This PR refactors the way we use buffers for O_DIRECT and to re-use those buffers for messagepack reader writer. After some extensive benchmarking found that not all objects have this benefit, and only objects smaller than 64KiB see this benefit overall. Benefits are seen from almost all objects from 1KiB - 32KiB Beyond this no objects see benefit with bulk call approach as the latency of bytes sent over the wire v/s streaming content directly from disk negate each other with no remarkable benefits. All other optimizations include reuse of msgp.Reader, msgp.Writer using sync.Pool's for all internode calls.
281 lines
9 KiB
Go
281 lines
9 KiB
Go
/*
|
|
* MinIO Cloud Storage, (C) 2016-2019 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 cmd
|
|
|
|
import (
|
|
"context"
|
|
"errors"
|
|
"hash/crc32"
|
|
|
|
"github.com/minio/minio/cmd/logger"
|
|
"github.com/minio/minio/pkg/sync/errgroup"
|
|
)
|
|
|
|
// Returns number of errors that occurred the most (incl. nil) and the
|
|
// corresponding error value. NB When there is more than one error value that
|
|
// occurs maximum number of times, the error value returned depends on how
|
|
// golang's map orders keys. This doesn't affect correctness as long as quorum
|
|
// value is greater than or equal to simple majority, since none of the equally
|
|
// maximal values would occur quorum or more number of times.
|
|
func reduceErrs(errs []error, ignoredErrs []error) (maxCount int, maxErr error) {
|
|
errorCounts := make(map[error]int)
|
|
for _, err := range errs {
|
|
if IsErrIgnored(err, ignoredErrs...) {
|
|
continue
|
|
}
|
|
errorCounts[err]++
|
|
}
|
|
|
|
max := 0
|
|
for err, count := range errorCounts {
|
|
switch {
|
|
case max < count:
|
|
max = count
|
|
maxErr = err
|
|
|
|
// Prefer `nil` over other error values with the same
|
|
// number of occurrences.
|
|
case max == count && err == nil:
|
|
maxErr = err
|
|
}
|
|
}
|
|
return max, maxErr
|
|
}
|
|
|
|
// reduceQuorumErrs behaves like reduceErrs by only for returning
|
|
// values of maximally occurring errors validated against a generic
|
|
// quorum number that can be read or write quorum depending on usage.
|
|
func reduceQuorumErrs(ctx context.Context, errs []error, ignoredErrs []error, quorum int, quorumErr error) error {
|
|
maxCount, maxErr := reduceErrs(errs, ignoredErrs)
|
|
if maxCount >= quorum {
|
|
return maxErr
|
|
}
|
|
return quorumErr
|
|
}
|
|
|
|
// reduceReadQuorumErrs behaves like reduceErrs but only for returning
|
|
// values of maximally occurring errors validated against readQuorum.
|
|
func reduceReadQuorumErrs(ctx context.Context, errs []error, ignoredErrs []error, readQuorum int) (maxErr error) {
|
|
return reduceQuorumErrs(ctx, errs, ignoredErrs, readQuorum, errErasureReadQuorum)
|
|
}
|
|
|
|
// reduceWriteQuorumErrs behaves like reduceErrs but only for returning
|
|
// values of maximally occurring errors validated against writeQuorum.
|
|
func reduceWriteQuorumErrs(ctx context.Context, errs []error, ignoredErrs []error, writeQuorum int) (maxErr error) {
|
|
return reduceQuorumErrs(ctx, errs, ignoredErrs, writeQuorum, errErasureWriteQuorum)
|
|
}
|
|
|
|
// Similar to 'len(slice)' but returns the actual elements count
|
|
// skipping the unallocated elements.
|
|
func diskCount(disks []StorageAPI) int {
|
|
diskCount := 0
|
|
for _, disk := range disks {
|
|
if disk == nil {
|
|
continue
|
|
}
|
|
diskCount++
|
|
}
|
|
return diskCount
|
|
}
|
|
|
|
// hashOrder - hashes input key to return consistent
|
|
// hashed integer slice. Returned integer order is salted
|
|
// with an input key. This results in consistent order.
|
|
// NOTE: collisions are fine, we are not looking for uniqueness
|
|
// in the slices returned.
|
|
func hashOrder(key string, cardinality int) []int {
|
|
if cardinality <= 0 {
|
|
// Returns an empty int slice for cardinality < 0.
|
|
return nil
|
|
}
|
|
|
|
nums := make([]int, cardinality)
|
|
keyCrc := crc32.Checksum([]byte(key), crc32.IEEETable)
|
|
|
|
start := int(keyCrc % uint32(cardinality))
|
|
for i := 1; i <= cardinality; i++ {
|
|
nums[i-1] = 1 + ((start + i) % cardinality)
|
|
}
|
|
return nums
|
|
}
|
|
|
|
// Reads all `xl.meta` metadata as a FileInfo slice.
|
|
// Returns error slice indicating the failed metadata reads.
|
|
func readAllFileInfo(ctx context.Context, disks []StorageAPI, bucket, object, versionID string, readData bool) ([]FileInfo, []error) {
|
|
metadataArray := make([]FileInfo, len(disks))
|
|
|
|
g := errgroup.WithNErrs(len(disks))
|
|
// Read `xl.meta` in parallel across disks.
|
|
for index := range disks {
|
|
index := index
|
|
g.Go(func() (err error) {
|
|
if disks[index] == nil {
|
|
return errDiskNotFound
|
|
}
|
|
metadataArray[index], err = disks[index].ReadVersion(ctx, bucket, object, versionID, readData)
|
|
if err != nil {
|
|
if !IsErr(err, errFileNotFound, errVolumeNotFound, errFileVersionNotFound, errDiskNotFound) {
|
|
logger.LogOnceIf(ctx, err, disks[index].String())
|
|
}
|
|
}
|
|
return err
|
|
}, index)
|
|
}
|
|
|
|
// Return all the metadata.
|
|
return metadataArray, g.Wait()
|
|
}
|
|
|
|
func shuffleDisksAndPartsMetadataByIndex(disks []StorageAPI, metaArr []FileInfo, distribution []int) (shuffledDisks []StorageAPI, shuffledPartsMetadata []FileInfo) {
|
|
shuffledDisks = make([]StorageAPI, len(disks))
|
|
shuffledPartsMetadata = make([]FileInfo, len(disks))
|
|
var inconsistent int
|
|
for i, meta := range metaArr {
|
|
if disks[i] == nil {
|
|
// Assuming offline drives as inconsistent,
|
|
// to be safe and fallback to original
|
|
// distribution order.
|
|
inconsistent++
|
|
continue
|
|
}
|
|
// check if erasure distribution order matches the index
|
|
// position if this is not correct we discard the disk
|
|
// and move to collect others
|
|
if distribution[i] != meta.Erasure.Index {
|
|
inconsistent++ // keep track of inconsistent entries
|
|
continue
|
|
}
|
|
shuffledDisks[meta.Erasure.Index-1] = disks[i]
|
|
shuffledPartsMetadata[meta.Erasure.Index-1] = metaArr[i]
|
|
}
|
|
|
|
// Inconsistent meta info is with in the limit of
|
|
// expected quorum, proceed with EcIndex based
|
|
// disk order.
|
|
if inconsistent < len(disks)/2 {
|
|
return shuffledDisks, shuffledPartsMetadata
|
|
}
|
|
|
|
// fall back to original distribution based order.
|
|
return shuffleDisksAndPartsMetadata(disks, metaArr, distribution)
|
|
}
|
|
|
|
// Return shuffled partsMetadata depending on distribution.
|
|
func shuffleDisksAndPartsMetadata(disks []StorageAPI, partsMetadata []FileInfo, distribution []int) (shuffledDisks []StorageAPI, shuffledPartsMetadata []FileInfo) {
|
|
if distribution == nil {
|
|
return disks, partsMetadata
|
|
}
|
|
shuffledDisks = make([]StorageAPI, len(disks))
|
|
shuffledPartsMetadata = make([]FileInfo, len(partsMetadata))
|
|
// Shuffle slice xl metadata for expected distribution.
|
|
for index := range partsMetadata {
|
|
blockIndex := distribution[index]
|
|
shuffledPartsMetadata[blockIndex-1] = partsMetadata[index]
|
|
shuffledDisks[blockIndex-1] = disks[index]
|
|
}
|
|
return shuffledDisks, shuffledPartsMetadata
|
|
}
|
|
|
|
// Return shuffled partsMetadata depending on distribution.
|
|
func shufflePartsMetadata(partsMetadata []FileInfo, distribution []int) (shuffledPartsMetadata []FileInfo) {
|
|
if distribution == nil {
|
|
return partsMetadata
|
|
}
|
|
shuffledPartsMetadata = make([]FileInfo, len(partsMetadata))
|
|
// Shuffle slice xl metadata for expected distribution.
|
|
for index := range partsMetadata {
|
|
blockIndex := distribution[index]
|
|
shuffledPartsMetadata[blockIndex-1] = partsMetadata[index]
|
|
}
|
|
return shuffledPartsMetadata
|
|
}
|
|
|
|
// shuffleDisks - shuffle input disks slice depending on the
|
|
// erasure distribution. Return shuffled slice of disks with
|
|
// their expected distribution.
|
|
func shuffleDisks(disks []StorageAPI, distribution []int) (shuffledDisks []StorageAPI) {
|
|
if distribution == nil {
|
|
return disks
|
|
}
|
|
shuffledDisks = make([]StorageAPI, len(disks))
|
|
// Shuffle disks for expected distribution.
|
|
for index := range disks {
|
|
blockIndex := distribution[index]
|
|
shuffledDisks[blockIndex-1] = disks[index]
|
|
}
|
|
return shuffledDisks
|
|
}
|
|
|
|
// evalDisks - returns a new slice of disks where nil is set if
|
|
// the corresponding error in errs slice is not nil
|
|
func evalDisks(disks []StorageAPI, errs []error) []StorageAPI {
|
|
if len(errs) != len(disks) {
|
|
logger.LogIf(GlobalContext, errors.New("unexpected disks/errors slice length"))
|
|
return nil
|
|
}
|
|
newDisks := make([]StorageAPI, len(disks))
|
|
for index := range errs {
|
|
if errs[index] == nil {
|
|
newDisks[index] = disks[index]
|
|
} else {
|
|
newDisks[index] = nil
|
|
}
|
|
}
|
|
return newDisks
|
|
}
|
|
|
|
// Errors specifically generated by calculatePartSizeFromIdx function.
|
|
var (
|
|
errPartSizeZero = errors.New("Part size cannot be zero")
|
|
errPartSizeIndex = errors.New("Part index cannot be smaller than 1")
|
|
)
|
|
|
|
// calculatePartSizeFromIdx calculates the part size according to input index.
|
|
// returns error if totalSize is -1, partSize is 0, partIndex is 0.
|
|
func calculatePartSizeFromIdx(ctx context.Context, totalSize int64, partSize int64, partIndex int) (currPartSize int64, err error) {
|
|
if totalSize < -1 {
|
|
logger.LogIf(ctx, errInvalidArgument)
|
|
return 0, errInvalidArgument
|
|
}
|
|
if partSize == 0 {
|
|
logger.LogIf(ctx, errPartSizeZero)
|
|
return 0, errPartSizeZero
|
|
}
|
|
if partIndex < 1 {
|
|
logger.LogIf(ctx, errPartSizeIndex)
|
|
return 0, errPartSizeIndex
|
|
}
|
|
if totalSize == -1 {
|
|
return -1, nil
|
|
}
|
|
if totalSize > 0 {
|
|
// Compute the total count of parts
|
|
partsCount := totalSize/partSize + 1
|
|
// Return the part's size
|
|
switch {
|
|
case int64(partIndex) < partsCount:
|
|
currPartSize = partSize
|
|
case int64(partIndex) == partsCount:
|
|
// Size of last part
|
|
currPartSize = totalSize % partSize
|
|
default:
|
|
currPartSize = 0
|
|
}
|
|
}
|
|
return currPartSize, nil
|
|
}
|