From c42155dc7d5dfd68b0dd2e27937fec90c6f804fa Mon Sep 17 00:00:00 2001
From: Grzegorz Nosek <grzegorz.nosek@sysdig.com>
Date: Tue, 4 Aug 2020 19:02:12 +0200
Subject: [PATCH] Make probe builder failures non-fatal

The probe builder still returns non-zero exit code, but only after
trying to build all probes. Results from all failed builds are logged
again at the end of the output for easier investigation.
---
 probe-builder/build-probe-binaries | 76 ++++++++++++++++++++++--------
 1 file changed, 56 insertions(+), 20 deletions(-)

diff --git a/probe-builder/build-probe-binaries b/probe-builder/build-probe-binaries
index e0c6389b64..c10998b8c4 100755
--- a/probe-builder/build-probe-binaries
+++ b/probe-builder/build-probe-binaries
@@ -34,6 +34,8 @@ RETRIES=10
 DOWNLOAD_TIMEOUT=300
 PROBE_VERSION=
 RUNNING_IN_DOCKER=
+FAILED=
+FAIL_LOG=$(mktemp /tmp/fail.log.XXXXXX)
 
 usage() {
 	cat >&2 <<EOF
@@ -273,19 +275,7 @@ function run_toolkit {
 }
 
 declare -A builders
-function build_probe {
-	# Skip Kernel 4.15.0-29 because probe does not build against it
-	if [ $KERNEL_RELEASE-$HASH = "4.15.0-29-generic-ea0aa038a6b9bdc4bb42152682bba6ce"  ]; then
-		echo "Temporarily skipping " $PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH.ko
-		return
-	fi
-
-	if [ -f $BASEDIR/output/$PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH.ko ] &&
-	   [ -f $BASEDIR/output/$PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH_ORIG.ko ]; then
-		echo "Skipping $PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH.ko (already built)"
-		return
-	fi
-
+function build_probe_impl {
 	PROBE_ROOT=${PROBE_ROOT:-/build/probe}
 	if [ -n "${DOCKER_BUILDER:-}" ]
 	then
@@ -309,17 +299,23 @@ function build_probe {
 		then
 			if [ -e $KERNELDIR/include/generated/compile.h ]
 			then
-				BUILDER_GCC_VERSION=$(grep -Po '(?<=^#define LINUX_COMPILER "gcc version )[0-9.]+' $KERNELDIR/include/generated/compile.h)
+				BUILDER_GCC_VERSION=$(grep -Po '(?<=^#define LINUX_COMPILER "gcc version )[0-9.]+' $KERNELDIR/include/generated/compile.h || true)
 			elif [ -e $KERNELDIR/include/linux/compile.h ]
 			then
 				# RHEL6
-				BUILDER_GCC_VERSION=$(grep -Po '(?<=^#define LINUX_COMPILER "gcc version )[0-9.]+' $KERNELDIR/include/linux/compile.h)
+				BUILDER_GCC_VERSION=$(grep -Po '(?<=^#define LINUX_COMPILER "gcc version )[0-9.]+' $KERNELDIR/include/linux/compile.h || true)
 			else
 				# ancient Ubuntu gets and ancient compiler
 				BUILDER_GCC_VERSION=4.8.0
 			fi
 		fi
 
+		if [ -z "$BUILDER_GCC_VERSION" ]
+		then
+			echo "Failed to find gcc version for $KERNELDIR"
+			return 1
+		fi
+
 		# We don't really care about the compiler patch levels, only the major/minor version
 		BUILDER_GCC_MINOR_VERSION="${BUILDER_GCC_VERSION%.*}"
 
@@ -380,15 +376,15 @@ function build_probe {
 	IMAGE_NAME="sysdig-probe-builder:${DOCKERFILE_TAG}"
 	CONTAINER_NAME="sysdig-probe-builder-${DOCKERFILE_TAG}"
 
-	docker ps -q -f "name=$CONTAINER_NAME" | xargs --no-run-if-empty docker rm
+	docker ps -q -f "name=$CONTAINER_NAME" | xargs --no-run-if-empty docker rm || return 1
 	if [ -n "$BUILDER_IMAGE_PREFIX" ]
 	then
 		IMAGE_NAME="${BUILDER_IMAGE_PREFIX}${IMAGE_NAME}"
 	else
 		if [ -z "${builders[$IMAGE_NAME]:-}" ]
 		then
-			docker build -t "$IMAGE_NAME" -f $DOCKERFILE --pull $BUILDER_SOURCE
-			docker images -q -f 'dangling=true' | xargs --no-run-if-empty docker rmi
+			docker build -t "$IMAGE_NAME" -f $DOCKERFILE --pull $BUILDER_SOURCE || return 1
+			docker images -q -f 'dangling=true' | xargs --no-run-if-empty docker rmi || true
 			builders[$IMAGE_NAME]=1
 		fi
 	fi
@@ -403,7 +399,34 @@ function build_probe {
 		-e HASH=$HASH \
 		-e HASH_ORIG=$HASH_ORIG \
 		--name $CONTAINER_NAME \
-		$IMAGE_NAME
+		$IMAGE_NAME || return 1
+}
+
+function build_probe {
+	PROBE_ID=$PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH.ko
+
+	# Skip Kernel 4.15.0-29 because probe does not build against it
+	if [ $KERNEL_RELEASE-$HASH = "4.15.0-29-generic-ea0aa038a6b9bdc4bb42152682bba6ce"  ]; then
+		echo "Temporarily skipping $PROBE_ID"
+		return
+	fi
+
+	if [ -f $BASEDIR/output/$PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH.ko ] &&
+	   [ -f $BASEDIR/output/$PROBE_NAME-$PROBE_VERSION-$ARCH-$KERNEL_RELEASE-$HASH_ORIG.ko ]; then
+		echo "Skipping $PROBE_ID (already built)"
+		return
+	fi
+
+    LOG=$(mktemp /tmp/log.XXXXXX)
+    if ! build_probe_impl &> $LOG
+    then
+        (echo "Build for $PROBE_ID failed"; cat $LOG) | tee -a $FAIL_LOG
+        FAILED=1
+    else
+		echo "Build for $PROBE_ID successful"
+		cat $LOG
+    fi
+    rm -f $LOG
 }
 
 function coreos_build {
@@ -990,6 +1013,19 @@ case "$KERNEL_TYPE" in
 		;;
 esac
 
-echo "Success."
+if [ -s "$FAIL_LOG" ]
+then
+	echo "Failed builds:"
+	echo "------------------------------"
+	cat $FAIL_LOG
+fi
+rm -f $FAIL_LOG
+if [ -n "$FAILED" ]
+then
+	echo "Build failed."
+	exit 1
+else
+	echo "Success."
+fi
 
 # vim: :set tabstop=8 shiftwidth=8 noexpandtab: