MyFavMirrors/minio
1
0
Fork 0
mirror of https://github.com/minio/minio.git synced 2025-11-07 21:02:58 -05:00
Code Issues Packages Projects Releases Wiki Activity

shuffle buckets randomly before being scanned (#17644)

Browse Source 
this randomness is needed to avoid scanning
the same buckets across different erasure sets,
in the same order.

allow random buckets to be scanned instead
allowing a wider spread of ILM, replication
checks.

Additionally do not loop over twice to fill
the channel, fill the channel regardless of
having bucket new or old.
This commit is contained in:
Harshavardhana
2023-07-14 02:25:40 -07:00
committed by GitHub
parent bb6921bf9c
commit bdddf597f6
4 changed files with 52 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
cmd/erasure.go
View File

@@ -277,31 +277,33 @@ func (er erasureObjects) getOnlineDisksWithHealing() (newDisks []StorageAPI, hea
var wg sync.WaitGroup
disks := er.getDisks()
infos := make([]DiskInfo, len(disks))
for _, i := range hashOrder(UTCNow().String(), len(disks)) {
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for _, i := range r.Perm(len(disks)) {
i := i
wg.Add(1)
go func() {
defer wg.Done()
disk := disks[i-1]
disk := disks[i]
if disk == nil {
infos[i-1].Error = "nil drive"
infos[i].Error = "offline drive"
return
}
di, err := disk.DiskInfo(context.Background())
if err != nil {
if err != nil || di.Healing {
// - Do not consume disks which are not reachable
// unformatted or simply not accessible for some reason.
//
//
// - Future: skip busy disks
infos[i-1].Error = err.Error()
if err != nil {
infos[i].Error = err.Error()
}
return
}
infos[i-1] = di
infos[i] = di
}()
}
wg.Wait()
@@ -373,23 +375,30 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, wa
// Put all buckets into channel.
bucketCh := make(chan BucketInfo, len(buckets))
// Shuffle buckets to ensure total randomness of buckets, being scanned.
// Otherwise same set of buckets get scanned across erasure sets always.
// at any given point in time. This allows different buckets to be scanned
// in different order per erasure set, this wider spread is needed when
// there are lots of buckets with different order of objects in them.
r := rand.New(rand.NewSource(time.Now().UnixNano()))
permutes := r.Perm(len(buckets))
// Add new buckets first
for _, b := range buckets {
if oldCache.find(b.Name) == nil {
for _, idx := range permutes {
b := buckets[idx]
if e := oldCache.find(b.Name); e == nil {
bucketCh <- b
}
}
// Add existing buckets.
for _, b := range buckets {
e := oldCache.find(b.Name)
if e != nil {
for _, idx := range permutes {
b := buckets[idx]
if e := oldCache.find(b.Name); e != nil {
cache.replace(b.Name, dataUsageRoot, *e)
bucketCh <- b
}
}
close(bucketCh)
bucketResults := make(chan dataUsageEntryInfo, len(disks))
// Start async collector/saver.
@@ -428,11 +437,6 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, wa
}
}()
// Shuffle disks to ensure a total randomness of bucket/disk association to ensure
// that objects that are not present in all disks are accounted and ILM applied.
r := rand.New(rand.NewSource(time.Now().UnixNano()))
r.Shuffle(len(disks), func(i, j int) { disks[i], disks[j] = disks[j], disks[i] })
// Restrict parallelism for disk usage scanner
// upto GOMAXPROCS if GOMAXPROCS is < len(disks)
maxProcs := runtime.GOMAXPROCS(0)