mirror of
https://github.com/minio/minio.git
synced 2025-11-10 14:09:48 -05:00
Optimize healObject by eliminating extra data passes (#4949)
This commit is contained in:
committed by
Dee Koder
parent
94670a387e
commit
4c9fae90ff
@@ -17,13 +17,16 @@
|
||||
package cmd
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"hash"
|
||||
"strings"
|
||||
)
|
||||
|
||||
// HealFile tries to reconstruct an erasure-coded file spread over all
|
||||
// available disks. HealFile will read the valid parts of the file,
|
||||
// reconstruct the missing data and write the reconstructed parts back
|
||||
// to `staleDisks`.
|
||||
// to `staleDisks` at the destination `dstVol/dstPath/`. Parts are
|
||||
// verified against the given BitrotAlgorithm and checksums.
|
||||
//
|
||||
// `staleDisks` is a slice of disks where each non-nil entry has stale
|
||||
// or no data, and so will be healed.
|
||||
@@ -34,19 +37,17 @@ import (
|
||||
// In addition, `staleDisks` and `s.disks` must have the same ordering
|
||||
// of disks w.r.t. erasure coding of the object.
|
||||
//
|
||||
// The function will try to read the valid parts from the file under
|
||||
// the given volume and path and tries to reconstruct the file under
|
||||
// the given healVolume and healPath (on staleDisks). The given
|
||||
// algorithm will be used to verify the valid parts and to protect the
|
||||
// reconstructed file.
|
||||
// Errors when writing to `staleDisks` are not propagated as long as
|
||||
// writes succeed for at least one disk. This allows partial healing
|
||||
// despite stale disks being faulty.
|
||||
//
|
||||
// It returns bitrot checksums for the non-nil staleDisks.
|
||||
func (s ErasureStorage) HealFile(staleDisks []StorageAPI, volume, path string,
|
||||
blocksize int64, healVolume, healPath string, size int64,
|
||||
algorithm BitrotAlgorithm, checksums [][]byte) (f ErasureFileInfo,
|
||||
err error) {
|
||||
// It returns bitrot checksums for the non-nil staleDisks on which
|
||||
// healing succeeded.
|
||||
func (s ErasureStorage) HealFile(staleDisks []StorageAPI, volume, path string, blocksize int64,
|
||||
dstVol, dstPath string, size int64, alg BitrotAlgorithm, checksums [][]byte) (
|
||||
f ErasureFileInfo, err error) {
|
||||
|
||||
if !algorithm.Available() {
|
||||
if !alg.Available() {
|
||||
return f, traceError(errBitrotHashAlgoInvalid)
|
||||
}
|
||||
|
||||
@@ -57,15 +58,15 @@ func (s ErasureStorage) HealFile(staleDisks []StorageAPI, volume, path string,
|
||||
for i, disk := range s.disks {
|
||||
switch {
|
||||
case staleDisks[i] != nil:
|
||||
hashers[i] = algorithm.New()
|
||||
hashers[i] = alg.New()
|
||||
case disk == nil:
|
||||
// disregard unavailable disk
|
||||
continue
|
||||
default:
|
||||
verifiers[i] = NewBitrotVerifier(algorithm, checksums[i])
|
||||
f.Checksums[i] = checksums[i]
|
||||
verifiers[i] = NewBitrotVerifier(alg, checksums[i])
|
||||
}
|
||||
}
|
||||
writeErrors := make([]error, len(s.disks))
|
||||
|
||||
// Scan part files on disk, block-by-block reconstruct it and
|
||||
// write to stale disks.
|
||||
@@ -125,27 +126,48 @@ func (s ErasureStorage) HealFile(staleDisks []StorageAPI, volume, path string,
|
||||
|
||||
// write computed shards as chunks on file in each
|
||||
// stale disk
|
||||
writeSucceeded := false
|
||||
for i, disk := range staleDisks {
|
||||
if disk == nil {
|
||||
// skip nil disk or disk that had error on
|
||||
// previous write
|
||||
if disk == nil || writeErrors[i] != nil {
|
||||
continue
|
||||
}
|
||||
|
||||
err = disk.AppendFile(healVolume, healPath, blocks[i])
|
||||
if err != nil {
|
||||
return f, traceError(err)
|
||||
writeErrors[i] = disk.AppendFile(dstVol, dstPath, blocks[i])
|
||||
if writeErrors[i] == nil {
|
||||
hashers[i].Write(blocks[i])
|
||||
writeSucceeded = true
|
||||
}
|
||||
hashers[i].Write(blocks[i])
|
||||
}
|
||||
|
||||
// If all disks had write errors we quit.
|
||||
if !writeSucceeded {
|
||||
// build error from all write errors
|
||||
return f, traceError(joinWriteErrors(writeErrors))
|
||||
}
|
||||
}
|
||||
|
||||
// copy computed file hashes into output variable
|
||||
f.Size = size
|
||||
f.Algorithm = algorithm
|
||||
f.Algorithm = alg
|
||||
for i, disk := range staleDisks {
|
||||
if disk == nil {
|
||||
if disk == nil || writeErrors[i] != nil {
|
||||
continue
|
||||
}
|
||||
f.Checksums[i] = hashers[i].Sum(nil)
|
||||
}
|
||||
return f, nil
|
||||
}
|
||||
|
||||
func joinWriteErrors(errs []error) error {
|
||||
msgs := []string{}
|
||||
for i, err := range errs {
|
||||
if err == nil {
|
||||
continue
|
||||
}
|
||||
msgs = append(msgs, fmt.Sprintf("disk %d: %v", i+1, err))
|
||||
}
|
||||
return fmt.Errorf("all stale disks had write errors during healing: %s",
|
||||
strings.Join(msgs, ", "))
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user