perf(diag): per-frame history export + checkpoint LOH/cache counters + soak capped mode (2026-07-24 audit review)

An adversarial performance review found our own instruments cannot
measure the project's own performance gates:

- FrameProfiler aggregated CPU/GPU/alloc/stage samples into ~5-second
  windows and reset the ring buffers after each report, so route-wide
  p50/p95/p99 distributions across a whole soak could not be
  reconstructed after the fact. ACDREAM_FRAME_HISTORY=<path> now opts
  into a separate per-frame history (one record per frame, ~72
  bytes/record, accumulated in memory with zero frame-thread I/O) that
  a shutdown-only Dispose() writes as CSV. The aggregated [frame-prof]
  report format and its existing metrics are unchanged.

- The canonical checkpoint JSON tracked cache residency (entry/byte
  counts) but never LOH size/fragmentation, process-wide allocated
  bytes, or cache hit/miss/eviction traffic — a committed audit JSON
  showed 65% LOH fragmentation that no tracked instrument recorded,
  and "does a revisit portal hit or miss the caches" was unanswerable
  from an artifact alone. WorldLifecycleResourceSnapshot now carries
  loh_size_bytes/loh_fragmentation_bytes (GCMemoryInfo.GenerationInfo
  index 3), process_total_allocated_bytes (GC.GetTotalAllocatedBytes),
  and Interlocked hit/miss/eviction counters for the CPU mesh cache,
  decoded-texture cache, and the four bounded DAT-object caches
  (portal/cell/highRes/language, aggregated).

- run-connected-r6-soak.ps1 unconditionally forced
  ACDREAM_UNCAPPED_RENDER=1 with no capped mode, while its sibling
  lifecycle-gate script correctly gated it behind a switch. Added
  -Uncapped (default capped, matching the sibling script's pattern),
  fixed the stationary dwell (12s -> 26s, past the 25s
  LiveEntityLivenessController deadline the adjacent comment already
  cited), and now write an env-disclosure.json into the automation
  artifact directory before every launch listing every ACDREAM_* var
  the script sets plus -Uncapped, since the prior audit could only see
  ACDREAM_DUMP_MOVE_TRUTH and nothing else was ever recorded anywhere.

Cache counters are wired via the existing composition path
(ObjectMeshManager already owns the CPU mesh cache and the mesh
extractor directly; content.Dats is threaded into
WorldLifecycleResourceSnapshotSource the same way every other
composition consumer receives it). The DAT-object cache lives behind
IDatReaderWriter, a third-party interface from the DatReaderWriter
package that cannot be extended; RuntimeDatCollection (the one
production implementation) exposes the aggregate stats directly and a
pattern match reads them, degrading to zero for any test double —
no new static registry was introduced (GpuMemoryTracker remains the
one precedented process-wide static).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(cherry picked from commit 1da2c33c875b41fa383dd79694ee2765f0e21896)
This commit is contained in:
Erik 2026-07-24 11:58:47 +02:00
parent bfc5e47365
commit 7b456b49d6
20 changed files with 477 additions and 9 deletions

View file

@ -1,6 +1,7 @@
using DatReaderWriter.DBObjs;
using DatReaderWriter.Lib.IO;
using System.Diagnostics.CodeAnalysis;
using System.Threading;
namespace AcDream.Content;
@ -42,6 +43,19 @@ internal sealed class BoundedDatObjectCache
private readonly object _gate = new();
private long _estimatedBytes;
// Hit/miss/eviction traffic counters (2026-07-24 measurement-tooling
// review). Read via Stats without the _gate lock, so plain fields with
// Interlocked increments rather than lock-protected longs.
private long _hits;
private long _misses;
private long _evictions;
/// <summary>Cumulative hit/miss/eviction counts since construction.</summary>
internal CacheStats Stats => new(
Interlocked.Read(ref _hits),
Interlocked.Read(ref _misses),
Interlocked.Read(ref _evictions));
internal BoundedDatObjectCache(
int entryLimit = DefaultEntryLimit,
long estimatedByteLimit = DefaultEstimatedByteLimit,
@ -80,10 +94,12 @@ internal sealed class BoundedDatObjectCache
{
if (!_byKey.TryGetValue(new CacheKey(typeof(T), fileId), out var node))
{
Interlocked.Increment(ref _misses);
value = default;
return false;
}
Interlocked.Increment(ref _hits);
MarkMostRecentlyUsed(node);
value = (T)node.Value.Value;
return true;
@ -146,6 +162,7 @@ internal sealed class BoundedDatObjectCache
_leastRecentlyUsed.RemoveFirst();
_byKey.Remove(node.Value.Key);
_estimatedBytes -= node.Value.EstimatedBytes;
Interlocked.Increment(ref _evictions);
}
private static long EstimateRetainedBytes(IDBObj value)

View file

@ -0,0 +1,18 @@
namespace AcDream.Content;
/// <summary>
/// Hit/miss/eviction counters for one bounded content cache. 2026-07-24
/// measurement-tooling review: the canonical checkpoint JSON tracked cache
/// residency (entry/byte counts) but never hit/miss traffic, so "does a
/// revisit portal hit or miss the caches" was unanswerable from the
/// artifact alone. Each cache accumulates its own counts via
/// <see cref="System.Threading.Interlocked"/> — mesh preparation runs on up
/// to four concurrent workers — and exposes them through this public,
/// cross-assembly-safe struct so <c>AcDream.App</c> diagnostics can read
/// them without taking the cache's internal lock.
/// </summary>
public readonly record struct CacheStats(long Hits, long Misses, long Evictions)
{
public static CacheStats operator +(CacheStats a, CacheStats b) =>
new(a.Hits + b.Hits, a.Misses + b.Misses, a.Evictions + b.Evictions);
}

View file

@ -56,6 +56,16 @@ public sealed class DatCollectionAdapter : IDatReaderWriter {
/// <summary>Source directory of the underlying DatCollection.</summary>
public string SourceDirectory => _dats.Options.DatDirectory ?? string.Empty;
/// <summary>
/// Aggregate hit/miss/eviction counts across the four bounded per-database
/// object caches (portal/cell/highRes/language). 2026-07-24
/// measurement-tooling review: answers "do revisit portals hit or miss
/// the caches" for the DAT-object layer specifically.
/// </summary>
public CacheStats ObjectCacheStats =>
_portal.ObjectCacheStats + _cell.ObjectCacheStats
+ _highRes.ObjectCacheStats + _language.ObjectCacheStats;
public IDatDatabase Portal => _portal;
public IDatDatabase Cell => _cell;
public ReadOnlyDictionary<uint, IDatDatabase> CellRegions => _cellRegions;
@ -173,6 +183,9 @@ public sealed class DatDatabaseWrapper : IDatDatabase {
public DatDatabase Db => _db;
public int Iteration => _db.Iteration?.CurrentIteration ?? 0;
/// <summary>This database's bounded DAT-object cache hit/miss/eviction counts (2026-07-24 measurement-tooling review).</summary>
public CacheStats ObjectCacheStats => _cache.Stats;
public IEnumerable<uint> GetAllIdsOfType<T>() where T : IDBObj =>
_db.GetAllIdsOfType<T>();

View file

@ -21,6 +21,19 @@ internal sealed class DecodedTextureCache {
private readonly int _maximumEntries;
private long _residentBytes;
// Hit/miss/eviction traffic counters (2026-07-24 measurement-tooling
// review). Read via Stats without the _gate lock, so plain fields with
// Interlocked increments rather than lock-protected longs.
private long _hits;
private long _misses;
private long _evictions;
/// <summary>Cumulative hit/miss/eviction counts since construction.</summary>
public CacheStats Stats => new(
Interlocked.Read(ref _hits),
Interlocked.Read(ref _misses),
Interlocked.Read(ref _evictions));
public DecodedTextureCache(long maximumBytes, int maximumEntries) {
ArgumentOutOfRangeException.ThrowIfNegative(maximumBytes);
ArgumentOutOfRangeException.ThrowIfNegative(maximumEntries);
@ -47,10 +60,12 @@ internal sealed class DecodedTextureCache {
public bool TryGet(DecodedTextureKey key, out byte[] pixels) {
lock (_gate) {
if (!_entries.TryGetValue(key, out var node)) {
Interlocked.Increment(ref _misses);
pixels = null!;
return false;
}
Interlocked.Increment(ref _hits);
Touch(node);
pixels = node.Value.Pixels;
return true;
@ -131,6 +146,7 @@ internal sealed class DecodedTextureCache {
_lru.Remove(node);
_entries.Remove(node.Value.Key);
_residentBytes -= node.Value.Pixels.LongLength;
Interlocked.Increment(ref _evictions);
}
}

View file

@ -49,6 +49,9 @@ public sealed class MeshExtractor {
// by up to MaxParallelLoads (4) decode workers.
private readonly System.Collections.Concurrent.ConcurrentDictionary<uint, byte[]> _solidColorTextureCache = new();
/// <summary>Decoded-texture cache hit/miss/eviction counts (2026-07-24 measurement-tooling review).</summary>
public CacheStats DecodedTextureCacheStats => _decodedTextureCache.Stats;
/// <summary>
/// MP1a mechanical seam: receives particle-preload meshes staged
/// mid-extraction (see <see cref="CollectEmittersFromScript"/>). The App