From 9246d01f719d7986b8f659ef43f37bd0ac668af6 Mon Sep 17 00:00:00 2001
From: "google-labs-jules[bot]"
 <161369871+google-labs-jules[bot]@users.noreply.github.com>
Date: Sun, 5 Apr 2026 05:13:58 +0000
Subject: [PATCH] Optimize ProcessMemoryScanner concurrency with sliding window

In `scanPIDs`, replace static array chunking with a continuous iterator-based
sliding window inside `withTaskGroup`.

This avoids tail-latency waiting where the system idles waiting for the slowest
task in a chunk to finish before starting the next chunk, maximizing throughput
when scanning hundreds of PIDs concurrently.

Co-authored-by: acebytes <2820910+acebytes@users.noreply.github.com>
---
 .jules/bolt.md                                |  3 ++
 .../Memory/ProcessMemoryScanner.swift         | 38 +++++++++++--------
 2 files changed, 25 insertions(+), 16 deletions(-)
 create mode 100644 .jules/bolt.md

diff --git a/.jules/bolt.md b/.jules/bolt.md
new file mode 100644
index 0000000..c7dc04d
--- /dev/null
+++ b/.jules/bolt.md
@@ -0,0 +1,3 @@
+## 2026-04-05 - Optimize ProcessMemoryScanner concurrency with sliding window
+**Learning:** In Swift structured concurrency, when processing high-volume tasks using `withTaskGroup`, static chunking limits throughput due to tail latency (waiting for the slowest task in a chunk). A sliding window with an iterator maintains maximum concurrent execution limits continuously.
+**Action:** Always use an iterator-based sliding window in `withTaskGroup` instead of static chunking for processing large collections to avoid tail latency bottlenecks.
diff --git a/Sources/Cacheout/Memory/ProcessMemoryScanner.swift b/Sources/Cacheout/Memory/ProcessMemoryScanner.swift
index 3f8e728..b653d4a 100644
--- a/Sources/Cacheout/Memory/ProcessMemoryScanner.swift
+++ b/Sources/Cacheout/Memory/ProcessMemoryScanner.swift
@@ -97,29 +97,35 @@ actor ProcessMemoryScanner {
     ///
     /// Returns the collected entries and the count of EPERM failures.
     private func scanPIDs(_ pids: [pid_t]) async -> (entries: [ProcessEntryDTO], epermCount: Int) {
-        // Chunk PIDs to cap concurrency at maxConcurrency.
-        let chunks = stride(from: 0, to: pids.count, by: maxConcurrency).map {
-            Array(pids[$0..<min($0 + maxConcurrency, pids.count)])
-        }
-
         var allEntries: [ProcessEntryDTO] = []
         var totalEperm = 0
 
-        for chunk in chunks {
-            await withTaskGroup(of: ScanPIDResult.self) { group in
-                for pid in chunk {
+        // Use a sliding window approach for maximum continuous concurrency.
+        // This avoids tail-latency waiting that occurs with static chunking.
+        await withTaskGroup(of: ScanPIDResult.self) { group in
+            var iterator = pids.makeIterator()
+
+            for _ in 0..<maxConcurrency {
+                if let pid = iterator.next() {
                     group.addTask { [self] in
                         self.scanSinglePID(pid)
                     }
                 }
-                for await result in group {
-                    switch result {
-                    case .success(let entry):
-                        allEntries.append(entry)
-                    case .eperm:
-                        totalEperm += 1
-                    case .otherError:
-                        break
+            }
+
+            for await result in group {
+                switch result {
+                case .success(let entry):
+                    allEntries.append(entry)
+                case .eperm:
+                    totalEperm += 1
+                case .otherError:
+                    break
+                }
+
+                if let nextPid = iterator.next() {
+                    group.addTask { [self] in
+                        self.scanSinglePID(nextPid)
                     }
                 }
             }