Live: stream resubmit on ds change, fix old ds settings in RunStream (#34130)

This commit is contained in:
Alexander Emelin
2021-05-18 21:39:56 +03:00
committed by GitHub
parent 18954aaa7b
commit e799257637
13 changed files with 502 additions and 85 deletions
+211 -44
View File
@@ -3,12 +3,13 @@ package runstream
import (
"context"
"errors"
"fmt"
"math"
"sync"
"time"
"github.com/grafana/grafana/pkg/infra/log"
"github.com/grafana/grafana/pkg/services/live/orgchannel"
"github.com/grafana/grafana/pkg/models"
"github.com/grafana/grafana-plugin-sdk-go/backend"
)
@@ -17,12 +18,16 @@ var (
logger = log.New("live.runstream")
)
//go:generate mockgen -destination=mock.go -package=runstream github.com/grafana/grafana/pkg/services/live/runstream StreamPacketSender,PresenceGetter,StreamRunner
//go:generate mockgen -destination=mock.go -package=runstream github.com/grafana/grafana/pkg/services/live/runstream StreamPacketSender,PresenceGetter,StreamRunner,PluginContextGetter
type StreamPacketSender interface {
Send(channel string, packet *backend.StreamPacket) error
}
type PluginContextGetter interface {
GetPluginContext(user *models.SignedInUser, pluginID string, datasourceUID string, skipCache bool) (backend.PluginContext, bool, error)
}
type PresenceGetter interface {
GetNumSubscribers(channel string) (int, error)
}
@@ -49,14 +54,18 @@ func (p *streamSender) Send(packet *backend.StreamPacket) error {
// Manager manages streams from Grafana to plugins (i.e. RunStream method).
type Manager struct {
mu sync.RWMutex
streams map[string]chan struct{}
presenceGetter PresenceGetter
packetSender StreamPacketSender
registerCh chan submitRequest
closedCh chan struct{}
checkInterval time.Duration
maxChecks int
mu sync.RWMutex
baseCtx context.Context
streams map[string]streamContext
datasourceStreams map[string]map[string]struct{}
presenceGetter PresenceGetter
pluginContextGetter PluginContextGetter
packetSender StreamPacketSender
registerCh chan submitRequest
closedCh chan struct{}
checkInterval time.Duration
maxChecks int
datasourceCheckInterval time.Duration
}
// ManagerOption modifies Manager behavior (used for tests for example).
@@ -71,20 +80,24 @@ func WithCheckConfig(interval time.Duration, maxChecks int) ManagerOption {
}
const (
defaultCheckInterval = 5 * time.Second
defaultMaxChecks = 3
defaultCheckInterval = 5 * time.Second
defaultDatasourceCheckInterval = 60 * time.Second
defaultMaxChecks = 3
)
// NewManager creates new Manager.
func NewManager(packetSender StreamPacketSender, presenceGetter PresenceGetter, opts ...ManagerOption) *Manager {
func NewManager(packetSender StreamPacketSender, presenceGetter PresenceGetter, pluginContextGetter PluginContextGetter, opts ...ManagerOption) *Manager {
sm := &Manager{
streams: make(map[string]chan struct{}),
packetSender: packetSender,
presenceGetter: presenceGetter,
registerCh: make(chan submitRequest),
closedCh: make(chan struct{}),
checkInterval: defaultCheckInterval,
maxChecks: defaultMaxChecks,
streams: make(map[string]streamContext),
datasourceStreams: map[string]map[string]struct{}{},
packetSender: packetSender,
presenceGetter: presenceGetter,
pluginContextGetter: pluginContextGetter,
registerCh: make(chan submitRequest),
closedCh: make(chan struct{}),
checkInterval: defaultCheckInterval,
maxChecks: defaultMaxChecks,
datasourceCheckInterval: defaultDatasourceCheckInterval,
}
for _, opt := range opts {
opt(sm)
@@ -92,25 +105,109 @@ func NewManager(packetSender StreamPacketSender, presenceGetter PresenceGetter,
return sm
}
func (s *Manager) HandleDatasourceDelete(orgID int64, dsUID string) error {
return s.handleDatasourceEvent(orgID, dsUID, false)
}
func (s *Manager) HandleDatasourceUpdate(orgID int64, dsUID string) error {
return s.handleDatasourceEvent(orgID, dsUID, true)
}
func (s *Manager) handleDatasourceEvent(orgID int64, dsUID string, resubmit bool) error {
dsKey := datasourceKey(orgID, dsUID)
s.mu.RLock()
dsStreams, ok := s.datasourceStreams[dsKey]
if !ok {
s.mu.RUnlock()
return nil
}
var resubmitRequests []streamRequest
var waitChannels []chan struct{}
for channel := range dsStreams {
streamCtx, ok := s.streams[channel]
if !ok {
continue
}
streamCtx.cancelFn()
waitChannels = append(waitChannels, streamCtx.CloseCh)
resubmitRequests = append(resubmitRequests, streamCtx.streamRequest)
}
s.mu.RUnlock()
// Wait for all streams to stop.
for _, ch := range waitChannels {
<-ch
}
if resubmit {
// Re-submit streams.
for _, sr := range resubmitRequests {
_, err := s.SubmitStream(s.baseCtx, sr.user, sr.Channel, sr.Path, sr.PluginContext, sr.StreamRunner, true)
if err != nil {
// Log error but do not prevent execution of caller routine.
logger.Error("Error re-submitting stream", "path", sr.Path, "error", err)
}
}
}
return nil
}
func datasourceKey(orgID int64, dsUID string) string {
return fmt.Sprintf("%d_%s", orgID, dsUID)
}
func (s *Manager) stopStream(sr streamRequest, cancelFn func()) {
s.mu.Lock()
defer s.mu.Unlock()
closeCh, ok := s.streams[sr.Channel]
streamCtx, ok := s.streams[sr.Channel]
if !ok {
return
}
closeCh := streamCtx.CloseCh
delete(s.streams, sr.Channel)
if sr.PluginContext.DataSourceInstanceSettings != nil {
dsUID := sr.PluginContext.DataSourceInstanceSettings.UID
dsKey := datasourceKey(sr.PluginContext.OrgID, dsUID)
delete(s.datasourceStreams[dsKey], sr.Channel)
}
cancelFn()
close(closeCh)
}
func (s *Manager) watchStream(ctx context.Context, cancelFn func(), sr streamRequest) {
numNoSubscribersChecks := 0
presenceTicker := time.NewTicker(s.checkInterval)
defer presenceTicker.Stop()
datasourceTicker := time.NewTicker(s.datasourceCheckInterval)
defer datasourceTicker.Stop()
for {
select {
case <-ctx.Done():
return
case <-time.After(s.checkInterval):
case <-datasourceTicker.C:
if sr.PluginContext.DataSourceInstanceSettings != nil {
dsUID := sr.PluginContext.DataSourceInstanceSettings.UID
pCtx, ok, err := s.pluginContextGetter.GetPluginContext(sr.user, sr.PluginContext.PluginID, dsUID, false)
if err != nil {
logger.Error("Error getting datasource context", "channel", sr.Channel, "path", sr.Path, "error", err)
continue
}
if !ok {
logger.Debug("Datasource not found, stop stream", "channel", sr.Channel, "path", sr.Path)
return
}
if pCtx.DataSourceInstanceSettings.Updated != sr.PluginContext.DataSourceInstanceSettings.Updated {
logger.Debug("Datasource changed, re-establish stream", "channel", sr.Channel, "path", sr.Path)
err := s.HandleDatasourceUpdate(pCtx.OrgID, dsUID)
if err != nil {
logger.Error("Error re-establishing stream", "channel", sr.Channel, "path", sr.Path, "error", err)
continue
}
return
}
}
case <-presenceTicker.C:
numSubscribers, err := s.presenceGetter.GetNumSubscribers(sr.Channel)
if err != nil {
logger.Error("Error checking num subscribers", "channel", sr.Channel, "path", sr.Path)
@@ -148,31 +245,24 @@ func getDelay(numErrors int) time.Duration {
// run stream until context canceled or stream finished without an error.
func (s *Manager) runStream(ctx context.Context, cancelFn func(), sr streamRequest) {
defer func() { s.stopStream(sr, cancelFn) }()
var numFastErrors int
var delay time.Duration
var isReconnect bool
startTime := time.Now()
for {
select {
case <-ctx.Done():
return
default:
}
startTime := time.Now()
err := sr.StreamRunner.RunStream(
ctx,
&backend.RunStreamRequest{
PluginContext: sr.PluginContext,
Path: sr.Path,
},
newStreamSender(sr.Channel, s.packetSender),
)
if err != nil {
if errors.Is(ctx.Err(), context.Canceled) {
logger.Debug("Stream cleanly finished", "path", sr.Path)
return
}
pluginCtx := sr.PluginContext
if isReconnect {
// Best effort to cool down re-establishment process. We don't have a
// nice way to understand whether we really need to wait here - so relying
// on RunStream duration time.
// on duration time of running a stream.
if time.Since(startTime) < streamDurationThreshold {
if delay < maxDelay {
// Due to not calling getDelay after we have delay larger than maxDelay
@@ -186,29 +276,85 @@ func (s *Manager) runStream(ctx context.Context, cancelFn func(), sr streamReque
delay = 0
numFastErrors = 0
}
select {
case <-ctx.Done():
return
case <-time.After(delay):
}
startTime = time.Now()
// Resolve new plugin context as it could be modified since last call.
// We are using the same user here which initiated stream originally.
var datasourceUID string
if pluginCtx.DataSourceInstanceSettings != nil {
datasourceUID = pluginCtx.DataSourceInstanceSettings.UID
}
newPluginCtx, ok, err := s.pluginContextGetter.GetPluginContext(sr.user, pluginCtx.PluginID, datasourceUID, false)
if err != nil {
logger.Error("Error getting plugin context", "path", sr.Path, "error", err)
isReconnect = true
continue
}
if !ok {
logger.Info("No plugin context found, stopping stream", "path", sr.Path)
return
}
pluginCtx = newPluginCtx
}
err := sr.StreamRunner.RunStream(
ctx,
&backend.RunStreamRequest{
PluginContext: pluginCtx,
Path: sr.Path,
},
newStreamSender(sr.Channel, s.packetSender),
)
if err != nil {
if errors.Is(ctx.Err(), context.Canceled) {
logger.Debug("Stream cleanly finished", "path", sr.Path)
return
}
logger.Error("Error running stream, re-establishing", "path", sr.Path, "error", err, "wait", delay)
time.Sleep(delay)
isReconnect = true
continue
}
logger.Debug("Stream finished without error, stopping it", "path", sr.Path)
s.stopStream(sr, cancelFn)
return
}
}
var errClosed = errors.New("stream manager closed")
type streamContext struct {
CloseCh chan struct{}
cancelFn func()
streamRequest streamRequest
}
func (s *Manager) registerStream(ctx context.Context, sr submitRequest) {
s.mu.Lock()
if closeCh, ok := s.streams[sr.streamRequest.Channel]; ok {
if streamCtx, ok := s.streams[sr.streamRequest.Channel]; ok {
s.mu.Unlock()
sr.responseCh <- submitResponse{Result: submitResult{StreamExists: true, CloseNotify: closeCh}}
sr.responseCh <- submitResponse{Result: submitResult{StreamExists: true, CloseNotify: streamCtx.CloseCh}}
return
}
ctx, cancel := context.WithCancel(ctx)
defer cancel()
closeCh := make(chan struct{})
s.streams[sr.streamRequest.Channel] = closeCh
s.streams[sr.streamRequest.Channel] = streamContext{
CloseCh: closeCh,
cancelFn: cancel,
streamRequest: sr.streamRequest,
}
if sr.streamRequest.PluginContext.DataSourceInstanceSettings != nil {
dsUID := sr.streamRequest.PluginContext.DataSourceInstanceSettings.UID
dsKey := datasourceKey(sr.streamRequest.PluginContext.OrgID, dsUID)
if _, ok := s.datasourceStreams[dsKey]; !ok {
s.datasourceStreams[dsKey] = map[string]struct{}{}
}
s.datasourceStreams[dsKey][sr.streamRequest.Channel] = struct{}{}
}
s.mu.Unlock()
sr.responseCh <- submitResponse{Result: submitResult{StreamExists: false, CloseNotify: closeCh}}
go s.watchStream(ctx, cancel, sr.streamRequest)
@@ -217,6 +363,7 @@ func (s *Manager) registerStream(ctx context.Context, sr submitRequest) {
// Run Manager till context canceled.
func (s *Manager) Run(ctx context.Context) error {
s.baseCtx = ctx
for {
select {
case sr := <-s.registerCh:
@@ -231,6 +378,7 @@ func (s *Manager) Run(ctx context.Context) error {
type streamRequest struct {
Channel string
Path string
user *models.SignedInUser
PluginContext backend.PluginContext
StreamRunner StreamRunner
}
@@ -252,13 +400,32 @@ type submitResponse struct {
Result submitResult
}
var errDatasourceNotFound = errors.New("datasource not found")
// SubmitStream submits stream handler in Manager to manage.
// The stream will be opened and kept till channel has active subscribers.
func (s *Manager) SubmitStream(ctx context.Context, orgID int64, channel string, path string, pCtx backend.PluginContext, streamRunner StreamRunner) (*submitResult, error) {
func (s *Manager) SubmitStream(ctx context.Context, user *models.SignedInUser, channel string, path string, pCtx backend.PluginContext, streamRunner StreamRunner, isResubmit bool) (*submitResult, error) {
if isResubmit {
// Resolve new plugin context as it could be modified since last call.
var datasourceUID string
if pCtx.DataSourceInstanceSettings != nil {
datasourceUID = pCtx.DataSourceInstanceSettings.UID
}
newPluginCtx, ok, err := s.pluginContextGetter.GetPluginContext(user, pCtx.PluginID, datasourceUID, false)
if err != nil {
return nil, err
}
if !ok {
return nil, errDatasourceNotFound
}
pCtx = newPluginCtx
}
req := submitRequest{
responseCh: make(chan submitResponse, 1),
streamRequest: streamRequest{
Channel: orgchannel.PrependOrgID(orgID, channel),
user: user,
Channel: channel,
Path: path,
PluginContext: pCtx,
StreamRunner: streamRunner,