github.com/blixtra/nomad@v0.7.2-0.20171221000451-da9a1d7bb050/nomad/fsm.go (about)

     1  package nomad
     2  
     3  import (
     4  	"fmt"
     5  	"io"
     6  	"log"
     7  	"reflect"
     8  	"sync"
     9  	"time"
    10  
    11  	"github.com/armon/go-metrics"
    12  	memdb "github.com/hashicorp/go-memdb"
    13  	"github.com/hashicorp/nomad/helper/uuid"
    14  	"github.com/hashicorp/nomad/nomad/state"
    15  	"github.com/hashicorp/nomad/nomad/structs"
    16  	"github.com/hashicorp/nomad/scheduler"
    17  	"github.com/hashicorp/raft"
    18  	"github.com/ugorji/go/codec"
    19  )
    20  
    21  const (
    22  	// timeTableGranularity is the granularity of index to time tracking
    23  	timeTableGranularity = 5 * time.Minute
    24  
    25  	// timeTableLimit is the maximum limit of our tracking
    26  	timeTableLimit = 72 * time.Hour
    27  )
    28  
    29  // SnapshotType is prefixed to a record in the FSM snapshot
    30  // so that we can determine the type for restore
    31  type SnapshotType byte
    32  
    33  const (
    34  	NodeSnapshot SnapshotType = iota
    35  	JobSnapshot
    36  	IndexSnapshot
    37  	EvalSnapshot
    38  	AllocSnapshot
    39  	TimeTableSnapshot
    40  	PeriodicLaunchSnapshot
    41  	JobSummarySnapshot
    42  	VaultAccessorSnapshot
    43  	JobVersionSnapshot
    44  	DeploymentSnapshot
    45  	ACLPolicySnapshot
    46  	ACLTokenSnapshot
    47  )
    48  
    49  // LogApplier is the definition of a function that can apply a Raft log
    50  type LogApplier func(buf []byte, index uint64) interface{}
    51  
    52  // LogAppliers is a mapping of the Raft MessageType to the appropriate log
    53  // applier
    54  type LogAppliers map[structs.MessageType]LogApplier
    55  
    56  // SnapshotRestorer is the definition of a function that can apply a Raft log
    57  type SnapshotRestorer func(restore *state.StateRestore, dec *codec.Decoder) error
    58  
    59  // SnapshotRestorers is a mapping of the SnapshotType to the appropriate
    60  // snapshot restorer.
    61  type SnapshotRestorers map[SnapshotType]SnapshotRestorer
    62  
    63  // nomadFSM implements a finite state machine that is used
    64  // along with Raft to provide strong consistency. We implement
    65  // this outside the Server to avoid exposing this outside the package.
    66  type nomadFSM struct {
    67  	evalBroker         *EvalBroker
    68  	blockedEvals       *BlockedEvals
    69  	periodicDispatcher *PeriodicDispatch
    70  	logger             *log.Logger
    71  	state              *state.StateStore
    72  	timetable          *TimeTable
    73  
    74  	// config is the FSM config
    75  	config *FSMConfig
    76  
    77  	// enterpriseAppliers holds the set of enterprise only LogAppliers
    78  	enterpriseAppliers LogAppliers
    79  
    80  	// enterpriseRestorers holds the set of enterprise only snapshot restorers
    81  	enterpriseRestorers SnapshotRestorers
    82  
    83  	// stateLock is only used to protect outside callers to State() from
    84  	// racing with Restore(), which is called by Raft (it puts in a totally
    85  	// new state store). Everything internal here is synchronized by the
    86  	// Raft side, so doesn't need to lock this.
    87  	stateLock sync.RWMutex
    88  }
    89  
    90  // nomadSnapshot is used to provide a snapshot of the current
    91  // state in a way that can be accessed concurrently with operations
    92  // that may modify the live state.
    93  type nomadSnapshot struct {
    94  	snap      *state.StateSnapshot
    95  	timetable *TimeTable
    96  }
    97  
    98  // snapshotHeader is the first entry in our snapshot
    99  type snapshotHeader struct {
   100  }
   101  
   102  // FSMConfig is used to configure the FSM
   103  type FSMConfig struct {
   104  	// EvalBroker is the evaluation broker evaluations should be added to
   105  	EvalBroker *EvalBroker
   106  
   107  	// Periodic is the periodic job dispatcher that periodic jobs should be
   108  	// added/removed from
   109  	Periodic *PeriodicDispatch
   110  
   111  	// BlockedEvals is the blocked eval tracker that blocked evaulations should
   112  	// be added to.
   113  	Blocked *BlockedEvals
   114  
   115  	// LogOutput is the writer logs should be written to
   116  	LogOutput io.Writer
   117  
   118  	// Region is the region of the server embedding the FSM
   119  	Region string
   120  }
   121  
   122  // NewFSMPath is used to construct a new FSM with a blank state
   123  func NewFSM(config *FSMConfig) (*nomadFSM, error) {
   124  	// Create a state store
   125  	sconfig := &state.StateStoreConfig{
   126  		LogOutput: config.LogOutput,
   127  		Region:    config.Region,
   128  	}
   129  	state, err := state.NewStateStore(sconfig)
   130  	if err != nil {
   131  		return nil, err
   132  	}
   133  
   134  	fsm := &nomadFSM{
   135  		evalBroker:          config.EvalBroker,
   136  		periodicDispatcher:  config.Periodic,
   137  		blockedEvals:        config.Blocked,
   138  		logger:              log.New(config.LogOutput, "", log.LstdFlags),
   139  		config:              config,
   140  		state:               state,
   141  		timetable:           NewTimeTable(timeTableGranularity, timeTableLimit),
   142  		enterpriseAppliers:  make(map[structs.MessageType]LogApplier, 8),
   143  		enterpriseRestorers: make(map[SnapshotType]SnapshotRestorer, 8),
   144  	}
   145  
   146  	// Register all the log applier functions
   147  	fsm.registerLogAppliers()
   148  
   149  	// Register all the snapshot restorer functions
   150  	fsm.registerSnapshotRestorers()
   151  
   152  	return fsm, nil
   153  }
   154  
   155  // Close is used to cleanup resources associated with the FSM
   156  func (n *nomadFSM) Close() error {
   157  	return nil
   158  }
   159  
   160  // State is used to return a handle to the current state
   161  func (n *nomadFSM) State() *state.StateStore {
   162  	n.stateLock.RLock()
   163  	defer n.stateLock.RUnlock()
   164  	return n.state
   165  }
   166  
   167  // TimeTable returns the time table of transactions
   168  func (n *nomadFSM) TimeTable() *TimeTable {
   169  	return n.timetable
   170  }
   171  
   172  func (n *nomadFSM) Apply(log *raft.Log) interface{} {
   173  	buf := log.Data
   174  	msgType := structs.MessageType(buf[0])
   175  
   176  	// Witness this write
   177  	n.timetable.Witness(log.Index, time.Now().UTC())
   178  
   179  	// Check if this message type should be ignored when unknown. This is
   180  	// used so that new commands can be added with developer control if older
   181  	// versions can safely ignore the command, or if they should crash.
   182  	ignoreUnknown := false
   183  	if msgType&structs.IgnoreUnknownTypeFlag == structs.IgnoreUnknownTypeFlag {
   184  		msgType &= ^structs.IgnoreUnknownTypeFlag
   185  		ignoreUnknown = true
   186  	}
   187  
   188  	switch msgType {
   189  	case structs.NodeRegisterRequestType:
   190  		return n.applyUpsertNode(buf[1:], log.Index)
   191  	case structs.NodeDeregisterRequestType:
   192  		return n.applyDeregisterNode(buf[1:], log.Index)
   193  	case structs.NodeUpdateStatusRequestType:
   194  		return n.applyStatusUpdate(buf[1:], log.Index)
   195  	case structs.NodeUpdateDrainRequestType:
   196  		return n.applyDrainUpdate(buf[1:], log.Index)
   197  	case structs.JobRegisterRequestType:
   198  		return n.applyUpsertJob(buf[1:], log.Index)
   199  	case structs.JobDeregisterRequestType:
   200  		return n.applyDeregisterJob(buf[1:], log.Index)
   201  	case structs.EvalUpdateRequestType:
   202  		return n.applyUpdateEval(buf[1:], log.Index)
   203  	case structs.EvalDeleteRequestType:
   204  		return n.applyDeleteEval(buf[1:], log.Index)
   205  	case structs.AllocUpdateRequestType:
   206  		return n.applyAllocUpdate(buf[1:], log.Index)
   207  	case structs.AllocClientUpdateRequestType:
   208  		return n.applyAllocClientUpdate(buf[1:], log.Index)
   209  	case structs.ReconcileJobSummariesRequestType:
   210  		return n.applyReconcileSummaries(buf[1:], log.Index)
   211  	case structs.VaultAccessorRegisterRequestType:
   212  		return n.applyUpsertVaultAccessor(buf[1:], log.Index)
   213  	case structs.VaultAccessorDegisterRequestType:
   214  		return n.applyDeregisterVaultAccessor(buf[1:], log.Index)
   215  	case structs.ApplyPlanResultsRequestType:
   216  		return n.applyPlanResults(buf[1:], log.Index)
   217  	case structs.DeploymentStatusUpdateRequestType:
   218  		return n.applyDeploymentStatusUpdate(buf[1:], log.Index)
   219  	case structs.DeploymentPromoteRequestType:
   220  		return n.applyDeploymentPromotion(buf[1:], log.Index)
   221  	case structs.DeploymentAllocHealthRequestType:
   222  		return n.applyDeploymentAllocHealth(buf[1:], log.Index)
   223  	case structs.DeploymentDeleteRequestType:
   224  		return n.applyDeploymentDelete(buf[1:], log.Index)
   225  	case structs.JobStabilityRequestType:
   226  		return n.applyJobStability(buf[1:], log.Index)
   227  	case structs.ACLPolicyUpsertRequestType:
   228  		return n.applyACLPolicyUpsert(buf[1:], log.Index)
   229  	case structs.ACLPolicyDeleteRequestType:
   230  		return n.applyACLPolicyDelete(buf[1:], log.Index)
   231  	case structs.ACLTokenUpsertRequestType:
   232  		return n.applyACLTokenUpsert(buf[1:], log.Index)
   233  	case structs.ACLTokenDeleteRequestType:
   234  		return n.applyACLTokenDelete(buf[1:], log.Index)
   235  	case structs.ACLTokenBootstrapRequestType:
   236  		return n.applyACLTokenBootstrap(buf[1:], log.Index)
   237  	}
   238  
   239  	// Check enterprise only message types.
   240  	if applier, ok := n.enterpriseAppliers[msgType]; ok {
   241  		return applier(buf[1:], log.Index)
   242  	}
   243  
   244  	// We didn't match anything, either panic or ignore
   245  	if ignoreUnknown {
   246  		n.logger.Printf("[WARN] nomad.fsm: ignoring unknown message type (%d), upgrade to newer version", msgType)
   247  		return nil
   248  	}
   249  
   250  	panic(fmt.Errorf("failed to apply request: %#v", buf))
   251  }
   252  
   253  func (n *nomadFSM) applyUpsertNode(buf []byte, index uint64) interface{} {
   254  	defer metrics.MeasureSince([]string{"nomad", "fsm", "register_node"}, time.Now())
   255  	var req structs.NodeRegisterRequest
   256  	if err := structs.Decode(buf, &req); err != nil {
   257  		panic(fmt.Errorf("failed to decode request: %v", err))
   258  	}
   259  
   260  	if err := n.state.UpsertNode(index, req.Node); err != nil {
   261  		n.logger.Printf("[ERR] nomad.fsm: UpsertNode failed: %v", err)
   262  		return err
   263  	}
   264  
   265  	// Unblock evals for the nodes computed node class if it is in a ready
   266  	// state.
   267  	if req.Node.Status == structs.NodeStatusReady {
   268  		n.blockedEvals.Unblock(req.Node.ComputedClass, index)
   269  	}
   270  
   271  	return nil
   272  }
   273  
   274  func (n *nomadFSM) applyDeregisterNode(buf []byte, index uint64) interface{} {
   275  	defer metrics.MeasureSince([]string{"nomad", "fsm", "deregister_node"}, time.Now())
   276  	var req structs.NodeDeregisterRequest
   277  	if err := structs.Decode(buf, &req); err != nil {
   278  		panic(fmt.Errorf("failed to decode request: %v", err))
   279  	}
   280  
   281  	if err := n.state.DeleteNode(index, req.NodeID); err != nil {
   282  		n.logger.Printf("[ERR] nomad.fsm: DeleteNode failed: %v", err)
   283  		return err
   284  	}
   285  	return nil
   286  }
   287  
   288  func (n *nomadFSM) applyStatusUpdate(buf []byte, index uint64) interface{} {
   289  	defer metrics.MeasureSince([]string{"nomad", "fsm", "node_status_update"}, time.Now())
   290  	var req structs.NodeUpdateStatusRequest
   291  	if err := structs.Decode(buf, &req); err != nil {
   292  		panic(fmt.Errorf("failed to decode request: %v", err))
   293  	}
   294  
   295  	if err := n.state.UpdateNodeStatus(index, req.NodeID, req.Status); err != nil {
   296  		n.logger.Printf("[ERR] nomad.fsm: UpdateNodeStatus failed: %v", err)
   297  		return err
   298  	}
   299  
   300  	// Unblock evals for the nodes computed node class if it is in a ready
   301  	// state.
   302  	if req.Status == structs.NodeStatusReady {
   303  		ws := memdb.NewWatchSet()
   304  		node, err := n.state.NodeByID(ws, req.NodeID)
   305  		if err != nil {
   306  			n.logger.Printf("[ERR] nomad.fsm: looking up node %q failed: %v", req.NodeID, err)
   307  			return err
   308  
   309  		}
   310  		n.blockedEvals.Unblock(node.ComputedClass, index)
   311  	}
   312  
   313  	return nil
   314  }
   315  
   316  func (n *nomadFSM) applyDrainUpdate(buf []byte, index uint64) interface{} {
   317  	defer metrics.MeasureSince([]string{"nomad", "fsm", "node_drain_update"}, time.Now())
   318  	var req structs.NodeUpdateDrainRequest
   319  	if err := structs.Decode(buf, &req); err != nil {
   320  		panic(fmt.Errorf("failed to decode request: %v", err))
   321  	}
   322  
   323  	if err := n.state.UpdateNodeDrain(index, req.NodeID, req.Drain); err != nil {
   324  		n.logger.Printf("[ERR] nomad.fsm: UpdateNodeDrain failed: %v", err)
   325  		return err
   326  	}
   327  	return nil
   328  }
   329  
   330  func (n *nomadFSM) applyUpsertJob(buf []byte, index uint64) interface{} {
   331  	defer metrics.MeasureSince([]string{"nomad", "fsm", "register_job"}, time.Now())
   332  	var req structs.JobRegisterRequest
   333  	if err := structs.Decode(buf, &req); err != nil {
   334  		panic(fmt.Errorf("failed to decode request: %v", err))
   335  	}
   336  
   337  	/* Handle upgrade paths:
   338  	 * - Empty maps and slices should be treated as nil to avoid
   339  	 *   un-intended destructive updates in scheduler since we use
   340  	 *   reflect.DeepEqual. Starting Nomad 0.4.1, job submission sanatizes
   341  	 *   the incoming job.
   342  	 * - Migrate from old style upgrade stanza that used only a stagger.
   343  	 */
   344  	req.Job.Canonicalize()
   345  
   346  	if err := n.state.UpsertJob(index, req.Job); err != nil {
   347  		n.logger.Printf("[ERR] nomad.fsm: UpsertJob failed: %v", err)
   348  		return err
   349  	}
   350  
   351  	// We always add the job to the periodic dispatcher because there is the
   352  	// possibility that the periodic spec was removed and then we should stop
   353  	// tracking it.
   354  	if err := n.periodicDispatcher.Add(req.Job); err != nil {
   355  		n.logger.Printf("[ERR] nomad.fsm: periodicDispatcher.Add failed: %v", err)
   356  		return err
   357  	}
   358  
   359  	// Create a watch set
   360  	ws := memdb.NewWatchSet()
   361  
   362  	// If it is an active periodic job, record the time it was inserted. This is
   363  	// necessary for recovering during leader election. It is possible that from
   364  	// the time it is added to when it was suppose to launch, leader election
   365  	// occurs and the job was not launched. In this case, we use the insertion
   366  	// time to determine if a launch was missed.
   367  	if req.Job.IsPeriodicActive() {
   368  		prevLaunch, err := n.state.PeriodicLaunchByID(ws, req.Namespace, req.Job.ID)
   369  		if err != nil {
   370  			n.logger.Printf("[ERR] nomad.fsm: PeriodicLaunchByID failed: %v", err)
   371  			return err
   372  		}
   373  
   374  		// Record the insertion time as a launch. We overload the launch table
   375  		// such that the first entry is the insertion time.
   376  		if prevLaunch == nil {
   377  			launch := &structs.PeriodicLaunch{
   378  				ID:        req.Job.ID,
   379  				Namespace: req.Namespace,
   380  				Launch:    time.Now(),
   381  			}
   382  			if err := n.state.UpsertPeriodicLaunch(index, launch); err != nil {
   383  				n.logger.Printf("[ERR] nomad.fsm: UpsertPeriodicLaunch failed: %v", err)
   384  				return err
   385  			}
   386  		}
   387  	}
   388  
   389  	// Check if the parent job is periodic and mark the launch time.
   390  	parentID := req.Job.ParentID
   391  	if parentID != "" {
   392  		parent, err := n.state.JobByID(ws, req.Namespace, parentID)
   393  		if err != nil {
   394  			n.logger.Printf("[ERR] nomad.fsm: JobByID(%v) lookup for parent failed: %v", parentID, err)
   395  			return err
   396  		} else if parent == nil {
   397  			// The parent has been deregistered.
   398  			return nil
   399  		}
   400  
   401  		if parent.IsPeriodic() && !parent.IsParameterized() {
   402  			t, err := n.periodicDispatcher.LaunchTime(req.Job.ID)
   403  			if err != nil {
   404  				n.logger.Printf("[ERR] nomad.fsm: LaunchTime(%v) failed: %v", req.Job.ID, err)
   405  				return err
   406  			}
   407  
   408  			launch := &structs.PeriodicLaunch{
   409  				ID:        parentID,
   410  				Namespace: req.Namespace,
   411  				Launch:    t,
   412  			}
   413  			if err := n.state.UpsertPeriodicLaunch(index, launch); err != nil {
   414  				n.logger.Printf("[ERR] nomad.fsm: UpsertPeriodicLaunch failed: %v", err)
   415  				return err
   416  			}
   417  		}
   418  	}
   419  
   420  	return nil
   421  }
   422  
   423  func (n *nomadFSM) applyDeregisterJob(buf []byte, index uint64) interface{} {
   424  	defer metrics.MeasureSince([]string{"nomad", "fsm", "deregister_job"}, time.Now())
   425  	var req structs.JobDeregisterRequest
   426  	if err := structs.Decode(buf, &req); err != nil {
   427  		panic(fmt.Errorf("failed to decode request: %v", err))
   428  	}
   429  
   430  	// If it is periodic remove it from the dispatcher
   431  	if err := n.periodicDispatcher.Remove(req.Namespace, req.JobID); err != nil {
   432  		n.logger.Printf("[ERR] nomad.fsm: periodicDispatcher.Remove failed: %v", err)
   433  		return err
   434  	}
   435  
   436  	if req.Purge {
   437  		if err := n.state.DeleteJob(index, req.Namespace, req.JobID); err != nil {
   438  			n.logger.Printf("[ERR] nomad.fsm: DeleteJob failed: %v", err)
   439  			return err
   440  		}
   441  
   442  		// We always delete from the periodic launch table because it is possible that
   443  		// the job was updated to be non-perioidic, thus checking if it is periodic
   444  		// doesn't ensure we clean it up properly.
   445  		n.state.DeletePeriodicLaunch(index, req.Namespace, req.JobID)
   446  	} else {
   447  		// Get the current job and mark it as stopped and re-insert it.
   448  		ws := memdb.NewWatchSet()
   449  		current, err := n.state.JobByID(ws, req.Namespace, req.JobID)
   450  		if err != nil {
   451  			n.logger.Printf("[ERR] nomad.fsm: JobByID lookup failed: %v", err)
   452  			return err
   453  		}
   454  
   455  		if current == nil {
   456  			return fmt.Errorf("job %q in namespace %q doesn't exist to be deregistered", req.JobID, req.Namespace)
   457  		}
   458  
   459  		stopped := current.Copy()
   460  		stopped.Stop = true
   461  
   462  		if err := n.state.UpsertJob(index, stopped); err != nil {
   463  			n.logger.Printf("[ERR] nomad.fsm: UpsertJob failed: %v", err)
   464  			return err
   465  		}
   466  	}
   467  
   468  	return nil
   469  }
   470  
   471  func (n *nomadFSM) applyUpdateEval(buf []byte, index uint64) interface{} {
   472  	defer metrics.MeasureSince([]string{"nomad", "fsm", "update_eval"}, time.Now())
   473  	var req structs.EvalUpdateRequest
   474  	if err := structs.Decode(buf, &req); err != nil {
   475  		panic(fmt.Errorf("failed to decode request: %v", err))
   476  	}
   477  
   478  	if err := n.state.UpsertEvals(index, req.Evals); err != nil {
   479  		n.logger.Printf("[ERR] nomad.fsm: UpsertEvals failed: %v", err)
   480  		return err
   481  	}
   482  
   483  	for _, eval := range req.Evals {
   484  		if eval.ShouldEnqueue() {
   485  			n.evalBroker.Enqueue(eval)
   486  		} else if eval.ShouldBlock() {
   487  			n.blockedEvals.Block(eval)
   488  		} else if eval.Status == structs.EvalStatusComplete &&
   489  			len(eval.FailedTGAllocs) == 0 {
   490  			// If we have a successful evaluation for a node, untrack any
   491  			// blocked evaluation
   492  			n.blockedEvals.Untrack(eval.JobID)
   493  		}
   494  	}
   495  	return nil
   496  }
   497  
   498  func (n *nomadFSM) applyDeleteEval(buf []byte, index uint64) interface{} {
   499  	defer metrics.MeasureSince([]string{"nomad", "fsm", "delete_eval"}, time.Now())
   500  	var req structs.EvalDeleteRequest
   501  	if err := structs.Decode(buf, &req); err != nil {
   502  		panic(fmt.Errorf("failed to decode request: %v", err))
   503  	}
   504  
   505  	if err := n.state.DeleteEval(index, req.Evals, req.Allocs); err != nil {
   506  		n.logger.Printf("[ERR] nomad.fsm: DeleteEval failed: %v", err)
   507  		return err
   508  	}
   509  	return nil
   510  }
   511  
   512  func (n *nomadFSM) applyAllocUpdate(buf []byte, index uint64) interface{} {
   513  	defer metrics.MeasureSince([]string{"nomad", "fsm", "alloc_update"}, time.Now())
   514  	var req structs.AllocUpdateRequest
   515  	if err := structs.Decode(buf, &req); err != nil {
   516  		panic(fmt.Errorf("failed to decode request: %v", err))
   517  	}
   518  
   519  	// Attach the job to all the allocations. It is pulled out in the
   520  	// payload to avoid the redundancy of encoding, but should be denormalized
   521  	// prior to being inserted into MemDB.
   522  	structs.DenormalizeAllocationJobs(req.Job, req.Alloc)
   523  
   524  	// Calculate the total resources of allocations. It is pulled out in the
   525  	// payload to avoid encoding something that can be computed, but should be
   526  	// denormalized prior to being inserted into MemDB.
   527  	for _, alloc := range req.Alloc {
   528  		if alloc.Resources != nil {
   529  			// COMPAT 0.4.1 -> 0.5
   530  			// Set the shared resources for allocations which don't have them
   531  			if alloc.SharedResources == nil {
   532  				alloc.SharedResources = &structs.Resources{
   533  					DiskMB: alloc.Resources.DiskMB,
   534  				}
   535  			}
   536  
   537  			continue
   538  		}
   539  
   540  		alloc.Resources = new(structs.Resources)
   541  		for _, task := range alloc.TaskResources {
   542  			alloc.Resources.Add(task)
   543  		}
   544  
   545  		// Add the shared resources
   546  		alloc.Resources.Add(alloc.SharedResources)
   547  	}
   548  
   549  	if err := n.state.UpsertAllocs(index, req.Alloc); err != nil {
   550  		n.logger.Printf("[ERR] nomad.fsm: UpsertAllocs failed: %v", err)
   551  		return err
   552  	}
   553  	return nil
   554  }
   555  
   556  func (n *nomadFSM) applyAllocClientUpdate(buf []byte, index uint64) interface{} {
   557  	defer metrics.MeasureSince([]string{"nomad", "fsm", "alloc_client_update"}, time.Now())
   558  	var req structs.AllocUpdateRequest
   559  	if err := structs.Decode(buf, &req); err != nil {
   560  		panic(fmt.Errorf("failed to decode request: %v", err))
   561  	}
   562  	if len(req.Alloc) == 0 {
   563  		return nil
   564  	}
   565  
   566  	// Create a watch set
   567  	ws := memdb.NewWatchSet()
   568  
   569  	// Updating the allocs with the job id and task group name
   570  	for _, alloc := range req.Alloc {
   571  		if existing, _ := n.state.AllocByID(ws, alloc.ID); existing != nil {
   572  			alloc.JobID = existing.JobID
   573  			alloc.TaskGroup = existing.TaskGroup
   574  		}
   575  	}
   576  
   577  	// Update all the client allocations
   578  	if err := n.state.UpdateAllocsFromClient(index, req.Alloc); err != nil {
   579  		n.logger.Printf("[ERR] nomad.fsm: UpdateAllocFromClient failed: %v", err)
   580  		return err
   581  	}
   582  
   583  	// Unblock evals for the nodes computed node class if the client has
   584  	// finished running an allocation.
   585  	for _, alloc := range req.Alloc {
   586  		if alloc.ClientStatus == structs.AllocClientStatusComplete ||
   587  			alloc.ClientStatus == structs.AllocClientStatusFailed {
   588  			nodeID := alloc.NodeID
   589  			node, err := n.state.NodeByID(ws, nodeID)
   590  			if err != nil || node == nil {
   591  				n.logger.Printf("[ERR] nomad.fsm: looking up node %q failed: %v", nodeID, err)
   592  				return err
   593  
   594  			}
   595  
   596  			// Unblock any associated quota
   597  			quota, err := n.allocQuota(alloc.ID)
   598  			if err != nil {
   599  				n.logger.Printf("[ERR] nomad.fsm: looking up quota associated with alloc %q failed: %v", alloc.ID, err)
   600  				return err
   601  			}
   602  
   603  			n.blockedEvals.UnblockClassAndQuota(node.ComputedClass, quota, index)
   604  		}
   605  	}
   606  
   607  	return nil
   608  }
   609  
   610  // applyReconcileSummaries reconciles summaries for all the jobs
   611  func (n *nomadFSM) applyReconcileSummaries(buf []byte, index uint64) interface{} {
   612  	if err := n.state.ReconcileJobSummaries(index); err != nil {
   613  		return err
   614  	}
   615  	return n.reconcileQueuedAllocations(index)
   616  }
   617  
   618  // applyUpsertVaultAccessor stores the Vault accessors for a given allocation
   619  // and task
   620  func (n *nomadFSM) applyUpsertVaultAccessor(buf []byte, index uint64) interface{} {
   621  	defer metrics.MeasureSince([]string{"nomad", "fsm", "upsert_vault_accessor"}, time.Now())
   622  	var req structs.VaultAccessorsRequest
   623  	if err := structs.Decode(buf, &req); err != nil {
   624  		panic(fmt.Errorf("failed to decode request: %v", err))
   625  	}
   626  
   627  	if err := n.state.UpsertVaultAccessor(index, req.Accessors); err != nil {
   628  		n.logger.Printf("[ERR] nomad.fsm: UpsertVaultAccessor failed: %v", err)
   629  		return err
   630  	}
   631  
   632  	return nil
   633  }
   634  
   635  // applyDeregisterVaultAccessor deregisters a set of Vault accessors
   636  func (n *nomadFSM) applyDeregisterVaultAccessor(buf []byte, index uint64) interface{} {
   637  	defer metrics.MeasureSince([]string{"nomad", "fsm", "deregister_vault_accessor"}, time.Now())
   638  	var req structs.VaultAccessorsRequest
   639  	if err := structs.Decode(buf, &req); err != nil {
   640  		panic(fmt.Errorf("failed to decode request: %v", err))
   641  	}
   642  
   643  	if err := n.state.DeleteVaultAccessors(index, req.Accessors); err != nil {
   644  		n.logger.Printf("[ERR] nomad.fsm: DeregisterVaultAccessor failed: %v", err)
   645  		return err
   646  	}
   647  
   648  	return nil
   649  }
   650  
   651  // applyPlanApply applies the results of a plan application
   652  func (n *nomadFSM) applyPlanResults(buf []byte, index uint64) interface{} {
   653  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_plan_results"}, time.Now())
   654  	var req structs.ApplyPlanResultsRequest
   655  	if err := structs.Decode(buf, &req); err != nil {
   656  		panic(fmt.Errorf("failed to decode request: %v", err))
   657  	}
   658  
   659  	if err := n.state.UpsertPlanResults(index, &req); err != nil {
   660  		n.logger.Printf("[ERR] nomad.fsm: ApplyPlan failed: %v", err)
   661  		return err
   662  	}
   663  
   664  	return nil
   665  }
   666  
   667  // applyDeploymentStatusUpdate is used to update the status of an existing
   668  // deployment
   669  func (n *nomadFSM) applyDeploymentStatusUpdate(buf []byte, index uint64) interface{} {
   670  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_deployment_status_update"}, time.Now())
   671  	var req structs.DeploymentStatusUpdateRequest
   672  	if err := structs.Decode(buf, &req); err != nil {
   673  		panic(fmt.Errorf("failed to decode request: %v", err))
   674  	}
   675  
   676  	if err := n.state.UpdateDeploymentStatus(index, &req); err != nil {
   677  		n.logger.Printf("[ERR] nomad.fsm: UpsertDeploymentStatusUpdate failed: %v", err)
   678  		return err
   679  	}
   680  
   681  	if req.Eval != nil && req.Eval.ShouldEnqueue() {
   682  		n.evalBroker.Enqueue(req.Eval)
   683  	}
   684  
   685  	return nil
   686  }
   687  
   688  // applyDeploymentPromotion is used to promote canaries in a deployment
   689  func (n *nomadFSM) applyDeploymentPromotion(buf []byte, index uint64) interface{} {
   690  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_deployment_promotion"}, time.Now())
   691  	var req structs.ApplyDeploymentPromoteRequest
   692  	if err := structs.Decode(buf, &req); err != nil {
   693  		panic(fmt.Errorf("failed to decode request: %v", err))
   694  	}
   695  
   696  	if err := n.state.UpdateDeploymentPromotion(index, &req); err != nil {
   697  		n.logger.Printf("[ERR] nomad.fsm: UpsertDeploymentPromotion failed: %v", err)
   698  		return err
   699  	}
   700  
   701  	if req.Eval != nil && req.Eval.ShouldEnqueue() {
   702  		n.evalBroker.Enqueue(req.Eval)
   703  	}
   704  
   705  	return nil
   706  }
   707  
   708  // applyDeploymentAllocHealth is used to set the health of allocations as part
   709  // of a deployment
   710  func (n *nomadFSM) applyDeploymentAllocHealth(buf []byte, index uint64) interface{} {
   711  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_deployment_alloc_health"}, time.Now())
   712  	var req structs.ApplyDeploymentAllocHealthRequest
   713  	if err := structs.Decode(buf, &req); err != nil {
   714  		panic(fmt.Errorf("failed to decode request: %v", err))
   715  	}
   716  
   717  	if err := n.state.UpdateDeploymentAllocHealth(index, &req); err != nil {
   718  		n.logger.Printf("[ERR] nomad.fsm: UpsertDeploymentAllocHealth failed: %v", err)
   719  		return err
   720  	}
   721  
   722  	if req.Eval != nil && req.Eval.ShouldEnqueue() {
   723  		n.evalBroker.Enqueue(req.Eval)
   724  	}
   725  
   726  	return nil
   727  }
   728  
   729  // applyDeploymentDelete is used to delete a set of deployments
   730  func (n *nomadFSM) applyDeploymentDelete(buf []byte, index uint64) interface{} {
   731  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_deployment_delete"}, time.Now())
   732  	var req structs.DeploymentDeleteRequest
   733  	if err := structs.Decode(buf, &req); err != nil {
   734  		panic(fmt.Errorf("failed to decode request: %v", err))
   735  	}
   736  
   737  	if err := n.state.DeleteDeployment(index, req.Deployments); err != nil {
   738  		n.logger.Printf("[ERR] nomad.fsm: DeleteDeployment failed: %v", err)
   739  		return err
   740  	}
   741  
   742  	return nil
   743  }
   744  
   745  // applyJobStability is used to set the stability of a job
   746  func (n *nomadFSM) applyJobStability(buf []byte, index uint64) interface{} {
   747  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_job_stability"}, time.Now())
   748  	var req structs.JobStabilityRequest
   749  	if err := structs.Decode(buf, &req); err != nil {
   750  		panic(fmt.Errorf("failed to decode request: %v", err))
   751  	}
   752  
   753  	if err := n.state.UpdateJobStability(index, req.Namespace, req.JobID, req.JobVersion, req.Stable); err != nil {
   754  		n.logger.Printf("[ERR] nomad.fsm: UpdateJobStability failed: %v", err)
   755  		return err
   756  	}
   757  
   758  	return nil
   759  }
   760  
   761  // applyACLPolicyUpsert is used to upsert a set of policies
   762  func (n *nomadFSM) applyACLPolicyUpsert(buf []byte, index uint64) interface{} {
   763  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_acl_policy_upsert"}, time.Now())
   764  	var req structs.ACLPolicyUpsertRequest
   765  	if err := structs.Decode(buf, &req); err != nil {
   766  		panic(fmt.Errorf("failed to decode request: %v", err))
   767  	}
   768  
   769  	if err := n.state.UpsertACLPolicies(index, req.Policies); err != nil {
   770  		n.logger.Printf("[ERR] nomad.fsm: UpsertACLPolicies failed: %v", err)
   771  		return err
   772  	}
   773  	return nil
   774  }
   775  
   776  // applyACLPolicyDelete is used to delete a set of policies
   777  func (n *nomadFSM) applyACLPolicyDelete(buf []byte, index uint64) interface{} {
   778  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_acl_policy_delete"}, time.Now())
   779  	var req structs.ACLPolicyDeleteRequest
   780  	if err := structs.Decode(buf, &req); err != nil {
   781  		panic(fmt.Errorf("failed to decode request: %v", err))
   782  	}
   783  
   784  	if err := n.state.DeleteACLPolicies(index, req.Names); err != nil {
   785  		n.logger.Printf("[ERR] nomad.fsm: DeleteACLPolicies failed: %v", err)
   786  		return err
   787  	}
   788  	return nil
   789  }
   790  
   791  // applyACLTokenUpsert is used to upsert a set of policies
   792  func (n *nomadFSM) applyACLTokenUpsert(buf []byte, index uint64) interface{} {
   793  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_acl_token_upsert"}, time.Now())
   794  	var req structs.ACLTokenUpsertRequest
   795  	if err := structs.Decode(buf, &req); err != nil {
   796  		panic(fmt.Errorf("failed to decode request: %v", err))
   797  	}
   798  
   799  	if err := n.state.UpsertACLTokens(index, req.Tokens); err != nil {
   800  		n.logger.Printf("[ERR] nomad.fsm: UpsertACLTokens failed: %v", err)
   801  		return err
   802  	}
   803  	return nil
   804  }
   805  
   806  // applyACLTokenDelete is used to delete a set of policies
   807  func (n *nomadFSM) applyACLTokenDelete(buf []byte, index uint64) interface{} {
   808  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_acl_token_delete"}, time.Now())
   809  	var req structs.ACLTokenDeleteRequest
   810  	if err := structs.Decode(buf, &req); err != nil {
   811  		panic(fmt.Errorf("failed to decode request: %v", err))
   812  	}
   813  
   814  	if err := n.state.DeleteACLTokens(index, req.AccessorIDs); err != nil {
   815  		n.logger.Printf("[ERR] nomad.fsm: DeleteACLTokens failed: %v", err)
   816  		return err
   817  	}
   818  	return nil
   819  }
   820  
   821  // applyACLTokenBootstrap is used to bootstrap an ACL token
   822  func (n *nomadFSM) applyACLTokenBootstrap(buf []byte, index uint64) interface{} {
   823  	defer metrics.MeasureSince([]string{"nomad", "fsm", "apply_acl_token_bootstrap"}, time.Now())
   824  	var req structs.ACLTokenBootstrapRequest
   825  	if err := structs.Decode(buf, &req); err != nil {
   826  		panic(fmt.Errorf("failed to decode request: %v", err))
   827  	}
   828  
   829  	if err := n.state.BootstrapACLTokens(index, req.ResetIndex, req.Token); err != nil {
   830  		n.logger.Printf("[ERR] nomad.fsm: BootstrapACLToken failed: %v", err)
   831  		return err
   832  	}
   833  	return nil
   834  }
   835  
   836  func (n *nomadFSM) Snapshot() (raft.FSMSnapshot, error) {
   837  	// Create a new snapshot
   838  	snap, err := n.state.Snapshot()
   839  	if err != nil {
   840  		return nil, err
   841  	}
   842  
   843  	ns := &nomadSnapshot{
   844  		snap:      snap,
   845  		timetable: n.timetable,
   846  	}
   847  	return ns, nil
   848  }
   849  
   850  func (n *nomadFSM) Restore(old io.ReadCloser) error {
   851  	defer old.Close()
   852  
   853  	// Create a new state store
   854  	config := &state.StateStoreConfig{
   855  		LogOutput: n.config.LogOutput,
   856  		Region:    n.config.Region,
   857  	}
   858  	newState, err := state.NewStateStore(config)
   859  	if err != nil {
   860  		return err
   861  	}
   862  
   863  	// Start the state restore
   864  	restore, err := newState.Restore()
   865  	if err != nil {
   866  		return err
   867  	}
   868  	defer restore.Abort()
   869  
   870  	// Create a decoder
   871  	dec := codec.NewDecoder(old, structs.MsgpackHandle)
   872  
   873  	// Read in the header
   874  	var header snapshotHeader
   875  	if err := dec.Decode(&header); err != nil {
   876  		return err
   877  	}
   878  
   879  	// Populate the new state
   880  	msgType := make([]byte, 1)
   881  	for {
   882  		// Read the message type
   883  		_, err := old.Read(msgType)
   884  		if err == io.EOF {
   885  			break
   886  		} else if err != nil {
   887  			return err
   888  		}
   889  
   890  		// Decode
   891  		snapType := SnapshotType(msgType[0])
   892  		switch snapType {
   893  		case TimeTableSnapshot:
   894  			if err := n.timetable.Deserialize(dec); err != nil {
   895  				return fmt.Errorf("time table deserialize failed: %v", err)
   896  			}
   897  
   898  		case NodeSnapshot:
   899  			node := new(structs.Node)
   900  			if err := dec.Decode(node); err != nil {
   901  				return err
   902  			}
   903  			if err := restore.NodeRestore(node); err != nil {
   904  				return err
   905  			}
   906  
   907  		case JobSnapshot:
   908  			job := new(structs.Job)
   909  			if err := dec.Decode(job); err != nil {
   910  				return err
   911  			}
   912  
   913  			/* Handle upgrade paths:
   914  			 * - Empty maps and slices should be treated as nil to avoid
   915  			 *   un-intended destructive updates in scheduler since we use
   916  			 *   reflect.DeepEqual. Starting Nomad 0.4.1, job submission sanatizes
   917  			 *   the incoming job.
   918  			 * - Migrate from old style upgrade stanza that used only a stagger.
   919  			 */
   920  			job.Canonicalize()
   921  
   922  			if err := restore.JobRestore(job); err != nil {
   923  				return err
   924  			}
   925  
   926  		case EvalSnapshot:
   927  			eval := new(structs.Evaluation)
   928  			if err := dec.Decode(eval); err != nil {
   929  				return err
   930  			}
   931  
   932  			// COMPAT: Handle upgrade to v0.7.0
   933  			if eval.Namespace == "" {
   934  				eval.Namespace = structs.DefaultNamespace
   935  			}
   936  
   937  			if err := restore.EvalRestore(eval); err != nil {
   938  				return err
   939  			}
   940  
   941  		case AllocSnapshot:
   942  			alloc := new(structs.Allocation)
   943  			if err := dec.Decode(alloc); err != nil {
   944  				return err
   945  			}
   946  
   947  			// COMPAT: Handle upgrade to v0.7.0
   948  			if alloc.Namespace == "" {
   949  				alloc.Namespace = structs.DefaultNamespace
   950  			}
   951  
   952  			if err := restore.AllocRestore(alloc); err != nil {
   953  				return err
   954  			}
   955  
   956  		case IndexSnapshot:
   957  			idx := new(state.IndexEntry)
   958  			if err := dec.Decode(idx); err != nil {
   959  				return err
   960  			}
   961  			if err := restore.IndexRestore(idx); err != nil {
   962  				return err
   963  			}
   964  
   965  		case PeriodicLaunchSnapshot:
   966  			launch := new(structs.PeriodicLaunch)
   967  			if err := dec.Decode(launch); err != nil {
   968  				return err
   969  			}
   970  
   971  			// COMPAT: Handle upgrade to v0.7.0
   972  			if launch.Namespace == "" {
   973  				launch.Namespace = structs.DefaultNamespace
   974  			}
   975  
   976  			if err := restore.PeriodicLaunchRestore(launch); err != nil {
   977  				return err
   978  			}
   979  
   980  		case JobSummarySnapshot:
   981  			summary := new(structs.JobSummary)
   982  			if err := dec.Decode(summary); err != nil {
   983  				return err
   984  			}
   985  
   986  			// COMPAT: Handle upgrade to v0.7.0
   987  			if summary.Namespace == "" {
   988  				summary.Namespace = structs.DefaultNamespace
   989  			}
   990  
   991  			if err := restore.JobSummaryRestore(summary); err != nil {
   992  				return err
   993  			}
   994  
   995  		case VaultAccessorSnapshot:
   996  			accessor := new(structs.VaultAccessor)
   997  			if err := dec.Decode(accessor); err != nil {
   998  				return err
   999  			}
  1000  			if err := restore.VaultAccessorRestore(accessor); err != nil {
  1001  				return err
  1002  			}
  1003  
  1004  		case JobVersionSnapshot:
  1005  			version := new(structs.Job)
  1006  			if err := dec.Decode(version); err != nil {
  1007  				return err
  1008  			}
  1009  
  1010  			// COMPAT: Handle upgrade to v0.7.0
  1011  			if version.Namespace == "" {
  1012  				version.Namespace = structs.DefaultNamespace
  1013  			}
  1014  
  1015  			if err := restore.JobVersionRestore(version); err != nil {
  1016  				return err
  1017  			}
  1018  
  1019  		case DeploymentSnapshot:
  1020  			deployment := new(structs.Deployment)
  1021  			if err := dec.Decode(deployment); err != nil {
  1022  				return err
  1023  			}
  1024  
  1025  			// COMPAT: Handle upgrade to v0.7.0
  1026  			if deployment.Namespace == "" {
  1027  				deployment.Namespace = structs.DefaultNamespace
  1028  			}
  1029  
  1030  			if err := restore.DeploymentRestore(deployment); err != nil {
  1031  				return err
  1032  			}
  1033  
  1034  		case ACLPolicySnapshot:
  1035  			policy := new(structs.ACLPolicy)
  1036  			if err := dec.Decode(policy); err != nil {
  1037  				return err
  1038  			}
  1039  			if err := restore.ACLPolicyRestore(policy); err != nil {
  1040  				return err
  1041  			}
  1042  
  1043  		case ACLTokenSnapshot:
  1044  			token := new(structs.ACLToken)
  1045  			if err := dec.Decode(token); err != nil {
  1046  				return err
  1047  			}
  1048  			if err := restore.ACLTokenRestore(token); err != nil {
  1049  				return err
  1050  			}
  1051  
  1052  		default:
  1053  			// Check if this is an enterprise only object being restored
  1054  			restorer, ok := n.enterpriseRestorers[snapType]
  1055  			if !ok {
  1056  				return fmt.Errorf("Unrecognized snapshot type: %v", msgType)
  1057  			}
  1058  
  1059  			// Restore the enterprise only object
  1060  			if err := restorer(restore, dec); err != nil {
  1061  				return err
  1062  			}
  1063  		}
  1064  	}
  1065  
  1066  	restore.Commit()
  1067  
  1068  	// Create Job Summaries
  1069  	// COMPAT 0.4 -> 0.4.1
  1070  	// We can remove this in 0.5. This exists so that the server creates job
  1071  	// summaries if they were not present previously. When users upgrade to 0.5
  1072  	// from 0.4.1, the snapshot will contain job summaries so it will be safe to
  1073  	// remove this block.
  1074  	index, err := newState.Index("job_summary")
  1075  	if err != nil {
  1076  		return fmt.Errorf("couldn't fetch index of job summary table: %v", err)
  1077  	}
  1078  
  1079  	// If the index is 0 that means there is no job summary in the snapshot so
  1080  	// we will have to create them
  1081  	if index == 0 {
  1082  		// query the latest index
  1083  		latestIndex, err := newState.LatestIndex()
  1084  		if err != nil {
  1085  			return fmt.Errorf("unable to query latest index: %v", index)
  1086  		}
  1087  		if err := newState.ReconcileJobSummaries(latestIndex); err != nil {
  1088  			return fmt.Errorf("error reconciling summaries: %v", err)
  1089  		}
  1090  	}
  1091  
  1092  	// External code might be calling State(), so we need to synchronize
  1093  	// here to make sure we swap in the new state store atomically.
  1094  	n.stateLock.Lock()
  1095  	stateOld := n.state
  1096  	n.state = newState
  1097  	n.stateLock.Unlock()
  1098  
  1099  	// Signal that the old state store has been abandoned. This is required
  1100  	// because we don't operate on it any more, we just throw it away, so
  1101  	// blocking queries won't see any changes and need to be woken up.
  1102  	stateOld.Abandon()
  1103  
  1104  	return nil
  1105  }
  1106  
  1107  // reconcileQueuedAllocations re-calculates the queued allocations for every job that we
  1108  // created a Job Summary during the snap shot restore
  1109  func (n *nomadFSM) reconcileQueuedAllocations(index uint64) error {
  1110  	// Get all the jobs
  1111  	ws := memdb.NewWatchSet()
  1112  	iter, err := n.state.Jobs(ws)
  1113  	if err != nil {
  1114  		return err
  1115  	}
  1116  
  1117  	snap, err := n.state.Snapshot()
  1118  	if err != nil {
  1119  		return fmt.Errorf("unable to create snapshot: %v", err)
  1120  	}
  1121  
  1122  	// Invoking the scheduler for every job so that we can populate the number
  1123  	// of queued allocations for every job
  1124  	for {
  1125  		rawJob := iter.Next()
  1126  		if rawJob == nil {
  1127  			break
  1128  		}
  1129  		job := rawJob.(*structs.Job)
  1130  		planner := &scheduler.Harness{
  1131  			State: &snap.StateStore,
  1132  		}
  1133  		// Create an eval and mark it as requiring annotations and insert that as well
  1134  		eval := &structs.Evaluation{
  1135  			ID:             uuid.Generate(),
  1136  			Namespace:      job.Namespace,
  1137  			Priority:       job.Priority,
  1138  			Type:           job.Type,
  1139  			TriggeredBy:    structs.EvalTriggerJobRegister,
  1140  			JobID:          job.ID,
  1141  			JobModifyIndex: job.JobModifyIndex + 1,
  1142  			Status:         structs.EvalStatusPending,
  1143  			AnnotatePlan:   true,
  1144  		}
  1145  		snap.UpsertEvals(100, []*structs.Evaluation{eval})
  1146  		// Create the scheduler and run it
  1147  		sched, err := scheduler.NewScheduler(eval.Type, n.logger, snap, planner)
  1148  		if err != nil {
  1149  			return err
  1150  		}
  1151  
  1152  		if err := sched.Process(eval); err != nil {
  1153  			return err
  1154  		}
  1155  
  1156  		// Get the job summary from the fsm state store
  1157  		originalSummary, err := n.state.JobSummaryByID(ws, job.Namespace, job.ID)
  1158  		if err != nil {
  1159  			return err
  1160  		}
  1161  		summary := originalSummary.Copy()
  1162  
  1163  		// Add the allocations scheduler has made to queued since these
  1164  		// allocations are never getting placed until the scheduler is invoked
  1165  		// with a real planner
  1166  		if l := len(planner.Plans); l != 1 {
  1167  			return fmt.Errorf("unexpected number of plans during restore %d. Please file an issue including the logs", l)
  1168  		}
  1169  		for _, allocations := range planner.Plans[0].NodeAllocation {
  1170  			for _, allocation := range allocations {
  1171  				tgSummary, ok := summary.Summary[allocation.TaskGroup]
  1172  				if !ok {
  1173  					return fmt.Errorf("task group %q not found while updating queued count", allocation.TaskGroup)
  1174  				}
  1175  				tgSummary.Queued += 1
  1176  				summary.Summary[allocation.TaskGroup] = tgSummary
  1177  			}
  1178  		}
  1179  
  1180  		// Add the queued allocations attached to the evaluation to the queued
  1181  		// counter of the job summary
  1182  		if l := len(planner.Evals); l != 1 {
  1183  			return fmt.Errorf("unexpected number of evals during restore %d. Please file an issue including the logs", l)
  1184  		}
  1185  		for tg, queued := range planner.Evals[0].QueuedAllocations {
  1186  			tgSummary, ok := summary.Summary[tg]
  1187  			if !ok {
  1188  				return fmt.Errorf("task group %q not found while updating queued count", tg)
  1189  			}
  1190  
  1191  			// We add instead of setting here because we want to take into
  1192  			// consideration what the scheduler with a mock planner thinks it
  1193  			// placed. Those should be counted as queued as well
  1194  			tgSummary.Queued += queued
  1195  			summary.Summary[tg] = tgSummary
  1196  		}
  1197  
  1198  		if !reflect.DeepEqual(summary, originalSummary) {
  1199  			summary.ModifyIndex = index
  1200  			if err := n.state.UpsertJobSummary(index, summary); err != nil {
  1201  				return err
  1202  			}
  1203  		}
  1204  	}
  1205  	return nil
  1206  }
  1207  
  1208  func (s *nomadSnapshot) Persist(sink raft.SnapshotSink) error {
  1209  	defer metrics.MeasureSince([]string{"nomad", "fsm", "persist"}, time.Now())
  1210  	// Register the nodes
  1211  	encoder := codec.NewEncoder(sink, structs.MsgpackHandle)
  1212  
  1213  	// Write the header
  1214  	header := snapshotHeader{}
  1215  	if err := encoder.Encode(&header); err != nil {
  1216  		sink.Cancel()
  1217  		return err
  1218  	}
  1219  
  1220  	// Write the time table
  1221  	sink.Write([]byte{byte(TimeTableSnapshot)})
  1222  	if err := s.timetable.Serialize(encoder); err != nil {
  1223  		sink.Cancel()
  1224  		return err
  1225  	}
  1226  
  1227  	// Write all the data out
  1228  	if err := s.persistIndexes(sink, encoder); err != nil {
  1229  		sink.Cancel()
  1230  		return err
  1231  	}
  1232  	if err := s.persistNodes(sink, encoder); err != nil {
  1233  		sink.Cancel()
  1234  		return err
  1235  	}
  1236  	if err := s.persistJobs(sink, encoder); err != nil {
  1237  		sink.Cancel()
  1238  		return err
  1239  	}
  1240  	if err := s.persistEvals(sink, encoder); err != nil {
  1241  		sink.Cancel()
  1242  		return err
  1243  	}
  1244  	if err := s.persistAllocs(sink, encoder); err != nil {
  1245  		sink.Cancel()
  1246  		return err
  1247  	}
  1248  	if err := s.persistPeriodicLaunches(sink, encoder); err != nil {
  1249  		sink.Cancel()
  1250  		return err
  1251  	}
  1252  	if err := s.persistJobSummaries(sink, encoder); err != nil {
  1253  		sink.Cancel()
  1254  		return err
  1255  	}
  1256  	if err := s.persistVaultAccessors(sink, encoder); err != nil {
  1257  		sink.Cancel()
  1258  		return err
  1259  	}
  1260  	if err := s.persistJobVersions(sink, encoder); err != nil {
  1261  		sink.Cancel()
  1262  		return err
  1263  	}
  1264  	if err := s.persistDeployments(sink, encoder); err != nil {
  1265  		sink.Cancel()
  1266  		return err
  1267  	}
  1268  	if err := s.persistACLPolicies(sink, encoder); err != nil {
  1269  		sink.Cancel()
  1270  		return err
  1271  	}
  1272  	if err := s.persistACLTokens(sink, encoder); err != nil {
  1273  		sink.Cancel()
  1274  		return err
  1275  	}
  1276  	if err := s.persistEnterpriseTables(sink, encoder); err != nil {
  1277  		sink.Cancel()
  1278  		return err
  1279  	}
  1280  	return nil
  1281  }
  1282  
  1283  func (s *nomadSnapshot) persistIndexes(sink raft.SnapshotSink,
  1284  	encoder *codec.Encoder) error {
  1285  	// Get all the indexes
  1286  	iter, err := s.snap.Indexes()
  1287  	if err != nil {
  1288  		return err
  1289  	}
  1290  
  1291  	for {
  1292  		// Get the next item
  1293  		raw := iter.Next()
  1294  		if raw == nil {
  1295  			break
  1296  		}
  1297  
  1298  		// Prepare the request struct
  1299  		idx := raw.(*state.IndexEntry)
  1300  
  1301  		// Write out a node registration
  1302  		sink.Write([]byte{byte(IndexSnapshot)})
  1303  		if err := encoder.Encode(idx); err != nil {
  1304  			return err
  1305  		}
  1306  	}
  1307  	return nil
  1308  }
  1309  
  1310  func (s *nomadSnapshot) persistNodes(sink raft.SnapshotSink,
  1311  	encoder *codec.Encoder) error {
  1312  	// Get all the nodes
  1313  	ws := memdb.NewWatchSet()
  1314  	nodes, err := s.snap.Nodes(ws)
  1315  	if err != nil {
  1316  		return err
  1317  	}
  1318  
  1319  	for {
  1320  		// Get the next item
  1321  		raw := nodes.Next()
  1322  		if raw == nil {
  1323  			break
  1324  		}
  1325  
  1326  		// Prepare the request struct
  1327  		node := raw.(*structs.Node)
  1328  
  1329  		// Write out a node registration
  1330  		sink.Write([]byte{byte(NodeSnapshot)})
  1331  		if err := encoder.Encode(node); err != nil {
  1332  			return err
  1333  		}
  1334  	}
  1335  	return nil
  1336  }
  1337  
  1338  func (s *nomadSnapshot) persistJobs(sink raft.SnapshotSink,
  1339  	encoder *codec.Encoder) error {
  1340  	// Get all the jobs
  1341  	ws := memdb.NewWatchSet()
  1342  	jobs, err := s.snap.Jobs(ws)
  1343  	if err != nil {
  1344  		return err
  1345  	}
  1346  
  1347  	for {
  1348  		// Get the next item
  1349  		raw := jobs.Next()
  1350  		if raw == nil {
  1351  			break
  1352  		}
  1353  
  1354  		// Prepare the request struct
  1355  		job := raw.(*structs.Job)
  1356  
  1357  		// Write out a job registration
  1358  		sink.Write([]byte{byte(JobSnapshot)})
  1359  		if err := encoder.Encode(job); err != nil {
  1360  			return err
  1361  		}
  1362  	}
  1363  	return nil
  1364  }
  1365  
  1366  func (s *nomadSnapshot) persistEvals(sink raft.SnapshotSink,
  1367  	encoder *codec.Encoder) error {
  1368  	// Get all the evaluations
  1369  	ws := memdb.NewWatchSet()
  1370  	evals, err := s.snap.Evals(ws)
  1371  	if err != nil {
  1372  		return err
  1373  	}
  1374  
  1375  	for {
  1376  		// Get the next item
  1377  		raw := evals.Next()
  1378  		if raw == nil {
  1379  			break
  1380  		}
  1381  
  1382  		// Prepare the request struct
  1383  		eval := raw.(*structs.Evaluation)
  1384  
  1385  		// Write out the evaluation
  1386  		sink.Write([]byte{byte(EvalSnapshot)})
  1387  		if err := encoder.Encode(eval); err != nil {
  1388  			return err
  1389  		}
  1390  	}
  1391  	return nil
  1392  }
  1393  
  1394  func (s *nomadSnapshot) persistAllocs(sink raft.SnapshotSink,
  1395  	encoder *codec.Encoder) error {
  1396  	// Get all the allocations
  1397  	ws := memdb.NewWatchSet()
  1398  	allocs, err := s.snap.Allocs(ws)
  1399  	if err != nil {
  1400  		return err
  1401  	}
  1402  
  1403  	for {
  1404  		// Get the next item
  1405  		raw := allocs.Next()
  1406  		if raw == nil {
  1407  			break
  1408  		}
  1409  
  1410  		// Prepare the request struct
  1411  		alloc := raw.(*structs.Allocation)
  1412  
  1413  		// Write out the evaluation
  1414  		sink.Write([]byte{byte(AllocSnapshot)})
  1415  		if err := encoder.Encode(alloc); err != nil {
  1416  			return err
  1417  		}
  1418  	}
  1419  	return nil
  1420  }
  1421  
  1422  func (s *nomadSnapshot) persistPeriodicLaunches(sink raft.SnapshotSink,
  1423  	encoder *codec.Encoder) error {
  1424  	// Get all the jobs
  1425  	ws := memdb.NewWatchSet()
  1426  	launches, err := s.snap.PeriodicLaunches(ws)
  1427  	if err != nil {
  1428  		return err
  1429  	}
  1430  
  1431  	for {
  1432  		// Get the next item
  1433  		raw := launches.Next()
  1434  		if raw == nil {
  1435  			break
  1436  		}
  1437  
  1438  		// Prepare the request struct
  1439  		launch := raw.(*structs.PeriodicLaunch)
  1440  
  1441  		// Write out a job registration
  1442  		sink.Write([]byte{byte(PeriodicLaunchSnapshot)})
  1443  		if err := encoder.Encode(launch); err != nil {
  1444  			return err
  1445  		}
  1446  	}
  1447  	return nil
  1448  }
  1449  
  1450  func (s *nomadSnapshot) persistJobSummaries(sink raft.SnapshotSink,
  1451  	encoder *codec.Encoder) error {
  1452  
  1453  	ws := memdb.NewWatchSet()
  1454  	summaries, err := s.snap.JobSummaries(ws)
  1455  	if err != nil {
  1456  		return err
  1457  	}
  1458  
  1459  	for {
  1460  		raw := summaries.Next()
  1461  		if raw == nil {
  1462  			break
  1463  		}
  1464  
  1465  		jobSummary := raw.(*structs.JobSummary)
  1466  
  1467  		sink.Write([]byte{byte(JobSummarySnapshot)})
  1468  		if err := encoder.Encode(jobSummary); err != nil {
  1469  			return err
  1470  		}
  1471  	}
  1472  	return nil
  1473  }
  1474  
  1475  func (s *nomadSnapshot) persistVaultAccessors(sink raft.SnapshotSink,
  1476  	encoder *codec.Encoder) error {
  1477  
  1478  	ws := memdb.NewWatchSet()
  1479  	accessors, err := s.snap.VaultAccessors(ws)
  1480  	if err != nil {
  1481  		return err
  1482  	}
  1483  
  1484  	for {
  1485  		raw := accessors.Next()
  1486  		if raw == nil {
  1487  			break
  1488  		}
  1489  
  1490  		accessor := raw.(*structs.VaultAccessor)
  1491  
  1492  		sink.Write([]byte{byte(VaultAccessorSnapshot)})
  1493  		if err := encoder.Encode(accessor); err != nil {
  1494  			return err
  1495  		}
  1496  	}
  1497  	return nil
  1498  }
  1499  
  1500  func (s *nomadSnapshot) persistJobVersions(sink raft.SnapshotSink,
  1501  	encoder *codec.Encoder) error {
  1502  	// Get all the jobs
  1503  	ws := memdb.NewWatchSet()
  1504  	versions, err := s.snap.JobVersions(ws)
  1505  	if err != nil {
  1506  		return err
  1507  	}
  1508  
  1509  	for {
  1510  		// Get the next item
  1511  		raw := versions.Next()
  1512  		if raw == nil {
  1513  			break
  1514  		}
  1515  
  1516  		// Prepare the request struct
  1517  		job := raw.(*structs.Job)
  1518  
  1519  		// Write out a job registration
  1520  		sink.Write([]byte{byte(JobVersionSnapshot)})
  1521  		if err := encoder.Encode(job); err != nil {
  1522  			return err
  1523  		}
  1524  	}
  1525  	return nil
  1526  }
  1527  
  1528  func (s *nomadSnapshot) persistDeployments(sink raft.SnapshotSink,
  1529  	encoder *codec.Encoder) error {
  1530  	// Get all the jobs
  1531  	ws := memdb.NewWatchSet()
  1532  	deployments, err := s.snap.Deployments(ws)
  1533  	if err != nil {
  1534  		return err
  1535  	}
  1536  
  1537  	for {
  1538  		// Get the next item
  1539  		raw := deployments.Next()
  1540  		if raw == nil {
  1541  			break
  1542  		}
  1543  
  1544  		// Prepare the request struct
  1545  		deployment := raw.(*structs.Deployment)
  1546  
  1547  		// Write out a job registration
  1548  		sink.Write([]byte{byte(DeploymentSnapshot)})
  1549  		if err := encoder.Encode(deployment); err != nil {
  1550  			return err
  1551  		}
  1552  	}
  1553  	return nil
  1554  }
  1555  
  1556  func (s *nomadSnapshot) persistACLPolicies(sink raft.SnapshotSink,
  1557  	encoder *codec.Encoder) error {
  1558  	// Get all the policies
  1559  	ws := memdb.NewWatchSet()
  1560  	policies, err := s.snap.ACLPolicies(ws)
  1561  	if err != nil {
  1562  		return err
  1563  	}
  1564  
  1565  	for {
  1566  		// Get the next item
  1567  		raw := policies.Next()
  1568  		if raw == nil {
  1569  			break
  1570  		}
  1571  
  1572  		// Prepare the request struct
  1573  		policy := raw.(*structs.ACLPolicy)
  1574  
  1575  		// Write out a policy registration
  1576  		sink.Write([]byte{byte(ACLPolicySnapshot)})
  1577  		if err := encoder.Encode(policy); err != nil {
  1578  			return err
  1579  		}
  1580  	}
  1581  	return nil
  1582  }
  1583  
  1584  func (s *nomadSnapshot) persistACLTokens(sink raft.SnapshotSink,
  1585  	encoder *codec.Encoder) error {
  1586  	// Get all the policies
  1587  	ws := memdb.NewWatchSet()
  1588  	tokens, err := s.snap.ACLTokens(ws)
  1589  	if err != nil {
  1590  		return err
  1591  	}
  1592  
  1593  	for {
  1594  		// Get the next item
  1595  		raw := tokens.Next()
  1596  		if raw == nil {
  1597  			break
  1598  		}
  1599  
  1600  		// Prepare the request struct
  1601  		token := raw.(*structs.ACLToken)
  1602  
  1603  		// Write out a token registration
  1604  		sink.Write([]byte{byte(ACLTokenSnapshot)})
  1605  		if err := encoder.Encode(token); err != nil {
  1606  			return err
  1607  		}
  1608  	}
  1609  	return nil
  1610  }
  1611  
  1612  // Release is a no-op, as we just need to GC the pointer
  1613  // to the state store snapshot. There is nothing to explicitly
  1614  // cleanup.
  1615  func (s *nomadSnapshot) Release() {}