Unified storage kvstore bulk import support (#113791)

* implement batchdelete in datastore

* implement bulkprocess in kv storage_backend

* convert bulkRVs to snowflake
This commit is contained in:
Will Assis
2025-12-04 09:26:50 -05:00
committed by GitHub
parent 64a3c298cf
commit ae4d2324d6
4 changed files with 326 additions and 0 deletions
+85
View File
@@ -6,10 +6,14 @@ import (
"fmt"
"io"
"net/http"
"sync"
"time"
"go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace"
"google.golang.org/grpc/metadata"
apierrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
authlib "github.com/grafana/authlib/types"
@@ -328,3 +332,84 @@ func (b *batchRunner) RollbackRequested() bool {
}
return false
}
type bulkRV struct {
max int64
counter int64
}
// Used when executing a bulk import so that we can generate snowflake RVs in the past
func newBulkRV() *bulkRV {
t := snowflakeFromTime(time.Now())
return &bulkRV{
max: t,
counter: 0,
}
}
func (x *bulkRV) next(obj metav1.Object) int64 {
ts := snowflakeFromTime(obj.GetCreationTimestamp().Time)
anno := obj.GetAnnotations()
if anno != nil {
v := anno[utils.AnnoKeyUpdatedTimestamp]
t, err := time.Parse(time.RFC3339, v)
if err == nil {
ts = snowflakeFromTime(t)
}
}
if ts > x.max || ts < 0 {
ts = x.max
}
x.counter++
return ts + x.counter
}
type BulkLock struct {
running map[string]bool
mu sync.Mutex
}
func NewBulkLock() *BulkLock {
return &BulkLock{
running: make(map[string]bool),
}
}
func (x *BulkLock) Start(keys []*resourcepb.ResourceKey) error {
x.mu.Lock()
defer x.mu.Unlock()
// First verify that it is not already running
ids := make([]string, len(keys))
for i, k := range keys {
id := NSGR(k)
if x.running[id] {
return &apierrors.StatusError{ErrStatus: metav1.Status{
Code: http.StatusPreconditionFailed,
Message: "bulk export is already running",
}}
}
ids[i] = id
}
// Then add the keys to the lock
for _, k := range ids {
x.running[k] = true
}
return nil
}
func (x *BulkLock) Finish(keys []*resourcepb.ResourceKey) {
x.mu.Lock()
defer x.mu.Unlock()
for _, k := range keys {
delete(x.running, NSGR(k))
}
}
func (x *BulkLock) Active() bool {
x.mu.Lock()
defer x.mu.Unlock()
return len(x.running) > 0
}
+21
View File
@@ -537,6 +537,27 @@ func (d *dataStore) Delete(ctx context.Context, key DataKey) error {
return d.kv.Delete(ctx, dataSection, key.String())
}
func (n *dataStore) batchDelete(ctx context.Context, keys []DataKey) error {
for len(keys) > 0 {
batch := keys
if len(batch) > dataBatchSize {
batch = batch[:dataBatchSize]
}
keys = keys[len(batch):]
stringKeys := make([]string, len(batch))
for _, dataKey := range batch {
stringKeys = append(stringKeys, dataKey.String())
}
if err := n.kv.BatchDelete(ctx, dataSection, stringKeys); err != nil {
return err
}
}
return nil
}
// ParseKey parses a string key into a DataKey struct
func ParseKey(key string) (DataKey, error) {
parts := strings.Split(key, "/")
@@ -2950,6 +2950,42 @@ func TestDataStore_getGroupResources(t *testing.T) {
}
}
func TestDataStore_BatchDelete(t *testing.T) {
ds := setupTestDataStore(t)
ctx := context.Background()
keys := make([]DataKey, 95)
for i := 0; i < 95; i++ {
rv := node.Generate().Int64()
keys[i] = DataKey{
Namespace: "test-namespace",
Group: "test-group",
Resource: "test-resource",
Name: fmt.Sprintf("test-name-%d", i),
ResourceVersion: rv,
Action: DataActionCreated,
Folder: "test-folder",
}
content := fmt.Sprintf("test-value-%d", i)
err := ds.Save(ctx, keys[i], bytes.NewReader([]byte(content)))
require.NoError(t, err)
}
err := ds.batchDelete(ctx, keys)
require.NoError(t, err)
// Verify all events were deleted
for i := 0; i < 95; i++ {
_, err := ds.Get(ctx, DataKey{
Namespace: "test-namespace",
Group: "test-group",
Resource: "test-resource",
Name: fmt.Sprintf("test-name-%d", i),
})
require.Error(t, err, "Resource should have been deleted")
}
}
func TestDataStore_BatchGet(t *testing.T) {
ds := setupTestDataStore(t)
ctx := context.Background()
@@ -18,6 +18,7 @@ import (
"github.com/prometheus/client_golang/prometheus"
"go.opentelemetry.io/otel/trace"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
"github.com/grafana/grafana/pkg/apimachinery/utils"
"github.com/grafana/grafana/pkg/storage/unified/resourcepb"
@@ -54,6 +55,7 @@ func convertEmptyToClusterNamespace(namespace string, withExperimentalClusterSco
type kvStorageBackend struct {
snowflake *snowflake.Node
kv KV
bulkLock *BulkLock
dataStore *dataStore
eventStore *eventStore
notifier *notifier
@@ -102,6 +104,7 @@ func NewKVStorageBackend(opts KVBackendOptions) (StorageBackend, error) {
backend := &kvStorageBackend{
kv: kv,
bulkLock: NewBulkLock(),
dataStore: newDataStore(kv),
eventStore: eventStore,
notifier: newNotifier(eventStore, notifierOptions{}),
@@ -1236,6 +1239,187 @@ func (k *kvStorageBackend) GetResourceLastImportTimes(ctx context.Context) iter.
}
}
func (b *kvStorageBackend) ProcessBulk(ctx context.Context, setting BulkSettings, iter BulkRequestIterator) *resourcepb.BulkResponse {
// TODO cross-node lock
err := b.bulkLock.Start(setting.Collection)
if err != nil {
return &resourcepb.BulkResponse{
Error: AsErrorResult(err),
}
}
defer b.bulkLock.Finish(setting.Collection)
bulkRvGenerator := newBulkRV()
summaries := make(map[string]*resourcepb.BulkResponse_Summary, len(setting.Collection))
rsp := &resourcepb.BulkResponse{}
if setting.RebuildCollection {
for _, key := range setting.Collection {
events := make([]string, 0)
for evtKeyStr, err := range b.eventStore.ListKeysSince(ctx, 1) {
if err != nil {
b.log.Error("failed to list event: %s", err)
return rsp
}
evtKey, err := ParseEventKey(evtKeyStr)
if err != nil {
b.log.Error("error parsing event key: %s", err)
return rsp
}
if evtKey.Group != key.Group || evtKey.Resource != key.Resource || evtKey.Namespace != key.Namespace {
continue
}
events = append(events, evtKeyStr)
}
if err := b.eventStore.batchDelete(ctx, events); err != nil {
b.log.Error("failed to delete events: %s", err)
return rsp
}
historyKeys := make([]DataKey, 0)
for dataKey, err := range b.dataStore.Keys(ctx, ListRequestKey{
Namespace: key.Namespace,
Group: key.Group,
Resource: key.Resource,
}, SortOrderAsc) {
if err != nil {
b.log.Error("failed to list collection before delete: %s", err)
return rsp
}
historyKeys = append(historyKeys, dataKey)
}
previousCount := int64(len(historyKeys))
if err := b.dataStore.batchDelete(ctx, historyKeys); err != nil {
b.log.Error("failed to delete collection: %s", err)
return rsp
}
summaries[NSGR(key)] = &resourcepb.BulkResponse_Summary{
Namespace: key.Namespace,
Group: key.Group,
Resource: key.Resource,
PreviousCount: previousCount,
}
}
} else {
for _, key := range setting.Collection {
summaries[NSGR(key)] = &resourcepb.BulkResponse_Summary{
Namespace: key.Namespace,
Group: key.Group,
Resource: key.Resource,
}
}
}
obj := &unstructured.Unstructured{}
saved := make([]DataKey, 0)
rollback := func() {
// we don't have transactions in the kv store, so we simply delete everything we created
err = b.dataStore.batchDelete(ctx, saved)
if err != nil {
b.log.Error("failed to delete during rollback: %s", err)
}
}
for iter.Next() {
if iter.RollbackRequested() {
rollback()
break
}
req := iter.Request()
if req == nil {
rollback()
rsp.Error = AsErrorResult(fmt.Errorf("missing request"))
break
}
rsp.Processed++
var action DataAction
switch resourcepb.WatchEvent_Type(req.Action) {
case resourcepb.WatchEvent_ADDED:
action = DataActionCreated
// Check if resource already exists for create operations
_, err := b.dataStore.GetLatestResourceKey(ctx, GetRequestKey{
Group: req.Key.Group,
Resource: req.Key.Resource,
Namespace: req.Key.Namespace,
Name: req.Key.Name,
})
if err == nil {
rsp.Rejected = append(rsp.Rejected, &resourcepb.BulkResponse_Rejected{
Key: req.Key,
Action: req.Action,
Error: "resource already exists",
})
continue
}
if !errors.Is(err, ErrNotFound) {
rsp.Rejected = append(rsp.Rejected, &resourcepb.BulkResponse_Rejected{
Key: req.Key,
Action: req.Action,
Error: fmt.Sprintf("failed to check if resource exists: %s", err),
})
continue
}
case resourcepb.WatchEvent_MODIFIED:
action = DataActionUpdated
case resourcepb.WatchEvent_DELETED:
action = DataActionDeleted
default:
rsp.Rejected = append(rsp.Rejected, &resourcepb.BulkResponse_Rejected{
Key: req.Key,
Action: req.Action,
Error: "invalid event type",
})
continue
}
err := obj.UnmarshalJSON(req.Value)
if err != nil {
rsp.Rejected = append(rsp.Rejected, &resourcepb.BulkResponse_Rejected{
Key: req.Key,
Action: req.Action,
Error: "unable to unmarshal json",
})
continue
}
dataKey := DataKey{
Group: req.Key.Group,
Resource: req.Key.Resource,
Namespace: req.Key.Namespace,
Name: req.Key.Name,
ResourceVersion: bulkRvGenerator.next(obj),
Action: action,
Folder: req.Folder,
}
err = b.dataStore.Save(ctx, dataKey, bytes.NewReader(req.Value))
if err != nil {
rsp.Rejected = append(rsp.Rejected, &resourcepb.BulkResponse_Rejected{
Key: req.Key,
Action: req.Action,
Error: fmt.Sprintf("failed to save resource: %s", err),
})
continue
}
saved = append(saved, dataKey)
}
// TODO update last import time
return rsp
}
// readAndClose reads all data from a ReadCloser and ensures it's closed,
// combining any errors from both operations.
func readAndClose(r io.ReadCloser) ([]byte, error) {