refactor(coordinator): remove debug api namespace from manager (#221)

Co-authored-by: colin <102356659+colinlyguo@users.noreply.github.com>

coordinator/api_debug.go

@@ -1,96 +0,0 @@
-package coordinator
-
-import (
-	"fmt"
-	"time"
-
-	"scroll-tech/database/orm"
-)
-
-// RollerDebugAPI roller api interface in order go get debug message.
-type RollerDebugAPI interface {
-	// ListRollers returns all live rollers
-	ListRollers() ([]*RollerInfo, error)
-	// GetSessionInfo returns the session information given the session id.
-	GetSessionInfo(sessionID string) (*SessionInfo, error)
-}
-
-// RollerInfo records the roller name, pub key and active session info (id, start time).
-type RollerInfo struct {
-	Name                   string    `json:"name"`
-	Version                string    `json:"version"`
-	PublicKey              string    `json:"public_key"`
-	ActiveSession          string    `json:"active_session,omitempty"`
-	ActiveSessionStartTime time.Time `json:"active_session_start_time"` // latest proof start time.
-}
-
-// SessionInfo records proof create or proof verify failed session.
-type SessionInfo struct {
-	ID              string    `json:"id"`
-	Status          string    `json:"status"`
-	StartTime       time.Time `json:"start_time"`
-	FinishTime      time.Time `json:"finish_time,omitempty"`      // set to 0 if not finished
-	AssignedRollers []string  `json:"assigned_rollers,omitempty"` // roller name list
-	Error           string    `json:"error,omitempty"`            // empty string if no error encountered
-}
-
-// ListRollers returns all live rollers.
-func (m *Manager) ListRollers() ([]*RollerInfo, error) {
-	m.mu.RLock()
-	defer m.mu.RUnlock()
-	var res []*RollerInfo
-	for _, pk := range m.rollerPool.Keys() {
-		node, exist := m.rollerPool.Get(pk)
-		if !exist {
-			continue
-		}
-		roller := node.(*rollerNode)
-		info := &RollerInfo{
-			Name:      roller.Name,
-			Version:   roller.Version,
-			PublicKey: pk,
-		}
-		for id, sess := range m.sessions {
-			if _, ok := sess.info.Rollers[pk]; ok {
-				info.ActiveSessionStartTime = time.Unix(sess.info.StartTimestamp, 0)
-				info.ActiveSession = id
-				break
-			}
-		}
-		res = append(res, info)
-	}
-
-	return res, nil
-}
-
-func newSessionInfo(sess *session, status orm.ProvingStatus, errMsg string, finished bool) *SessionInfo {
-	now := time.Now()
-	var nameList []string
-	for pk := range sess.info.Rollers {
-		nameList = append(nameList, sess.info.Rollers[pk].Name)
-	}
-	info := SessionInfo{
-		ID:              sess.info.ID,
-		Status:          status.String(),
-		AssignedRollers: nameList,
-		StartTime:       time.Unix(sess.info.StartTimestamp, 0),
-		Error:           errMsg,
-	}
-	if finished {
-		info.FinishTime = now
-	}
-	return &info
-}
-
-// GetSessionInfo returns the session information given the session id.
-func (m *Manager) GetSessionInfo(sessionID string) (*SessionInfo, error) {
-	m.mu.RLock()
-	defer m.mu.RUnlock()
-	if info, ok := m.failedSessionInfos[sessionID]; ok {
-		return info, nil
-	}
-	if s, ok := m.sessions[sessionID]; ok {
-		return newSessionInfo(s, orm.ProvingTaskAssigned, "", false), nil
-	}
-	return nil, fmt.Errorf("no such session, sessionID: %s", sessionID)
-}

coordinator/manager.go

@@ -65,9 +65,6 @@ type Manager struct {
 	// A map containing proof failed or verify failed proof.
 	rollerPool cmap.ConcurrentMap
 
-	// TODO: once put into use, should add to graceful restart.
-	failedSessionInfos map[string]*SessionInfo
-
 	// A direct connection to the Halo2 verifier, used to verify
 	// incoming proofs.
 	verifier *verifier.Verifier
@@ -94,15 +91,14 @@ func New(ctx context.Context, cfg *config.RollerManagerConfig, orm database.OrmF
 
 	log.Info("Start coordinator successfully.")
 	return &Manager{
-		ctx:                ctx,
-		cfg:                cfg,
-		rollerPool:         cmap.New(),
-		sessions:           make(map[string]*session),
-		failedSessionInfos: make(map[string]*SessionInfo),
-		verifier:           v,
-		orm:                orm,
-		Client:             client,
-		tokenCache:         cache.New(time.Duration(cfg.TokenTimeToLive)*time.Second, 1*time.Hour),
+		ctx:        ctx,
+		cfg:        cfg,
+		rollerPool: cmap.New(),
+		sessions:   make(map[string]*session),
+		verifier:   v,
+		orm:        orm,
+		Client:     client,
+		tokenCache: cache.New(time.Duration(cfg.TokenTimeToLive)*time.Second, 1*time.Hour),
 	}, nil
 }
 
@@ -279,6 +275,9 @@ func (m *Manager) handleZkProof(pk string, msg *message.ProofDetail) error {
 			"roller pk", roller.PublicKey,
 			"error", msg.Error,
 		)
+		if dbErr = m.orm.UpdateProvingStatus(msg.ID, orm.ProvingTaskFailed); dbErr != nil {
+			log.Error("failed to update task status as failed", "error", dbErr)
+		}
 		return nil
 	}
 
@@ -346,7 +345,6 @@ func (m *Manager) CollectProofs(sess *session) {
 		if len(participatingRollers) == 0 {
 			// record failed session.
 			errMsg := "proof generation session ended without receiving any valid proofs"
-			m.addFailedSession(sess, errMsg)
 			log.Warn(errMsg, "session id", sess.info.ID)
 			// Set status as skipped.
 			// Note that this is only a workaround for testnet here.
@@ -382,11 +380,6 @@ func (m *Manager) APIs() []rpc.API {
 			Service:   RollerAPI(m),
 			Public:    true,
 		},
-		{
-			Namespace: "debug",
-			Public:    true,
-			Service:   RollerDebugAPI(m),
-		},
 	}
 }
 
@@ -506,10 +499,6 @@ func (m *Manager) IsRollerIdle(hexPk string) bool {
 	return true
 }
 
-func (m *Manager) addFailedSession(sess *session, errMsg string) {
-	m.failedSessionInfos[sess.info.ID] = newSessionInfo(sess, orm.ProvingTaskFailed, errMsg, true)
-}
-
 // VerifyToken verifies pukey for token and expiration time
 func (m *Manager) VerifyToken(authMsg *message.AuthMsg) (bool, error) {
 	pubkey, _ := authMsg.PublicKey()

coordinator/manager_test.go

@@ -318,10 +318,6 @@ func testGracefulRestart(t *testing.T) {
 	}()
 
 	for i := range ids {
-		info, err := newRollerManager.GetSessionInfo(ids[i])
-		assert.Equal(t, orm.ProvingTaskAssigned.String(), info.Status)
-		assert.NoError(t, err)
-
 		// at this point, roller haven't submitted
 		status, err := l2db.GetProvingStatusByID(ids[i])
 		assert.NoError(t, err)