rclone/cmd/bisync/deltas.go
nielash fd95511091 bisync: generate listings concurrently with march -- fixes #7332
Before this change, bisync needed to build a full listing for Path1, then a
full listing for Path2, then compare them -- and each of those tasks needed to
finish before the next one could start. In addition to being slow and
inefficient, it also caused real problems if a file changed between the time
bisync checked it on Path1 and the time it checked the corresponding file on
Path2.

This change solves these problems by listing both paths concurrently, using
the same March infrastructure that check and sync use to traverse two
directories in lock-step, optimized by Go's robust concurrency support.
Listings should now be much faster, and any given path is now checked
nearly-instantaneously on both sides, minimizing room for error.

Further discussion:
https://forum.rclone.org/t/bisync-bugs-and-feature-requests/37636#:~:text=4.%20Listings%20should%20alternate%20between%20paths%20to%20minimize%20errors
2024-01-20 14:50:08 -05:00

455 lines
12 KiB
Go

// Package bisync implements bisync
// Copyright (c) 2017-2020 Chris Nelson
package bisync
import (
"bytes"
"context"
"fmt"
"path/filepath"
"sort"
"strings"
"github.com/rclone/rclone/cmd/bisync/bilib"
"github.com/rclone/rclone/cmd/check"
"github.com/rclone/rclone/fs"
"github.com/rclone/rclone/fs/accounting"
"github.com/rclone/rclone/fs/filter"
"github.com/rclone/rclone/fs/operations"
)
// delta
type delta uint8
const (
deltaZero delta = 0
deltaNew delta = 1 << iota
deltaNewer
deltaOlder
deltaSize
deltaHash
deltaDeleted
)
const (
deltaModified delta = deltaNewer | deltaOlder | deltaSize | deltaHash | deltaDeleted
deltaOther delta = deltaNew | deltaNewer | deltaOlder
)
func (d delta) is(cond delta) bool {
return d&cond != 0
}
// deltaSet
type deltaSet struct {
deltas map[string]delta
opt *Options
fs fs.Fs // base filesystem
msg string // filesystem name for logging
oldCount int // original number of files (for "excess deletes" check)
deleted int // number of deleted files (for "excess deletes" check)
foundSame bool // true if found at least one unchanged file
checkFiles bilib.Names
}
func (ds *deltaSet) empty() bool {
return len(ds.deltas) == 0
}
func (ds *deltaSet) sort() (sorted []string) {
if ds.empty() {
return
}
sorted = make([]string, 0, len(ds.deltas))
for file := range ds.deltas {
sorted = append(sorted, file)
}
sort.Strings(sorted)
return
}
func (ds *deltaSet) printStats() {
if ds.empty() {
return
}
nAll := len(ds.deltas)
nNew := 0
nNewer := 0
nOlder := 0
nDeleted := 0
for _, d := range ds.deltas {
if d.is(deltaNew) {
nNew++
}
if d.is(deltaNewer) {
nNewer++
}
if d.is(deltaOlder) {
nOlder++
}
if d.is(deltaDeleted) {
nDeleted++
}
}
fs.Infof(nil, "%s: %4d changes: %4d new, %4d newer, %4d older, %4d deleted",
ds.msg, nAll, nNew, nNewer, nOlder, nDeleted)
}
// check potential conflicts (to avoid renaming if already identical)
func (b *bisyncRun) checkconflicts(ctxCheck context.Context, filterCheck *filter.Filter, fs1, fs2 fs.Fs) (bilib.Names, error) {
matches := bilib.Names{}
if filterCheck.HaveFilesFrom() {
fs.Debugf(nil, "There are potential conflicts to check.")
opt, close, checkopterr := check.GetCheckOpt(b.fs1, b.fs2)
if checkopterr != nil {
b.critical = true
b.retryable = true
fs.Debugf(nil, "GetCheckOpt error: %v", checkopterr)
return matches, checkopterr
}
defer close()
opt.Match = new(bytes.Buffer)
// TODO: consider using custom CheckFn to act like cryptcheck, if either fs is a crypt remote and -c has been passed
// note that cryptCheck() is not currently exported
fs.Infof(nil, "Checking potential conflicts...")
check := operations.Check(ctxCheck, opt)
fs.Infof(nil, "Finished checking the potential conflicts. %s", check)
//reset error count, because we don't want to count check errors as bisync errors
accounting.Stats(ctxCheck).ResetErrors()
//return the list of identical files to check against later
if len(fmt.Sprint(opt.Match)) > 0 {
matches = bilib.ToNames(strings.Split(fmt.Sprint(opt.Match), "\n"))
}
if matches.NotEmpty() {
fs.Debugf(nil, "The following potential conflicts were determined to be identical. %v", matches)
} else {
fs.Debugf(nil, "None of the conflicts were determined to be identical.")
}
}
return matches, nil
}
// findDeltas
func (b *bisyncRun) findDeltas(fctx context.Context, f fs.Fs, oldListing string, now *fileList, msg string) (ds *deltaSet, err error) {
var old *fileList
newListing := oldListing + "-new"
old, err = b.loadListing(oldListing)
if err != nil {
fs.Errorf(nil, "Failed loading prior %s listing: %s", msg, oldListing)
b.abort = true
return
}
if err = b.checkListing(old, oldListing, "prior "+msg); err != nil {
return
}
if err == nil {
err = b.checkListing(now, newListing, "current "+msg)
}
if err != nil {
return
}
ds = &deltaSet{
deltas: map[string]delta{},
fs: f,
msg: msg,
oldCount: len(old.list),
opt: b.opt,
checkFiles: bilib.Names{},
}
for _, file := range old.list {
d := deltaZero
if !now.has(file) {
b.indent(msg, file, "File was deleted")
ds.deleted++
d |= deltaDeleted
} else {
// skip dirs here, as we only care if they are new/deleted, not newer/older
if !now.isDir(file) {
if old.getTime(file) != now.getTime(file) {
if old.beforeOther(now, file) {
fs.Debugf(file, "(old: %v current: %v", old.getTime(file), now.getTime(file))
b.indent(msg, file, "File is newer")
d |= deltaNewer
} else { // Current version is older than prior sync.
fs.Debugf(file, "(old: %v current: %v", old.getTime(file), now.getTime(file))
b.indent(msg, file, "File is OLDER")
d |= deltaOlder
}
}
// TODO Compare sizes and hashes
}
}
if d.is(deltaModified) {
ds.deltas[file] = d
} else {
// Once we've found at least one unchanged file,
// we know that not everything has changed,
// as with a DST time change
ds.foundSame = true
}
}
for _, file := range now.list {
if !old.has(file) {
b.indent(msg, file, "File is new")
ds.deltas[file] = deltaNew
}
}
if b.opt.CheckAccess {
// checkFiles is a small structure compared with the `now`, so we
// return it alone and let the full delta map be garbage collected.
for _, file := range now.list {
if filepath.Base(file) == b.opt.CheckFilename {
ds.checkFiles.Add(file)
}
}
}
return
}
// applyDeltas
func (b *bisyncRun) applyDeltas(ctx context.Context, ds1, ds2 *deltaSet) (changes1, changes2 bool, results2to1, results1to2 []Results, queues queues, err error) {
path1 := bilib.FsPath(b.fs1)
path2 := bilib.FsPath(b.fs2)
copy1to2 := bilib.Names{}
copy2to1 := bilib.Names{}
delete1 := bilib.Names{}
delete2 := bilib.Names{}
handled := bilib.Names{}
renamed1 := bilib.Names{}
renamed2 := bilib.Names{}
renameSkipped := bilib.Names{}
deletedonboth := bilib.Names{}
skippedDirs1 := newFileList()
skippedDirs2 := newFileList()
ctxMove := b.opt.setDryRun(ctx)
// efficient isDir check
// we load the listing just once and store only the dirs
dirs1, dirs1Err := b.listDirsOnly(1)
if dirs1Err != nil {
b.critical = true
b.retryable = true
fs.Debugf(nil, "Error generating dirsonly list for path1: %v", dirs1Err)
return
}
dirs2, dirs2Err := b.listDirsOnly(2)
if dirs2Err != nil {
b.critical = true
b.retryable = true
fs.Debugf(nil, "Error generating dirsonly list for path2: %v", dirs2Err)
return
}
// build a list of only the "deltaOther"s so we don't have to check more files than necessary
// this is essentially the same as running rclone check with a --files-from filter, then exempting the --match results from being renamed
// we therefore avoid having to list the same directory more than once.
// we are intentionally overriding DryRun here because we need to perform the check, even during a dry run, or the results would be inaccurate.
// check is a read-only operation by its nature, so it's already "dry" in that sense.
ctxNew, ciCheck := fs.AddConfig(ctx)
ciCheck.DryRun = false
ctxCheck, filterCheck := filter.AddConfig(ctxNew)
for _, file := range ds1.sort() {
d1 := ds1.deltas[file]
if d1.is(deltaOther) {
d2 := ds2.deltas[file]
if d2.is(deltaOther) {
if err := filterCheck.AddFile(file); err != nil {
fs.Debugf(nil, "Non-critical error adding file to list of potential conflicts to check: %s", err)
} else {
fs.Debugf(nil, "Added file to list of potential conflicts to check: %s", file)
}
}
}
}
//if there are potential conflicts to check, check them all here (outside the loop) in one fell swoop
matches, err := b.checkconflicts(ctxCheck, filterCheck, b.fs1, b.fs2)
for _, file := range ds1.sort() {
p1 := path1 + file
p2 := path2 + file
d1 := ds1.deltas[file]
if d1.is(deltaOther) {
d2, in2 := ds2.deltas[file]
if !in2 {
b.indent("Path1", p2, "Queue copy to Path2")
copy1to2.Add(file)
} else if d2.is(deltaDeleted) {
b.indent("Path1", p2, "Queue copy to Path2")
copy1to2.Add(file)
handled.Add(file)
} else if d2.is(deltaOther) {
b.indent("!WARNING", file, "New or changed in both paths")
//if files are identical, leave them alone instead of renaming
if dirs1.has(file) && dirs2.has(file) {
fs.Debugf(nil, "This is a directory, not a file. Skipping equality check and will not rename: %s", file)
ls1.getPut(file, skippedDirs1)
ls2.getPut(file, skippedDirs2)
} else {
equal := matches.Has(file)
if equal {
fs.Infof(nil, "Files are equal! Skipping: %s", file)
renameSkipped.Add(file)
} else {
fs.Debugf(nil, "Files are NOT equal: %s", file)
b.indent("!Path1", p1+"..path1", "Renaming Path1 copy")
if err = operations.MoveFile(ctxMove, b.fs1, b.fs1, file+"..path1", file); err != nil {
err = fmt.Errorf("path1 rename failed for %s: %w", p1, err)
b.critical = true
return
}
if b.opt.DryRun {
renameSkipped.Add(file)
} else {
renamed1.Add(file)
}
b.indent("!Path1", p2+"..path1", "Queue copy to Path2")
copy1to2.Add(file + "..path1")
b.indent("!Path2", p2+"..path2", "Renaming Path2 copy")
if err = operations.MoveFile(ctxMove, b.fs2, b.fs2, file+"..path2", file); err != nil {
err = fmt.Errorf("path2 rename failed for %s: %w", file, err)
return
}
if b.opt.DryRun {
renameSkipped.Add(file)
} else {
renamed2.Add(file)
}
b.indent("!Path2", p1+"..path2", "Queue copy to Path1")
copy2to1.Add(file + "..path2")
}
}
handled.Add(file)
}
} else {
// Path1 deleted
d2, in2 := ds2.deltas[file]
if !in2 {
b.indent("Path2", p2, "Queue delete")
delete2.Add(file)
copy1to2.Add(file)
} else if d2.is(deltaOther) {
b.indent("Path2", p1, "Queue copy to Path1")
copy2to1.Add(file)
handled.Add(file)
} else if d2.is(deltaDeleted) {
handled.Add(file)
deletedonboth.Add(file)
}
}
}
for _, file := range ds2.sort() {
p1 := path1 + file
d2 := ds2.deltas[file]
if handled.Has(file) {
continue
}
if d2.is(deltaOther) {
b.indent("Path2", p1, "Queue copy to Path1")
copy2to1.Add(file)
} else {
// Deleted
b.indent("Path1", p1, "Queue delete")
delete1.Add(file)
copy2to1.Add(file)
}
}
// Do the batch operation
if copy2to1.NotEmpty() {
changes1 = true
b.indent("Path2", "Path1", "Do queued copies to")
results2to1, err = b.fastCopy(ctx, b.fs2, b.fs1, copy2to1, "copy2to1")
if err != nil {
return
}
//copy empty dirs from path2 to path1 (if --create-empty-src-dirs)
b.syncEmptyDirs(ctx, b.fs1, copy2to1, dirs2, &results2to1, "make")
}
if copy1to2.NotEmpty() {
changes2 = true
b.indent("Path1", "Path2", "Do queued copies to")
results1to2, err = b.fastCopy(ctx, b.fs1, b.fs2, copy1to2, "copy1to2")
if err != nil {
return
}
//copy empty dirs from path1 to path2 (if --create-empty-src-dirs)
b.syncEmptyDirs(ctx, b.fs2, copy1to2, dirs1, &results1to2, "make")
}
if delete1.NotEmpty() {
if err = b.saveQueue(delete1, "delete1"); err != nil {
return
}
//propagate deletions of empty dirs from path2 to path1 (if --create-empty-src-dirs)
b.syncEmptyDirs(ctx, b.fs1, delete1, dirs1, &results2to1, "remove")
}
if delete2.NotEmpty() {
if err = b.saveQueue(delete2, "delete2"); err != nil {
return
}
//propagate deletions of empty dirs from path1 to path2 (if --create-empty-src-dirs)
b.syncEmptyDirs(ctx, b.fs2, delete2, dirs2, &results1to2, "remove")
}
queues.copy1to2 = copy1to2
queues.copy2to1 = copy2to1
queues.renamed1 = renamed1
queues.renamed2 = renamed2
queues.renameSkipped = renameSkipped
queues.deletedonboth = deletedonboth
queues.skippedDirs1 = skippedDirs1
queues.skippedDirs2 = skippedDirs2
return
}
// excessDeletes checks whether number of deletes is within allowed range
func (ds *deltaSet) excessDeletes() bool {
maxDelete := ds.opt.MaxDelete
maxRatio := float64(maxDelete) / 100.0
curRatio := 0.0
if ds.deleted > 0 && ds.oldCount > 0 {
curRatio = float64(ds.deleted) / float64(ds.oldCount)
}
if curRatio <= maxRatio {
return false
}
fs.Errorf("Safety abort",
"too many deletes (>%d%%, %d of %d) on %s %s. Run with --force if desired.",
maxDelete, ds.deleted, ds.oldCount, ds.msg, quotePath(bilib.FsPath(ds.fs)))
return true
}