1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
5 package deduplicationreport
13 "git.arvados.org/arvados.git/sdk/go/arvados"
14 "git.arvados.org/arvados.git/sdk/go/arvadosclient"
15 "git.arvados.org/arvados.git/sdk/go/manifest"
17 "github.com/dustin/go-humanize"
18 "github.com/sirupsen/logrus"
21 func deDuplicate(inputs []string) (trimmed []string) {
22 seen := make(map[string]bool)
23 for _, uuid := range inputs {
26 trimmed = append(trimmed, uuid)
32 func parseFlags(prog string, args []string, logger *logrus.Logger, stderr io.Writer) ([]string, error) {
33 flags := flag.NewFlagSet("", flag.ContinueOnError)
34 flags.SetOutput(stderr)
35 flags.Usage = func() {
36 fmt.Fprintf(flags.Output(), `
38 %s [options ...] <collection-uuid> <collection-uuid> ...
40 %s [options ...] <collection-pdh>,<collection_uuid> \
41 <collection-pdh>,<collection_uuid> ...
43 This program analyzes the overlap in blocks used by 2 or more collections. It
44 prints a deduplication report that shows the nominal space used by the
45 collections, as well as the actual size and the amount of space that is saved
46 by Keep's deduplication.
48 The list of collections may be provided in two ways. A list of collection
49 uuids is sufficient. Alternatively, the PDH for each collection may also be
50 provided. This is will greatly speed up operation when the list contains
51 multiple collections with the same PDH.
53 Exit status will be zero if there were no errors generating the report.
57 Use the 'arv' and 'jq' commands to get the list of the 100
58 largest collections and generate the deduplication report:
60 arv collection list --order 'file_size_total desc' --limit 100 | \
61 jq -r '.items[] | [.portable_data_hash,.uuid] |@csv' | \
62 sed -e 's/"//g'|tr '\n' ' ' | \
69 loglevel := flags.String("log-level", "info", "logging level (debug, info, ...)")
70 err := flags.Parse(args)
71 if err == flag.ErrHelp {
73 } else if err != nil {
77 inputs := flags.Args()
79 inputs = deDuplicate(inputs)
82 err = fmt.Errorf("Error: no collections provided")
86 lvl, err := logrus.ParseLevel(*loglevel)
94 func blockList(collection arvados.Collection) (blocks map[string]int) {
95 blocks = make(map[string]int)
96 m := manifest.Manifest{Text: collection.ManifestText}
97 blockChannel := m.BlockIterWithDuplicates()
98 for b := range blockChannel {
99 blocks[b.Digest.String()] = b.Size
104 func report(prog string, args []string, logger *logrus.Logger, stdout, stderr io.Writer) (exitcode int) {
108 inputs, err = parseFlags(prog, args, logger, stderr)
110 if err != flag.ErrHelp {
111 logger.Error(err.Error())
117 // Arvados Client setup
118 arv, err := arvadosclient.MakeArvadosClient()
120 logger.Errorf("Error creating Arvados object: %s", err)
130 blocks := make(map[string]map[string]int)
131 pdhs := make(map[string]Col)
132 var nominalSize int64
134 for _, input := range inputs {
137 if strings.Contains(input, ",") {
138 // The input is in the format pdh,uuid. This will allow us to save time on duplicate pdh's
139 tmp := strings.Split(input, ",")
143 // The input must be a plain uuid
146 if !strings.Contains(uuid, "-4zz18-") {
147 logger.Errorf("Error: uuid must refer to collection object")
151 if _, ok := pdhs[pdh]; ok {
152 // We've processed a collection with this pdh already. Simply add its
153 // size to the totals and move on to the next one.
154 // Note that we simply trust the PDH matches the collection UUID here,
155 // in other words, we use it over the UUID. If they don't match, the report
157 nominalSize += pdhs[pdh].FileSizeTotal
159 var collection arvados.Collection
160 err = arv.Get("collections", uuid, nil, &collection)
162 logger.Errorf("Error: unable to retrieve collection: %s", err)
166 blocks[uuid] = make(map[string]int)
167 blocks[uuid] = blockList(collection)
168 if pdh != "" && collection.PortableDataHash != pdh {
169 logger.Errorf("Error: the collection with UUID %s has PDH %s, but a different PDH was provided in the arguments: %s", uuid, collection.PortableDataHash, pdh)
174 pdh = collection.PortableDataHash
178 if collection.FileSizeTotal != 0 || collection.FileCount != 0 {
179 nominalSize += collection.FileSizeTotal
180 col.FileSizeTotal = collection.FileSizeTotal
181 col.FileCount = int64(collection.FileCount)
183 // Collections created with old Arvados versions do not always have the total file size and count cached in the collections object
185 for _, size := range blocks[uuid] {
186 collSize += int64(size)
188 nominalSize += collSize
189 col.FileSizeTotal = collSize
194 if pdhs[pdh].FileCount != 0 {
195 fmt.Fprintf(stdout, "Collection %s: pdh %s; nominal size %d (%s); file count %d\n", uuid, pdh, pdhs[pdh].FileSizeTotal, humanize.IBytes(uint64(pdhs[pdh].FileSizeTotal)), pdhs[pdh].FileCount)
197 fmt.Fprintf(stdout, "Collection %s: pdh %s; nominal size %d (%s)\n", uuid, pdh, pdhs[pdh].FileSizeTotal, humanize.IBytes(uint64(pdhs[pdh].FileSizeTotal)))
202 seen := make(map[string]bool)
203 for _, v := range blocks {
204 for pdh, size := range v {
207 totalSize += int64(size)
212 fmt.Fprintf(stdout, "Collections: %15d\n", len(inputs))
213 fmt.Fprintf(stdout, "Nominal size of stored data: %15d bytes (%s)\n", nominalSize, humanize.IBytes(uint64(nominalSize)))
214 fmt.Fprintf(stdout, "Actual size of stored data: %15d bytes (%s)\n", totalSize, humanize.IBytes(uint64(totalSize)))
215 fmt.Fprintf(stdout, "Saved by Keep deduplication: %15d bytes (%s)\n", nominalSize-totalSize, humanize.IBytes(uint64(nominalSize-totalSize)))