heal: Fix regression in healing a new fresh drive (#19615)

Makefile

@@ -131,6 +131,7 @@ verify-healing: ## verify healing and replacing disks with minio binary
 	@echo "Verify healing build with race"
 	@GORACE=history_size=7 CGO_ENABLED=1 go build -race -tags kqueue -trimpath --ldflags "$(LDFLAGS)" -o $(PWD)/minio 1>/dev/null
 	@(env bash $(PWD)/buildscripts/verify-healing.sh)
+	@(env bash $(PWD)/buildscripts/verify-healing-empty-erasure-set.sh)
 	@(env bash $(PWD)/buildscripts/heal-inconsistent-versions.sh)
 
 verify-healing-with-root-disks: ## verify healing root disks

buildscripts/verify-healing-empty-erasure-set.sh

@@ -0,0 +1,140 @@
+#!/bin/bash -e
+#
+
+set -E
+set -o pipefail
+
+if [ ! -x "$PWD/minio" ]; then
+	echo "minio executable binary not found in current directory"
+	exit 1
+fi
+
+WORK_DIR="$PWD/.verify-$RANDOM"
+MINIO_CONFIG_DIR="$WORK_DIR/.minio"
+MINIO=("$PWD/minio" --config-dir "$MINIO_CONFIG_DIR" server)
+
+function start_minio_3_node() {
+	export MINIO_ROOT_USER=minio
+	export MINIO_ROOT_PASSWORD=minio123
+	export MINIO_ERASURE_SET_DRIVE_COUNT=6
+	export MINIO_CI_CD=1
+
+	start_port=$2
+	args=""
+	for i in $(seq 1 3); do
+		args="$args http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/1/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/2/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/3/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/4/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/5/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/6/"
+	done
+
+	"${MINIO[@]}" --address ":$((start_port + 1))" $args >"${WORK_DIR}/dist-minio-server1.log" 2>&1 &
+	pid1=$!
+	disown ${pid1}
+
+	"${MINIO[@]}" --address ":$((start_port + 2))" $args >"${WORK_DIR}/dist-minio-server2.log" 2>&1 &
+	pid2=$!
+	disown $pid2
+
+	"${MINIO[@]}" --address ":$((start_port + 3))" $args >"${WORK_DIR}/dist-minio-server3.log" 2>&1 &
+	pid3=$!
+	disown $pid3
+
+	sleep "$1"
+
+	if ! ps -p $pid1 1>&2 >/dev/null; then
+		echo "server1 log:"
+		cat "${WORK_DIR}/dist-minio-server1.log"
+		echo "FAILED"
+		purge "$WORK_DIR"
+		exit 1
+	fi
+
+	if ! ps -p $pid2 1>&2 >/dev/null; then
+		echo "server2 log:"
+		cat "${WORK_DIR}/dist-minio-server2.log"
+		echo "FAILED"
+		purge "$WORK_DIR"
+		exit 1
+	fi
+
+	if ! ps -p $pid3 1>&2 >/dev/null; then
+		echo "server3 log:"
+		cat "${WORK_DIR}/dist-minio-server3.log"
+		echo "FAILED"
+		purge "$WORK_DIR"
+		exit 1
+	fi
+
+	if ! pkill minio; then
+		for i in $(seq 1 3); do
+			echo "server$i log:"
+			cat "${WORK_DIR}/dist-minio-server$i.log"
+		done
+		echo "FAILED"
+		purge "$WORK_DIR"
+		exit 1
+	fi
+
+	sleep 1
+	if pgrep minio; then
+		# forcibly killing, to proceed further properly.
+		if ! pkill -9 minio; then
+			echo "no minio process running anymore, proceed."
+		fi
+	fi
+}
+
+function check_online() {
+	if ! grep -q 'Status:' ${WORK_DIR}/dist-minio-*.log; then
+		echo "1"
+	fi
+}
+
+function purge() {
+	rm -rf "$1"
+}
+
+function __init__() {
+	echo "Initializing environment"
+	mkdir -p "$WORK_DIR"
+	mkdir -p "$MINIO_CONFIG_DIR"
+
+	## version is purposefully set to '3' for minio to migrate configuration file
+	echo '{"version": "3", "credential": {"accessKey": "minio", "secretKey": "minio123"}, "region": "us-east-1"}' >"$MINIO_CONFIG_DIR/config.json"
+}
+
+function perform_test() {
+	start_minio_3_node 120 $2
+
+	echo "Testing Distributed Erasure setup healing of drives"
+	echo "Remove the contents of the disks belonging to '${1}' erasure set"
+
+	rm -rf ${WORK_DIR}/${1}/*/
+
+	set -x
+	start_minio_3_node 120 $2
+
+	rv=$(check_online)
+	if [ "$rv" == "1" ]; then
+		for i in $(seq 1 3); do
+			echo "server$i log:"
+			cat "${WORK_DIR}/dist-minio-server$i.log"
+		done
+		pkill -9 minio
+		echo "FAILED"
+		purge "$WORK_DIR"
+		exit 1
+	fi
+}
+
+function main() {
+	# use same ports for all tests
+	start_port=$(shuf -i 10000-65000 -n 1)
+
+	perform_test "2" ${start_port}
+	perform_test "1" ${start_port}
+	perform_test "3" ${start_port}
+}
+
+(__init__ "$@" && main "$@")
+rv=$?
+purge "$WORK_DIR"
+exit "$rv"

buildscripts/verify-healing.sh

@@ -12,6 +12,7 @@ fi
 WORK_DIR="$PWD/.verify-$RANDOM"
 MINIO_CONFIG_DIR="$WORK_DIR/.minio"
 MINIO=("$PWD/minio" --config-dir "$MINIO_CONFIG_DIR" server)
+GOPATH=/tmp/gopath
 
 function start_minio_3_node() {
 	export MINIO_ROOT_USER=minio
@@ -19,10 +20,14 @@ function start_minio_3_node() {
 	export MINIO_ERASURE_SET_DRIVE_COUNT=6
 	export MINIO_CI_CD=1
 
+	first_time=$(find ${WORK_DIR}/ | grep format.json | wc -l)
+
 	start_port=$2
 	args=""
-	for i in $(seq 1 3); do
+	for d in $(seq 1 3 5); do
-		args="$args http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/1/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/2/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/3/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/4/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/5/ http://127.0.0.1:$((start_port + i))${WORK_DIR}/$i/6/"
+		args="$args http://127.0.0.1:$((start_port + 1))${WORK_DIR}/1/${d}/ http://127.0.0.1:$((start_port + 2))${WORK_DIR}/2/${d}/ http://127.0.0.1:$((start_port + 3))${WORK_DIR}/3/${d}/ "
+		d=$((d + 1))
+		args="$args http://127.0.0.1:$((start_port + 1))${WORK_DIR}/1/${d}/ http://127.0.0.1:$((start_port + 2))${WORK_DIR}/2/${d}/ http://127.0.0.1:$((start_port + 3))${WORK_DIR}/3/${d}/ "
 	done
 
 	"${MINIO[@]}" --address ":$((start_port + 1))" $args >"${WORK_DIR}/dist-minio-server1.log" 2>&1 &
@@ -39,6 +44,8 @@ function start_minio_3_node() {
 
 	sleep "$1"
 
+	[ ${first_time} -eq 0 ] && upload_objects $start_port
+
 	if ! ps -p $pid1 1>&2 >/dev/null; then
 		echo "server1 log:"
 		cat "${WORK_DIR}/dist-minio-server1.log"
@@ -82,10 +89,23 @@ function start_minio_3_node() {
 	fi
 }
 
-function check_online() {
+function check_heal() {
 	if ! grep -q 'Status:' ${WORK_DIR}/dist-minio-*.log; then
-		echo "1"
+		return 1
 	fi
+
+	for ((i = 0; i < 20; i++)); do
+		test -f ${WORK_DIR}/$1/1/.minio.sys/format.json
+		v1=$?
+		nextInES=$(($1 + 1)) && [ $nextInES -gt 3 ] && nextInES=1
+		foundFiles1=$(find ${WORK_DIR}/$1/1/ | grep -v .minio.sys | grep xl.meta | wc -l)
+		foundFiles2=$(find ${WORK_DIR}/$nextInES/1/ | grep -v .minio.sys | grep xl.meta | wc -l)
+		test $foundFiles1 -eq $foundFiles2
+		v2=$?
+		[ $v1 == 0 -a $v2 == 0 ] && return 0
+		sleep 10
+	done
+	return 1
 }
 
 function purge() {
@@ -99,20 +119,39 @@ function __init__() {
 
 	## version is purposefully set to '3' for minio to migrate configuration file
 	echo '{"version": "3", "credential": {"accessKey": "minio", "secretKey": "minio123"}, "region": "us-east-1"}' >"$MINIO_CONFIG_DIR/config.json"
+
+	if [ ! -f /tmp/mc ]; then
+		wget --quiet -O /tmp/mc https://dl.minio.io/client/mc/release/linux-amd64/mc &&
+			chmod +x /tmp/mc
+	fi
+}
+
+function upload_objects() {
+	start_port=$1
+
+	/tmp/mc alias set myminio http://127.0.0.1:$((start_port + 1)) minio minio123 --api=s3v4
+	/tmp/mc ready myminio
+	/tmp/mc mb myminio/testbucket/
+	for ((i = 0; i < 20; i++)); do
+		echo "my content" | /tmp/mc pipe myminio/testbucket/file-$i
+	done
 }
 
 function perform_test() {
-	start_minio_3_node 120 $2
+	start_port=$2
+
+	start_minio_3_node 120 $start_port
 
 	echo "Testing Distributed Erasure setup healing of drives"
-	echo "Remove the contents of the disks belonging to '${1}' erasure set"
+	echo "Remove the contents of the disks belonging to '${1}' node"
 
 	rm -rf ${WORK_DIR}/${1}/*/
 
 	set -x
-	start_minio_3_node 120 $2
+	start_minio_3_node 120 $start_port
 
-	rv=$(check_online)
+	check_heal ${1}
+	rv=$?
 	if [ "$rv" == "1" ]; then
 		for i in $(seq 1 3); do
 			echo "server$i log:"

cmd/global-heal.go

@@ -28,6 +28,10 @@ import (
 
 	"github.com/dustin/go-humanize"
 	"github.com/minio/madmin-go/v3"
+	"github.com/minio/minio/internal/bucket/lifecycle"
+	objectlock "github.com/minio/minio/internal/bucket/object/lock"
+	"github.com/minio/minio/internal/bucket/replication"
+	"github.com/minio/minio/internal/bucket/versioning"
 	"github.com/minio/minio/internal/color"
 	"github.com/minio/minio/internal/config/storageclass"
 	xioutil "github.com/minio/minio/internal/ioutil"
@@ -214,34 +218,40 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 			continue
 		}
 
-		vc, err := globalBucketVersioningSys.Get(bucket)
+		var (
-		if err != nil {
+			vc   *versioning.Versioning
-			retErr = err
+			lc   *lifecycle.Lifecycle
-			healingLogIf(ctx, err)
+			lr   objectlock.Retention
-			continue
+			rcfg *replication.Config
-		}
+		)
 
-		// Check if the current bucket has a configured lifecycle policy
+		if !isMinioMetaBucketName(bucket) {
-		lc, err := globalLifecycleSys.Get(bucket)
+			vc, err = globalBucketVersioningSys.Get(bucket)
-		if err != nil && !errors.Is(err, BucketLifecycleNotFound{Bucket: bucket}) {
+			if err != nil {
-			retErr = err
+				retErr = err
-			healingLogIf(ctx, err)
+				healingLogIf(ctx, err)
-			continue
+				continue
-		}
+			}
-
+			// Check if the current bucket has a configured lifecycle policy
-		// Check if bucket is object locked.
+			lc, err = globalLifecycleSys.Get(bucket)
-		lr, err := globalBucketObjectLockSys.Get(bucket)
+			if err != nil && !errors.Is(err, BucketLifecycleNotFound{Bucket: bucket}) {
-		if err != nil {
+				retErr = err
-			retErr = err
+				healingLogIf(ctx, err)
-			healingLogIf(ctx, err)
+				continue
-			continue
+			}
-		}
+			// Check if bucket is object locked.
-
+			lr, err = globalBucketObjectLockSys.Get(bucket)
-		rcfg, err := getReplicationConfig(ctx, bucket)
+			if err != nil {
-		if err != nil {
+				retErr = err
-			retErr = err
+				healingLogIf(ctx, err)
-			healingLogIf(ctx, err)
+				continue
-			continue
+			}
+			rcfg, err = getReplicationConfig(ctx, bucket)
+			if err != nil {
+				retErr = err
+				healingLogIf(ctx, err)
+				continue
+			}
 		}
 
 		if serverDebugLog {