From 03f08942faaeccf930a9727c359ea2ea34ed3ddd Mon Sep 17 00:00:00 2001
From: duanmeng <duanmeng@outlook.com>
Date: Wed, 4 Dec 2024 11:38:10 -0800
Subject: [PATCH] fix: Flaky HashJoinReplayerTest.partialDriverIds (#11742)

Summary:
The trace data file for driver 0 might be empty. In such a case, the replayer
would disregard the associated re-execution, thereby not triggering the injected
failure. Add a file size check before the exception asserts in `HashJoinReplayerTest.partialDriverIds`.

Fix https://github.com/facebookincubator/velox/issues/11695

Pull Request resolved: https://github.com/facebookincubator/velox/pull/11742

Reviewed By: Yuhta

Differential Revision: D66767559

Pulled By: xiaoxmeng

fbshipit-source-id: 75df35ca2a27f13f62831bceccd88415e7826e6e
---
 .../tool/trace/tests/HashJoinReplayerTest.cpp | 25 +++++++++++--------
 1 file changed, 14 insertions(+), 11 deletions(-)

diff --git a/velox/tool/trace/tests/HashJoinReplayerTest.cpp b/velox/tool/trace/tests/HashJoinReplayerTest.cpp
index 53567d48f027..dcd840d217c0 100644
--- a/velox/tool/trace/tests/HashJoinReplayerTest.cpp
+++ b/velox/tool/trace/tests/HashJoinReplayerTest.cpp
@@ -289,22 +289,25 @@ TEST_F(HashJoinReplayerTest, partialDriverIds) {
   const auto opTraceDataFile = exec::trace::getOpTraceInputFilePath(opTraceDir);
   auto faultyFs = faultyFileSystem();
   faultyFs->setFileInjectionHook([&](FaultFileOperation* op) {
-    if (op->type == FaultFileOperation::Type::kRead &&
+    if ((op->type == FaultFileOperation::Type::kRead ||
+         op->type == FaultFileOperation::Type::kReadv) &&
         op->path == opTraceDataFile) {
       VELOX_FAIL("Read wrong data file {}", opTraceDataFile);
     }
   });
 
-  VELOX_ASSERT_THROW(
-      HashJoinReplayer(
-          traceRoot,
-          task->queryCtx()->queryId(),
-          task->taskId(),
-          traceNodeId_,
-          "HashJoin",
-          "0")
-          .run(),
-      "Read wrong data file");
+  if (faultyFs->openFileForRead(opTraceDataFile)->size() > 0) {
+    VELOX_ASSERT_THROW(
+        HashJoinReplayer(
+            traceRoot,
+            task->queryCtx()->queryId(),
+            task->taskId(),
+            traceNodeId_,
+            "HashJoin",
+            "0")
+            .run(),
+        "Read wrong data file");
+  }
   HashJoinReplayer(
       traceRoot,
       task->queryCtx()->queryId(),