mirror of
https://github.com/pgsty/minio.git
synced 2026-11-12 02:25:34 +03:00
fix(site-replication): snapshot state before bucket healing and peer hooks
Snapshot site-replication topology and release the state read lock before bucket healing, metadata healing, IAM deletion healing, and replication hooks perform storage I/O or peer RPCs. Guard remaining peer-state reads and cloning edits so concurrent configuration updates neither deadlock waiting writers nor race on state maps. Signed-off-by: Feng Ruohang <rh@vonng.com>
This commit is contained in:
@@ -5,6 +5,7 @@ package cmd
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/base64"
|
||||
"encoding/json"
|
||||
"maps"
|
||||
"net/http"
|
||||
@@ -196,3 +197,323 @@ func TestSiteReplicationMetaInfoCachedPeerState(t *testing.T) {
|
||||
t.Fatal("returned cached peer state that was not requested")
|
||||
}
|
||||
}
|
||||
|
||||
type siteBucketDeletePausedLayer struct {
|
||||
ObjectLayer
|
||||
once sync.Once
|
||||
deleting, resume chan struct{}
|
||||
}
|
||||
|
||||
func (l *siteBucketDeletePausedLayer) DeleteBucket(ctx context.Context, bucket string, opts DeleteBucketOptions) error {
|
||||
l.once.Do(func() { close(l.deleting); <-l.resume })
|
||||
return l.ObjectLayer.DeleteBucket(ctx, bucket, opts)
|
||||
}
|
||||
|
||||
func TestSiteReplicationHealAndHooksReleaseLock(t *testing.T) {
|
||||
cases := []string{
|
||||
"make-bucket-hook",
|
||||
"delete-bucket-hook",
|
||||
"bucket-meta-hook",
|
||||
"iam-change-hook",
|
||||
"configure-repl",
|
||||
"peer-bucket-delete",
|
||||
"heal-bucket-make",
|
||||
"heal-bucket-delete",
|
||||
"heal-bucket-repl-config",
|
||||
"heal-bucket-config",
|
||||
"heal-cors",
|
||||
"heal-ilm-expiry-config",
|
||||
"heal-bucket-ilm-expiry",
|
||||
}
|
||||
for _, tc := range cases {
|
||||
t.Run(tc, func(t *testing.T) {
|
||||
site := newSvcAccSite(t)
|
||||
c := &globalSiteReplicationSys
|
||||
const bucket = "sr-lock-test-bucket"
|
||||
mustIAM(t, site.obj.MakeBucket(site.ctx, bucket, MakeBucketOptions{VersioningEnabled: true}))
|
||||
_, err := globalBucketMetadataSys.Update(site.ctx, bucket, bucketVersioningConfig, enabledBucketVersioningConfig)
|
||||
mustIAM(t, err)
|
||||
|
||||
entered, proceed := make(chan struct{}), make(chan struct{})
|
||||
var once sync.Once
|
||||
pause := func() {
|
||||
once.Do(func() { close(entered) })
|
||||
<-proceed
|
||||
}
|
||||
release := sync.OnceFunc(func() { close(proceed) })
|
||||
defer release()
|
||||
|
||||
peer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
pause()
|
||||
if strings.HasSuffix(r.URL.Path, iamRevisionPeerPath) {
|
||||
w.Header().Set("Content-Type", "application/json")
|
||||
_ = json.NewEncoder(w).Encode(iamRevisionResponse{
|
||||
iamRevisionStatus: iamRevisionStatus{Version: iamRevisionProtocol, Node: "node", Instance: "boot", Digest: "fixture"},
|
||||
})
|
||||
return
|
||||
}
|
||||
if r.Method == http.MethodGet {
|
||||
if _, ok := r.URL.Query()["versioning"]; ok {
|
||||
w.Header().Set("Content-Type", "application/xml")
|
||||
_, _ = w.Write([]byte(`<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"><Status>Enabled</Status></VersioningConfiguration>`))
|
||||
return
|
||||
}
|
||||
if _, ok := r.URL.Query()["location"]; ok {
|
||||
w.Header().Set("Content-Type", "application/xml")
|
||||
_, _ = w.Write([]byte(`<LocationConstraint xmlns="http://s3.amazonaws.com/doc/2006-03-01/"></LocationConstraint>`))
|
||||
return
|
||||
}
|
||||
}
|
||||
w.WriteHeader(http.StatusOK)
|
||||
}))
|
||||
defer peer.Close()
|
||||
|
||||
pausedObj := &siteBucketDeletePausedLayer{
|
||||
ObjectLayer: site.obj,
|
||||
deleting: entered,
|
||||
resume: proceed,
|
||||
}
|
||||
if tc == "peer-bucket-delete" || tc == "heal-bucket-delete" {
|
||||
setObjectLayer(pausedObj)
|
||||
defer setObjectLayer(site.obj)
|
||||
}
|
||||
|
||||
local := globalDeploymentID()
|
||||
now := UTCNow().Truncate(time.Second)
|
||||
peers := site.peers(peer.URL)
|
||||
for id, p := range peers {
|
||||
p.ReplicateILMExpiry = true
|
||||
peers[id] = p
|
||||
}
|
||||
c.Lock()
|
||||
c.enabled = true
|
||||
c.state = srState{
|
||||
Name: "local",
|
||||
ServiceAccountAccessKey: site.parent.AccessKey,
|
||||
Peers: peers,
|
||||
UpdatedAt: now,
|
||||
}
|
||||
c.Unlock()
|
||||
|
||||
done := make(chan error, 1)
|
||||
go func() {
|
||||
switch tc {
|
||||
case "make-bucket-hook":
|
||||
done <- c.MakeBucketHook(site.ctx, bucket, MakeBucketOptions{CreatedAt: now})
|
||||
case "delete-bucket-hook":
|
||||
done <- c.DeleteBucketHook(site.ctx, bucket, true)
|
||||
case "bucket-meta-hook":
|
||||
done <- c.BucketMetaHook(site.ctx, madmin.SRBucketMeta{
|
||||
Type: madmin.SRBucketMetaTypeTags,
|
||||
Bucket: bucket,
|
||||
UpdatedAt: now,
|
||||
})
|
||||
case "iam-change-hook":
|
||||
site.create(t, "sr-lock-hook-sa")
|
||||
done <- c.IAMChangeHook(site.ctx, madmin.SRIAMItem{
|
||||
Type: madmin.SRIAMItemSvcAcc,
|
||||
SvcAccChange: &madmin.SRSvcAccChange{Create: &madmin.SRSvcAccCreate{AccessKey: "sr-lock-hook-sa"}},
|
||||
UpdatedAt: now,
|
||||
})
|
||||
case "configure-repl":
|
||||
done <- c.PeerBucketConfigureReplHandler(site.ctx, bucket)
|
||||
case "peer-bucket-delete":
|
||||
done <- c.PeerBucketDeleteHandler(site.ctx, bucket, DeleteBucketOptions{Force: true})
|
||||
case "heal-bucket-make":
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now}}},
|
||||
"remote": {HasBucket: false},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healBucket(site.ctx, site.obj, bucket, info)
|
||||
case "heal-bucket-delete":
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute), DeletedAt: now}}},
|
||||
"remote": {HasBucket: false, BucketMarkedDeleted: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{DeletedAt: now.Add(-2 * time.Minute)}}},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healBucket(site.ctx, pausedObj, bucket, info)
|
||||
case "heal-bucket-repl-config":
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, ReplicationCfgMismatch: true},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healBucketReplicationConfig(site.ctx, site.obj, bucket, info, &validateReplicationDestinationOptions{CheckReady: true})
|
||||
case "heal-bucket-config":
|
||||
tagPayload := base64.StdEncoding.EncodeToString([]byte(`<Tagging><TagSet><Tag><Key>k</Key><Value>v</Value></Tag></TagSet></Tagging>`))
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute), Tags: &tagPayload, TagConfigUpdatedAt: now}}},
|
||||
"remote": {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute)}}},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healBucketConfig(site.ctx, bucket, bucketTaggingConfig, info)
|
||||
case "heal-cors":
|
||||
corsPayload := base64.StdEncoding.EncodeToString([]byte(`<CORSConfiguration><CORSRule><AllowedOrigin>*</AllowedOrigin><AllowedMethod>GET</AllowedMethod></CORSRule></CORSConfiguration>`))
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute), CorsConfig: &corsPayload, CorsConfigUpdatedAt: now}}},
|
||||
"remote": {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute)}}},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healCORSMetadata(site.ctx, site.obj, bucket, info)
|
||||
case "heal-ilm-expiry-config":
|
||||
stalePeers := site.peers(peer.URL)
|
||||
for id, p := range stalePeers {
|
||||
p.ReplicateILMExpiry = false
|
||||
stalePeers[id] = p
|
||||
}
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
PeerStates: map[string]madmin.SRStateInfo{
|
||||
local: {Name: "local", Peers: peers, UpdatedAt: now},
|
||||
"remote": {Name: "remote", Peers: stalePeers, UpdatedAt: now.Add(-time.Minute)},
|
||||
},
|
||||
}
|
||||
done <- c.healILMExpiryConfig(site.ctx, site.obj, info)
|
||||
case "heal-bucket-ilm-expiry":
|
||||
lcXML := `<LifecycleConfiguration><Rule><ID>exp</ID><Status>Enabled</Status><Expiration><Days>1</Days></Expiration></Rule></LifecycleConfiguration>`
|
||||
lcPayload := base64.StdEncoding.EncodeToString([]byte(lcXML))
|
||||
info := srStatusInfo{
|
||||
Sites: site.peers(peer.URL),
|
||||
BucketStats: map[string]map[string]srBucketStatsSummary{
|
||||
bucket: {
|
||||
local: {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute), ExpiryLCConfig: &lcPayload, ExpiryLCConfigUpdatedAt: now}}},
|
||||
"remote": {HasBucket: true, meta: srBucketMetaInfo{SRBucketInfo: madmin.SRBucketInfo{CreatedAt: now.Add(-time.Minute)}}},
|
||||
},
|
||||
},
|
||||
}
|
||||
done <- c.healBucketILMExpiry(site.ctx, site.obj, bucket, info)
|
||||
}
|
||||
}()
|
||||
|
||||
select {
|
||||
case <-entered:
|
||||
case err := <-done:
|
||||
t.Fatalf("operation returned before reaching storage/peer pause point: %v", err)
|
||||
case <-time.After(10 * time.Second):
|
||||
t.Fatal("timed out waiting for storage/peer pause point")
|
||||
}
|
||||
|
||||
if c.TryLock() {
|
||||
c.state.Name = "mutated-while-in-flight"
|
||||
delete(c.state.Peers, "remote")
|
||||
c.state.UpdatedAt = now.Add(time.Second)
|
||||
c.Unlock()
|
||||
} else {
|
||||
t.Error("SiteReplicationSys lock held across storage I/O or peer RPC")
|
||||
}
|
||||
|
||||
release()
|
||||
select {
|
||||
case err := <-done:
|
||||
if err != nil {
|
||||
t.Fatalf("unexpected error: %v", err)
|
||||
}
|
||||
case <-time.After(10 * time.Second):
|
||||
t.Fatal("operation deadlocked after writer mutation")
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
func TestSiteReplicationConcurrentStateUpdates(t *testing.T) {
|
||||
site := newSvcAccSite(t)
|
||||
c := &globalSiteReplicationSys
|
||||
const bucket = "sr-concurrent-state-bucket"
|
||||
mustIAM(t, site.obj.MakeBucket(site.ctx, bucket, MakeBucketOptions{VersioningEnabled: true}))
|
||||
_, err := globalBucketMetadataSys.Update(site.ctx, bucket, bucketVersioningConfig, enabledBucketVersioningConfig)
|
||||
mustIAM(t, err)
|
||||
|
||||
peer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
if strings.Contains(r.URL.Path, "metainfo") {
|
||||
w.Header().Set("Content-Type", "application/json")
|
||||
_ = json.NewEncoder(w).Encode(madmin.SRInfo{DeploymentID: "remote"})
|
||||
return
|
||||
}
|
||||
if r.Method == http.MethodGet {
|
||||
if _, ok := r.URL.Query()["versioning"]; ok {
|
||||
w.Header().Set("Content-Type", "application/xml")
|
||||
_, _ = w.Write([]byte(`<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"><Status>Enabled</Status></VersioningConfiguration>`))
|
||||
return
|
||||
}
|
||||
if _, ok := r.URL.Query()["location"]; ok {
|
||||
w.Header().Set("Content-Type", "application/xml")
|
||||
_, _ = w.Write([]byte(`<LocationConstraint xmlns="http://s3.amazonaws.com/doc/2006-03-01/"></LocationConstraint>`))
|
||||
return
|
||||
}
|
||||
}
|
||||
w.WriteHeader(http.StatusOK)
|
||||
}))
|
||||
defer peer.Close()
|
||||
|
||||
basePeers := site.peers(peer.URL)
|
||||
now := UTCNow().Truncate(time.Second)
|
||||
c.Lock()
|
||||
c.enabled = true
|
||||
c.state = srState{
|
||||
Name: "local",
|
||||
ServiceAccountAccessKey: site.parent.AccessKey,
|
||||
Peers: maps.Clone(basePeers),
|
||||
UpdatedAt: now,
|
||||
}
|
||||
c.Unlock()
|
||||
|
||||
var wg sync.WaitGroup
|
||||
wg.Add(4)
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
for i := range 10 {
|
||||
_ = c.PeerEditReq(site.ctx, madmin.PeerInfo{
|
||||
DeploymentID: "remote",
|
||||
Name: "remote",
|
||||
Endpoint: peer.URL,
|
||||
ReplicateILMExpiry: i%2 == 0,
|
||||
})
|
||||
}
|
||||
}()
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
for i := range 10 {
|
||||
peers := site.peers(peer.URL)
|
||||
p := peers["remote"]
|
||||
p.ReplicateILMExpiry = i%2 == 1
|
||||
peers["remote"] = p
|
||||
_ = c.PeerStateEditReq(site.ctx, madmin.SRStateEditReq{
|
||||
Peers: peers,
|
||||
UpdatedAt: now.Add(time.Duration(i+1) * time.Second),
|
||||
})
|
||||
}
|
||||
}()
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
for range 10 {
|
||||
_ = c.PeerBucketConfigureReplHandler(site.ctx, bucket)
|
||||
}
|
||||
}()
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
for range 5 {
|
||||
_ = c.healBuckets(site.ctx, site.obj)
|
||||
}
|
||||
}()
|
||||
wg.Wait()
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user