From 38fad0c145748518fdbeba244a89f5e0f346c2fe Mon Sep 17 00:00:00 2001
From: Ed Savage <ed.savage@elastic.co>
Date: Thu, 2 Apr 2026 11:12:22 +1300
Subject: [PATCH] [ML] Fix flaky CIoManagerTest/testFileIoGood on linux-x86_64

The reader hits a premature EOF at 8192 bytes (two filesystem pages)
while the remaining ~1.8KB of data hasn't been flushed from the
kernel buffer yet. With MAX_EOF_RETRIES=10 and 40ms sleeps (400ms
total), the reader gives up before the flush completes on loaded CI
agents.

Increase MAX_EOF_RETRIES from 10 to 50 (2 seconds total patience)
which should be ample time for the kernel to flush a few KB of data.

Fixes #2890

Made-with: Cursor
---
 lib/test/CThreadDataReader.cc | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)

diff --git a/lib/test/CThreadDataReader.cc b/lib/test/CThreadDataReader.cc
index d9851207b..24b2f7096 100644
--- a/lib/test/CThreadDataReader.cc
+++ b/lib/test/CThreadDataReader.cc
@@ -65,7 +65,22 @@ void CThreadDataReader::run() {
 
     static const std::streamsize BUF_SIZE{512};
     char buffer[BUF_SIZE];
-    while (strm.good()) {
+
+    // For regular files the reader can open the file while the writer
+    // is still flushing, hit a premature EOF, and stop too early.
+    // After hitting EOF we clear the stream state and retry a limited
+    // number of times, sleeping between each attempt.  Any successful
+    // read resets the counter so we only give up after the writer has
+    // truly finished.
+    //
+    // 50 retries * 40ms sleep = 2 seconds total patience after the last
+    // successful read.  The previous value of 10 (~400ms) was too low
+    // for loaded CI agents where kernel buffer flushing can be delayed.
+    // See https://github.com/elastic/ml-cpp/issues/2890.
+    static const std::size_t MAX_EOF_RETRIES{50};
+    std::size_t eofRetries{0};
+
+    for (;;) {
         if (m_Shutdown) {
             return;
         }
@@ -75,6 +90,7 @@ void CThreadDataReader::run() {
             return;
         }
         if (strm.gcount() > 0) {
+            eofRetries = 0;
             core::CScopedLock lock(m_Mutex);
             // This code deals with the test character we write to
             // detect the short-lived connection problem on Windows
@@ -88,6 +104,16 @@ void CThreadDataReader::run() {
                 m_Data.append(copyFrom, copyLen);
             }
         }
+        if (strm.eof()) {
+            if (strm.gcount() == 0) {
+                ++eofRetries;
+                if (eofRetries > MAX_EOF_RETRIES) {
+                    break;
+                }
+                std::this_thread::sleep_for(std::chrono::milliseconds(m_SleepTimeMs));
+            }
+            strm.clear();
+        }
     }
 }