admin: add plugin lock coordination

This commit is contained in:
Chris Lu committed 2026-03-02 13:29:16 -08:00
1 parent 2644816692
commit 1e0003e283
6 files changed
+165

No files matched your search

+38
View File
@@ -98,6 +98,7 @@ type AdminServer struct {
// Maintenance system
maintenanceManager *maintenance.MaintenanceManager
plugin *adminplugin.Plugin
pluginLock *AdminLockManager
// Topic retention purger
topicRetentionPurger *TopicRetentionPurger
@@ -134,6 +135,8 @@ func NewAdminServer(masters string, templateFS http.FileSystem, dataDir string,
ctx := context.Background()
go masterClient.KeepConnectedToMaster(ctx)
lockManager := NewAdminLockManager(masterClient, adminLockClientName)
server := &AdminServer{
masterClient: masterClient,
templateFS: templateFS,
@@ -145,6 +148,7 @@ func NewAdminServer(masters string, templateFS http.FileSystem, dataDir string,
collectionStatsCacheThreshold: defaultStatsCacheTimeout,
s3TablesManager: newS3TablesManager(),
icebergPort: icebergPort,
pluginLock: lockManager,
}
// Initialize topic retention purger
@@ -228,6 +232,7 @@ func NewAdminServer(masters string, templateFS http.FileSystem, dataDir string,
ClusterContextProvider: func(_ context.Context) (*plugin_pb.ClusterContext, error) {
return server.buildDefaultPluginClusterContext(), nil
},
LockManager: lockManager,
})
if err != nil && dataDir != "" {
glog.Warningf("Failed to initialize plugin with dataDir=%q: %v. Falling back to in-memory plugin state.", dataDir, err)
@@ -236,6 +241,7 @@ func NewAdminServer(masters string, templateFS http.FileSystem, dataDir string,
ClusterContextProvider: func(_ context.Context) (*plugin_pb.ClusterContext, error) {
return server.buildDefaultPluginClusterContext(), nil
},
LockManager: lockManager,
})
}
if err != nil {
@@ -889,6 +895,13 @@ func (s *AdminServer) GetPlugin() *adminplugin.Plugin {
return s.plugin
}
func (s *AdminServer) acquirePluginLock(reason string) (func(), error) {
if s == nil || s.pluginLock == nil {
return func() {}, nil
}
return s.pluginLock.Acquire(reason)
}
// RequestPluginJobTypeDescriptor asks one worker for job type schema and returns the descriptor.
func (s *AdminServer) RequestPluginJobTypeDescriptor(ctx context.Context, jobType string, forceRefresh bool) (*plugin_pb.JobTypeDescriptor, error) {
if s.plugin == nil {
@@ -931,6 +944,13 @@ func (s *AdminServer) RunPluginDetection(
if s.plugin == nil {
return nil, fmt.Errorf("plugin is not enabled")
}
releaseLock, err := s.acquirePluginLock(fmt.Sprintf("plugin detection %s", jobType))
if err != nil {
return nil, err
}
if releaseLock != nil {
defer releaseLock()
}
return s.plugin.RunDetection(ctx, jobType, clusterContext, maxResults)
}
@@ -956,6 +976,13 @@ func (s *AdminServer) RunPluginDetectionWithReport(
if s.plugin == nil {
return nil, fmt.Errorf("plugin is not enabled")
}
releaseLock, err := s.acquirePluginLock(fmt.Sprintf("plugin detection %s", jobType))
if err != nil {
return nil, err
}
if releaseLock != nil {
defer releaseLock()
}
return s.plugin.RunDetectionWithReport(ctx, jobType, clusterContext, maxResults)
}
@@ -969,6 +996,17 @@ func (s *AdminServer) ExecutePluginJob(
if s.plugin == nil {
return nil, fmt.Errorf("plugin is not enabled")
}
jobType := ""
if job != nil {
jobType = strings.TrimSpace(job.JobType)
}
releaseLock, err := s.acquirePluginLock(fmt.Sprintf("plugin execution %s", jobType))
if err != nil {
return nil, err
}
if releaseLock != nil {
defer releaseLock()
}
return s.plugin.ExecuteJob(ctx, job, clusterContext, attempt)
}