                Metrics.RecordToolDuration(
+                    GetToolType(),
+                    Stopwatch.GetElapsedTime(startTimestamp).TotalMilliseconds
+                );
+            }
         }
 
         public bool GetClosedCaptions(string fileName, out bool hasClosedCaptions)

PlexCleaner/MediaTool.cs

@@ -1,3 +1,4 @@
+using System.Diagnostics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Text;
@@ -173,6 +174,7 @@ public bool Execute(
     {
         bufferedCommandResult = null!;
         int processId = -1;
+        long startTimestamp = Stopwatch.GetTimestamp();
         try
         {
             StringBuilder stdOutBuilder = new();
@@ -223,6 +225,13 @@ public bool Execute(
         {
             return false;
         }
+        finally
+        {
+            Metrics.RecordToolDuration(
+                GetToolType(),
+                Stopwatch.GetElapsedTime(startTimestamp).TotalMilliseconds
+            );
+        }
     }
 
     public bool ExecuteStreamStdErr(
@@ -234,6 +243,7 @@ public bool ExecuteStreamStdErr(
     {
         exitCode = -1;
         int processId = -1;
+        long startTimestamp = Stopwatch.GetTimestamp();
         try
         {
             // Stream stderr line by line to the caller instead of buffering it
@@ -285,6 +295,13 @@ public bool ExecuteStreamStdErr(
         {
             return false;
         }
+        finally
+        {
+            Metrics.RecordToolDuration(
+                GetToolType(),
+                Stopwatch.GetElapsedTime(startTimestamp).TotalMilliseconds
+            );
+        }
     }
 
     public static PipeTarget ToStringBuilder(StringBuilder stringBuilder) =>

PlexCleaner/Metrics.cs

@@ -0,0 +1,193 @@
+using System.Diagnostics;
+using System.Diagnostics.Metrics;
+
+namespace PlexCleaner;
+
+internal static class Metrics
+{
+    private static readonly Meter s_meter = new("PlexCleaner.Process");
+
+    // Cumulative for the process lifetime, not reset between runs.
+    private static readonly Counter<long> s_filesCompleted = s_meter.CreateCounter<long>(
+        "plexcleaner.files.completed",
+        description: "Files finished, any outcome"
+    );
+    private static readonly Counter<long> s_filesModified = s_meter.CreateCounter<long>(
+        "plexcleaner.files.modified",
+        description: "Files whose media was changed"
+    );
+    private static readonly Counter<long> s_filesErrors = s_meter.CreateCounter<long>(
+        "plexcleaner.files.errors",
+        description: "Files that errored"
+    );
+    private static readonly Counter<long> s_filesVerifyFailed = s_meter.CreateCounter<long>(
+        "plexcleaner.files.verifyfailed",
+        description: "Files that failed verification"
+    );
+    private static readonly Counter<long> s_filesProcessed = s_meter.CreateCounter<long>(
+        "plexcleaner.files.processed",
+        description: "Per-outcome tally, tagged by each State flag set"
+    );
+
+    private static readonly Histogram<double> s_fileDuration = s_meter.CreateHistogram<double>(
+        "plexcleaner.file.duration",
+        unit: "ms",
+        description: "Per-file wall-clock time"
+    );
+    private static readonly Histogram<double> s_toolDuration = s_meter.CreateHistogram<double>(
+        "plexcleaner.tool.duration",
+        unit: "ms",
+        description: "Per media-tool invocation time, tagged by tool"
+    );
+
+    // Run-scoped state, reset by BeginRun, read by the observable gauges.
+    // All access is via Interlocked so the parallel loop needs no lock.
+    private static long s_runFilesTotal;
+    private static long s_runBytesTotal;
+    private static long s_runBytesCompleted;
+    private static long s_runInflight;
+    private static long s_runStartTimestamp;
+
+    // Each State flag (minus None) with its tag, pre-built once so RecordStates allocates nothing per file.
+    private static readonly (
+        SidecarFile.StatesType Flag,
+        KeyValuePair<string, object?> Tag
+    )[] s_stateTags =
+    [
+        .. Enum.GetValues<SidecarFile.StatesType>()
+            .Where(flag => flag != SidecarFile.StatesType.None)
+            .Select(flag => (flag, new KeyValuePair<string, object?>("state", flag.ToString()))),
+    ];
+
+    // Each tool with its pre-built tag, so RecordToolDuration allocates nothing per invocation.
+    private static readonly Dictionary<
+        MediaTool.ToolType,
+        KeyValuePair<string, object?>
+    > s_toolTags = Enum.GetValues<MediaTool.ToolType>()
+        .ToDictionary(
+            tool => tool,
+            tool => new KeyValuePair<string, object?>("tool", tool.ToString())
+        );
+
+    static Metrics()
+    {
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.files.total",
+            () => Interlocked.Read(ref s_runFilesTotal),
+            description: "Files in the current run"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.files.inflight",
+            () => Interlocked.Read(ref s_runInflight),
+            description: "Files currently processing"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.threads.active",
+            () => (long)(Program.Options?.ThreadCount ?? 0),
+            description: "Configured worker threads"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.bytes.total",
+            () => Interlocked.Read(ref s_runBytesTotal),
+            unit: "By",
+            description: "Sum of input sizes in the current run"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.bytes.completed",
+            () => Interlocked.Read(ref s_runBytesCompleted),
+            unit: "By",
+            description: "Bytes of finished files (no partial credit for in-flight files)"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.progress.ratio",
+            ComputeProgress,
+            description: "Byte-weighted overall progress [0..1]"
+        );
+        _ = s_meter.CreateObservableGauge(
+            "plexcleaner.eta.seconds",
+            ComputeEtaSeconds,
+            unit: "s",
+            description: "Estimated time remaining"
+        );
+    }
+
+    // Reset the run-scoped gauges and ETA clock, called once per ProcessFiles run.
+    internal static void BeginRun(long totalFiles, long totalBytes)
+    {
+        _ = Interlocked.Exchange(ref s_runFilesTotal, totalFiles);
+        _ = Interlocked.Exchange(ref s_runBytesTotal, totalBytes);
+        _ = Interlocked.Exchange(ref s_runBytesCompleted, 0);
+        _ = Interlocked.Exchange(ref s_runInflight, 0);
+        _ = Interlocked.Exchange(ref s_runStartTimestamp, Stopwatch.GetTimestamp());
+    }
+
+    internal static void FileStarted() => Interlocked.Increment(ref s_runInflight);
+
+    internal static void FileInflightDone() => Interlocked.Decrement(ref s_runInflight);
+
+    // A finished file credits its whole size (no partial credit in v1) and its wall-clock time.
+    internal static void FileCompleted(long sizeBytes, TimeSpan wall)
+    {
+        _ = Interlocked.Add(ref s_runBytesCompleted, sizeBytes);
+        s_filesCompleted.Add(1);
+        s_fileDuration.Record(wall.TotalMilliseconds);
+    }
+
+    internal static void FileErrored() => s_filesErrors.Add(1);
+
+    internal static void RecordModified() => s_filesModified.Add(1);
+
+    internal static void RecordVerifyFailed() => s_filesVerifyFailed.Add(1);
+
+    internal static void RecordStates(SidecarFile.StatesType state)
+    {
+        foreach ((SidecarFile.StatesType flag, KeyValuePair<string, object?> tag) in s_stateTags)
+        {
+            if ((state & flag) == flag)
+            {
+                s_filesProcessed.Add(1, tag);
+            }
+        }
+    }
+
+    internal static void RecordToolDuration(MediaTool.ToolType tool, double milliseconds) =>
+        s_toolDuration.Record(milliseconds, s_toolTags[tool]);
+
+    internal static void Dispose() => s_meter.Dispose();
+
+    // Byte-weighted progress, guards a zero (or not-yet-started) total.
+    internal static double ComputeProgress()
+    {
+        long total = Interlocked.Read(ref s_runBytesTotal);
+        return total <= 0 ? 0.0 : (double)Interlocked.Read(ref s_runBytesCompleted) / total;
+    }
+
+    // Linear extrapolation from weighted progress and elapsed time.
+    // Returns 0 before any progress and never a non-finite value.
+    internal static double ComputeEtaSeconds()
+    {
+        double ratio = ComputeProgress();
+        if (ratio <= 0.0)
+        {
+            return 0.0;
+        }
+        double elapsed = Stopwatch
+            .GetElapsedTime(Interlocked.Read(ref s_runStartTimestamp))
+            .TotalSeconds;
+        double eta = elapsed * (1.0 - ratio) / ratio;
+        return double.IsFinite(eta) ? eta : 0.0;
+    }
+
+    internal static IEnumerable<SidecarFile.StatesType> EnumerateSetStates(
+        SidecarFile.StatesType state
+    )
+    {
+        foreach ((SidecarFile.StatesType flag, _) in s_stateTags)
+        {
+            if ((state & flag) == flag)
+            {
+                yield return flag;
+            }
+        }
+    }
+}

PlexCleaner/Process.cs

@@ -506,6 +506,17 @@ out string? failedOperation
                     return processResult;
                 }
 
+                // Per-outcome metrics: this is the only place the State flags are known
+                Metrics.RecordStates(state);
+                if (modified)
+                {
+                    Metrics.RecordModified();
+                }
+                if ((state & SidecarFile.StatesType.VerifyFailed) != 0)
+                {
+                    Metrics.RecordVerifyFailed();
+                }
+
                 // Save result
                 lock (resultLock)
                 {