From ba007d9373797565802638132cac5b54f8e0f8a6 Mon Sep 17 00:00:00 2001
From: yyforyongyu <yy2452@columbia.edu>
Date: Sun, 20 Aug 2023 11:50:16 +0800
Subject: [PATCH] itest: fix itest logs upload

When one of the itest tranches fails, because we are running tests in
parallel, other tranches will still be running, which caused 7z to fail
at zipping the logs because race reads. This commit fixes it by making
sure we are waiting other tranches to finish before moving to zipping
and uploading the logs.
---
 Makefile                  |  2 +-
 scripts/itest_parallel.sh | 38 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+), 1 deletion(-)
 create mode 100755 scripts/itest_parallel.sh

diff --git a/Makefile b/Makefile
index e31c0adde..a6ca526ba 100644
--- a/Makefile
+++ b/Makefile
@@ -174,7 +174,7 @@ itest-race: build-itest-race itest-only
 itest-parallel: build-itest db-instance
 	@$(call print, "Running tests")
 	rm -rf itest/*.log itest/.logs-*; date
-	EXEC_SUFFIX=$(EXEC_SUFFIX) echo "$$(seq 0 $$(expr $(ITEST_PARALLELISM) - 1))" | xargs -P $(ITEST_PARALLELISM) -n 1 -I {} scripts/itest_part.sh {} $(NUM_ITEST_TRANCHES) $(TEST_FLAGS) $(ITEST_FLAGS)
+	EXEC_SUFFIX=$(EXEC_SUFFIX) scripts/itest_parallel.sh $(ITEST_PARALLELISM) $(NUM_ITEST_TRANCHES) $(TEST_FLAGS) $(ITEST_FLAGS)
 
 itest-clean:
 	@$(call print, "Cleaning old itest processes")
diff --git a/scripts/itest_parallel.sh b/scripts/itest_parallel.sh
new file mode 100755
index 000000000..e540a02ec
--- /dev/null
+++ b/scripts/itest_parallel.sh
@@ -0,0 +1,38 @@
+#!/bin/bash
+
+# Get all the variables.
+PROCESSES=$1
+TRANCHES=$2
+TEST_FLAGS=$3
+ITEST_FLAGS=$4
+
+# Create a variable to hold the final exit code.
+exit_code=0
+
+# Run commands using xargs in parallel and capture their PIDs
+pids=()
+for ((i=0; i<PROCESSES; i++)); do 
+    scripts/itest_part.sh $i $TRANCHES $TEST_FLAGS $ITEST_FLAGS &
+    pids+=($!)
+done
+
+
+# Wait for the processes created by xargs to finish.
+for pid in "${pids[@]}"; do
+    wait $pid
+
+    # Once finished, grab its exit code.
+    current_exit_code=$?
+
+    # Overwrite the exit code if current itest doesn't return 0.
+    if [ $current_exit_code -ne 0 ]; then
+	# Only write the exit code of the first failing itest.
+	if [ $exit_code -eq 0 ]; then
+            exit_code=$current_exit_code
+	fi
+    fi
+done
+
+
+# Exit with the exit code of the first failing itest or 0.
+exit $exit_code