Add support converting non-inlined to inlined (#15444)

This is a feature to allow for inode compaction on
large clusters that use a lot of small files spread
across a large heirarchy.

.github/workflows/go-healing.yml

@@ -49,3 +49,4 @@ jobs:
           make verify-healing
           make verify-healing-inconsistent-versions
           make verify-healing-with-root-disks
+          make verify-healing-with-rewrite

Makefile

@@ -15,7 +15,7 @@ checks: ## check dependencies
 	@(env bash $(PWD)/buildscripts/checkdeps.sh)
 
 help: ## print this help
-	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' Makefile | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-30s\033[0m %s\n", $$1, $$2}'
+	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' Makefile | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-40s\033[0m %s\n", $$1, $$2}'
 
 getdeps: ## fetch necessary dependencies
 	@mkdir -p ${GOPATH}/bin
@@ -90,11 +90,16 @@ verify-healing: ## verify healing and replacing disks with minio binary
 	@(env bash $(PWD)/buildscripts/verify-healing.sh)
 	@(env bash $(PWD)/buildscripts/unaligned-healing.sh)
 
-verify-healing-with-root-disks:
+verify-healing-with-root-disks: ## verify healing root disks
 	@echo "Verify healing with root disks"
 	@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-with-root-disks.sh)
 
+verify-healing-with-rewrite: ## verify healing to rewrite old xl.meta -> new xl.meta
+	@echo "Verify healing with rewrite"
+	@GORACE=history_size=7 CGO_ENABLED=1 go build -race -tags kqueue -trimpath --ldflags "$(LDFLAGS)" -o $(PWD)/minio 1>/dev/null
+	@(env bash $(PWD)/buildscripts/rewrite-old-new.sh)
+
 verify-healing-inconsistent-versions: ## verify resolving inconsistent versions
 	@echo "Verify resolving inconsistent versions build with race"
 	@GORACE=history_size=7 CGO_ENABLED=1 go build -race -tags kqueue -trimpath --ldflags "$(LDFLAGS)" -o $(PWD)/minio 1>/dev/null

buildscripts/heal-manual.go

@@ -0,0 +1,87 @@
+//go:build ignore
+// +build ignore
+
+//
+// MinIO Object Storage (c) 2022 MinIO, Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"log"
+	"os"
+	"time"
+
+	"github.com/minio/madmin-go"
+)
+
+func main() {
+	// Note: YOUR-ACCESSKEYID, YOUR-SECRETACCESSKEY are
+	// dummy values, please replace them with original values.
+
+	// API requests are secure (HTTPS) if secure=true and insecure (HTTP) otherwise.
+	// New returns an MinIO Admin client object.
+	madmClnt, err := madmin.New(os.Args[1], os.Args[2], os.Args[3], false)
+	if err != nil {
+		log.Fatalln(err)
+	}
+
+	opts := madmin.HealOpts{
+		Recursive: true,                  // recursively heal all objects at 'prefix'
+		Remove:    true,                  // remove content that has lost quorum and not recoverable
+		Recreate:  true,                  // rewrite all old non-inlined xl.meta to new xl.meta
+		ScanMode:  madmin.HealNormalScan, // by default do not do 'deep' scanning
+	}
+
+	start, _, err := madmClnt.Heal(context.Background(), "healing-rewrite-bucket", "", opts, "", false, false)
+	if err != nil {
+		log.Fatalln(err)
+	}
+	fmt.Println("Healstart sequence ===")
+	enc := json.NewEncoder(os.Stdout)
+	if err = enc.Encode(&start); err != nil {
+		log.Fatalln(err)
+	}
+
+	fmt.Println()
+	for {
+		_, status, err := madmClnt.Heal(context.Background(), "healing-rewrite-bucket", "", opts, start.ClientToken, false, false)
+		if status.Summary == "finished" {
+			fmt.Println("Healstatus on items ===")
+			for _, item := range status.Items {
+				if err = enc.Encode(&item); err != nil {
+					log.Fatalln(err)
+				}
+			}
+			break
+		}
+		if status.Summary == "stopped" {
+			fmt.Println("Healstatus on items ===")
+			fmt.Println("Heal failed with", status.FailureDetail)
+			break
+		}
+
+		for _, item := range status.Items {
+			if err = enc.Encode(&item); err != nil {
+				log.Fatalln(err)
+			}
+		}
+
+		time.Sleep(time.Second)
+	}
+}

buildscripts/rewrite-old-new.sh

@@ -0,0 +1,151 @@
+#!/bin/bash -e
+
+set -E
+set -o pipefail
+set -x
+
+WORK_DIR="$PWD/.verify-$RANDOM"
+MINIO_CONFIG_DIR="$WORK_DIR/.minio"
+MINIO_OLD=( "$PWD/minio.RELEASE.2020-10-28T08-16-50Z" --config-dir "$MINIO_CONFIG_DIR" server )
+MINIO=( "$PWD/minio" --config-dir "$MINIO_CONFIG_DIR" server )
+
+if [ ! -x "$PWD/minio" ]; then
+    echo "minio executable binary not found in current directory"
+    exit 1
+fi
+
+function download_old_release() {
+    if [ ! -f minio.RELEASE.2020-10-28T08-16-50Z ]; then
+	curl --silent -O https://dl.minio.io/server/minio/release/linux-amd64/archive/minio.RELEASE.2020-10-28T08-16-50Z
+	chmod a+x minio.RELEASE.2020-10-28T08-16-50Z
+    fi
+}
+
+function verify_rewrite() {
+    start_port=$1
+
+    export MINIO_ACCESS_KEY=minio
+    export MINIO_SECRET_KEY=minio123
+    export MC_HOST_minio="http://minio:minio123@127.0.0.1:${start_port}/"
+    unset MINIO_KMS_AUTO_ENCRYPTION # do not auto-encrypt objects
+    export MINIO_CI_CD=1
+
+    MC_BUILD_DIR="mc-$RANDOM"
+    if ! git clone --quiet https://github.com/minio/mc "$MC_BUILD_DIR"; then
+	echo "failed to download https://github.com/minio/mc"
+	purge "${MC_BUILD_DIR}"
+	exit 1
+    fi
+
+    (cd "${MC_BUILD_DIR}" && go build -o "$WORK_DIR/mc")
+
+    # remove mc source.
+    purge "${MC_BUILD_DIR}"
+
+    "${MINIO_OLD[@]}" --address ":$start_port" "${WORK_DIR}/xl{1...16}" > "${WORK_DIR}/server1.log" 2>&1 &
+    pid=$!
+    disown $pid
+    sleep 10
+
+    if ! ps -p ${pid} 1>&2 >/dev/null; then
+	echo "server1 log:"
+	cat "${WORK_DIR}/server1.log"
+	echo "FAILED"
+	purge "$WORK_DIR"
+	exit 1
+    fi
+
+    "${WORK_DIR}/mc" mb minio/healing-rewrite-bucket --quiet --with-lock
+    "${WORK_DIR}/mc" cp \
+		     buildscripts/verify-build.sh \
+		     minio/healing-rewrite-bucket/ \
+		     --disable-multipart --quiet
+
+    "${WORK_DIR}/mc" cp \
+		     buildscripts/verify-build.sh \
+		     minio/healing-rewrite-bucket/ \
+		     --disable-multipart --quiet
+
+    "${WORK_DIR}/mc" cp \
+		     buildscripts/verify-build.sh \
+		     minio/healing-rewrite-bucket/ \
+		     --disable-multipart --quiet
+
+    kill ${pid}
+    sleep 3
+
+    "${MINIO[@]}" --address ":$start_port" "${WORK_DIR}/xl{1...16}" > "${WORK_DIR}/server1.log" 2>&1 &
+    pid=$!
+    disown $pid
+    sleep 10
+
+    if ! ps -p ${pid} 1>&2 >/dev/null; then
+	echo "server1 log:"
+	cat "${WORK_DIR}/server1.log"
+	echo "FAILED"
+	purge "$WORK_DIR"
+	exit 1
+    fi
+
+    go build ./docs/debugging/s3-check-md5/
+    if ! ./s3-check-md5 \
+	 -debug \
+	 -versions \
+	 -access-key minio \
+	 -secret-key minio123 \
+	 -endpoint http://127.0.0.1:${start_port}/ 2>&1 | grep INTACT; then
+	echo "server1 log:"
+	cat "${WORK_DIR}/server1.log"
+	echo "FAILED"
+	mkdir -p inspects
+	(cd inspects; "${WORK_DIR}/mc" admin inspect minio/healing-rewrite-bucket/verify-build.sh/**)
+
+	"${WORK_DIR}/mc" mb play/inspects
+	"${WORK_DIR}/mc" mirror inspects play/inspects
+
+	purge "$WORK_DIR"
+	exit 1
+    fi
+
+    go run ./buildscripts/heal-manual.go "127.0.0.1:${start_port}" "minio" "minio123"
+    sleep 1
+
+    if ! ./s3-check-md5 \
+	 -debug \
+	 -versions \
+	 -access-key minio \
+	 -secret-key minio123 \
+	 -endpoint http://127.0.0.1:${start_port}/ 2>&1 | grep INTACT; then
+	echo "server1 log:"
+	cat "${WORK_DIR}/server1.log"
+	echo "FAILED"
+	mkdir -p inspects
+	(cd inspects; "${WORK_DIR}/mc" admin inspect minio/healing-rewrite-bucket/verify-build.sh/**)
+
+	"${WORK_DIR}/mc" mb play/inspects
+	"${WORK_DIR}/mc" mirror inspects play/inspects
+
+	purge "$WORK_DIR"
+	exit 1
+    fi
+
+    kill ${pid}
+}
+
+function main() {
+    download_old_release
+
+    start_port=$(shuf -i 10000-65000 -n 1)
+
+    verify_rewrite ${start_port}
+}
+
+function purge()
+{
+    rm -rf "$1"
+}
+
+( main "$@" )
+rv=$?
+purge "$WORK_DIR"
+exit "$rv"

cmd/erasure-healing.go

@@ -243,7 +243,7 @@ func listAllBuckets(ctx context.Context, storageDisks []StorageAPI, healBuckets
 
 // Only heal on disks where we are sure that healing is needed. We can expand
 // this list as and when we figure out more errors can be added to this list safely.
-func shouldHealObjectOnDisk(erErr, dataErr error, meta FileInfo, latestMeta FileInfo) bool {
+func shouldHealObjectOnDisk(erErr, dataErr error, meta FileInfo, latestMeta FileInfo, doinline bool) bool {
 	switch {
 	case errors.Is(erErr, errFileNotFound) || errors.Is(erErr, errFileVersionNotFound):
 		return true
@@ -256,6 +256,10 @@ func shouldHealObjectOnDisk(erErr, dataErr error, meta FileInfo, latestMeta File
 			// always check first.
 			return true
 		}
+		if doinline {
+			// convert small files to 'inline'
+			return true
+		}
 		if !meta.Deleted && !meta.IsRemote() {
 			// If xl.meta was read fine but there may be problem with the part.N files.
 			if IsErr(dataErr, []error{
@@ -342,6 +346,26 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 	availableDisks, dataErrs, diskMTime := disksWithAllParts(ctx, onlineDisks, partsMetadata,
 		errs, latestMeta, bucket, object, scanMode)
 
+	var erasure Erasure
+	var recreate bool
+	if !latestMeta.Deleted && !latestMeta.IsRemote() {
+		// Initialize erasure coding
+		erasure, err = NewErasure(ctx, latestMeta.Erasure.DataBlocks,
+			latestMeta.Erasure.ParityBlocks, latestMeta.Erasure.BlockSize)
+		if err != nil {
+			return result, err
+		}
+
+		// Is only 'true' if the opts.Recreate is true and
+		// the object shardSize < smallFileThreshold do not
+		// set this to 'true' arbitrarily and must be only
+		// 'true' with caller ask.
+		recreate = (opts.Recreate &&
+			!latestMeta.InlineData() &&
+			len(latestMeta.Parts) == 1 &&
+			erasure.ShardFileSize(latestMeta.Parts[0].ActualSize) < smallFileThreshold)
+	}
+
 	// Loop to find number of disks with valid data, per-drive
 	// data state and a list of outdated disks on which data needs
 	// to be healed.
@@ -368,7 +392,7 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 			driveState = madmin.DriveStateCorrupt
 		}
 
-		if shouldHealObjectOnDisk(errs[i], dataErrs[i], partsMetadata[i], latestMeta) {
+		if shouldHealObjectOnDisk(errs[i], dataErrs[i], partsMetadata[i], latestMeta, recreate) {
 			outDatedDisks[i] = storageDisks[i]
 			disksToHealCount++
 			result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
@@ -422,7 +446,7 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 		return result, nil
 	}
 
-	if !latestMeta.XLV1 && !latestMeta.Deleted && disksToHealCount > latestMeta.Erasure.ParityBlocks {
+	if !latestMeta.XLV1 && !latestMeta.Deleted && !recreate && disksToHealCount > latestMeta.Erasure.ParityBlocks {
 		// When disk to heal count is greater than parity blocks we should simply error out.
 		err := fmt.Errorf("more disks are expected to heal than parity, returned errors: %v (dataErrs %v) -> %s/%s(%s)", errs, dataErrs, bucket, object, versionID)
 		logger.LogIf(ctx, err)
@@ -478,18 +502,9 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 	}
 
 	var inlineBuffers []*bytes.Buffer
-	if latestMeta.InlineData() {
-		inlineBuffers = make([]*bytes.Buffer, len(outDatedDisks))
-	}
-
 	if !latestMeta.Deleted && !latestMeta.IsRemote() {
-		// Heal each part. erasureHealFile() will write the healed
+		if latestMeta.InlineData() || recreate {
-		// part to .minio/tmp/uuid/ which needs to be renamed later to
+			inlineBuffers = make([]*bytes.Buffer, len(outDatedDisks))
-		// the final location.
-		erasure, err := NewErasure(ctx, latestMeta.Erasure.DataBlocks,
-			latestMeta.Erasure.ParityBlocks, latestMeta.Erasure.BlockSize)
-		if err != nil {
-			return result, err
 		}
 
 		erasureInfo := latestMeta.Erasure
@@ -525,6 +540,10 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 						tillOffset, DefaultBitrotAlgorithm, erasure.ShardSize())
 				}
 			}
+
+			// Heal each part. erasure.Heal() will write the healed
+			// part to .minio/tmp/uuid/ which needs to be renamed
+			// later to the final location.
 			err = erasure.Heal(ctx, writers, readers, partSize)
 			closeBitrotReaders(readers)
 			closeBitrotWriters(writers)
@@ -556,6 +575,7 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 				})
 				if len(inlineBuffers) > 0 && inlineBuffers[i] != nil {
 					partsMetadata[i].Data = inlineBuffers[i].Bytes()
+					partsMetadata[i].SetInlineData()
 				} else {
 					partsMetadata[i].Data = nil
 				}
@@ -587,8 +607,9 @@ func (er erasureObjects) healObject(ctx context.Context, bucket string, object s
 			return result, err
 		}
 
-		// Remove any remaining parts from outdated disks from before transition.
+		// - Remove any parts from healed disks after its been inlined.
-		if partsMetadata[i].IsRemote() {
+		// - Remove any remaining parts from outdated disks from before transition.
+		if recreate || partsMetadata[i].IsRemote() {
 			rmDataDir := partsMetadata[i].DataDir
 			disk.DeleteVol(ctx, pathJoin(bucket, encodeDirObject(object), rmDataDir), true)
 		}

cmd/erasure-server-pool.go

@@ -1769,7 +1769,9 @@ func (z *erasureServerPools) HealBucket(ctx context.Context, bucket string, opts
 	}
 
 	// Attempt heal on the bucket metadata, ignore any failures
-	defer z.HealObject(ctx, minioMetaBucket, pathJoin(bucketMetaPrefix, bucket, bucketMetadataFile), "", opts)
+	hopts := opts
+	hopts.Recreate = false
+	defer z.HealObject(ctx, minioMetaBucket, pathJoin(bucketMetaPrefix, bucket, bucketMetadataFile), "", hopts)
 
 	for _, pool := range z.serverPools {
 		result, err := pool.HealBucket(ctx, bucket, opts)