package structs import ( "bytes" "fmt" "time" "github.com/hashicorp/go-msgpack/codec" ) var ( ErrNoLeader = fmt.Errorf("No cluster leader") ErrNoRegionPath = fmt.Errorf("No path to region") ) type MessageType uint8 const ( NodeRegisterRequestType MessageType = iota NodeDeregisterRequestType NodeUpdateStatusRequestType JobRegisterRequestType JobDeregisterRequestType EvalUpdateRequestType EvalDeleteRequestType AllocUpdateRequestType ) const ( // IgnoreUnknownTypeFlag is set along with a MessageType // to indicate that the message type can be safely ignored // if it is not recognized. This is for future proofing, so // that new commands can be added in a way that won't cause // old servers to crash when the FSM attempts to process them. IgnoreUnknownTypeFlag MessageType = 128 ) // RPCInfo is used to describe common information about query type RPCInfo interface { RequestRegion() string IsRead() bool AllowStaleRead() bool } // QueryOptions is used to specify various flags for read queries type QueryOptions struct { // The target region for this query Region string // If set, wait until query exceeds given index. Must be provided // with MaxQueryTime. MinQueryIndex uint64 // Provided with MinQueryIndex to wait for change. MaxQueryTime time.Duration // If set, any follower can service the request. Results // may be arbitrarily stale. AllowStale bool } func (q QueryOptions) RequestRegion() string { return q.Region } // QueryOption only applies to reads, so always true func (q QueryOptions) IsRead() bool { return true } func (q QueryOptions) AllowStaleRead() bool { return q.AllowStale } type WriteRequest struct { // The target region for this write Region string } func (w WriteRequest) RequestRegion() string { // The target region for this request return w.Region } // WriteRequest only applies to writes, always false func (w WriteRequest) IsRead() bool { return false } func (w WriteRequest) AllowStaleRead() bool { return false } // QueryMeta allows a query response to include potentially // useful metadata about a query type QueryMeta struct { // This is the index associated with the read Index uint64 // If AllowStale is used, this is time elapsed since // last contact between the follower and leader. This // can be used to gauge staleness. LastContact time.Duration // Used to indicate if there is a known leader node KnownLeader bool } // WriteMeta allows a write response to includ e potentially // useful metadata about the write type WriteMeta struct { // This is the index associated with the write Index uint64 } // NodeRegisterRequest is used for Client.Register endpoint // to register a node as being a schedulable entity. type NodeRegisterRequest struct { Node *Node WriteRequest } // NodeDeregisterRequest is used for Client.Deregister endpoint // to deregister a node as being a schedulable entity. type NodeDeregisterRequest struct { NodeID string WriteRequest } // UpdateStatusRequest is used for Client.UpdateStatus endpoint // to update the status of a node. type NodeUpdateStatusRequest struct { NodeID string Status string WriteRequest } // NodeEvaluateRequest is used to re-evaluate the ndoe type NodeEvaluateRequest struct { NodeID string WriteRequest } // NodeSpecificRequest is used when we just need to specify a target node type NodeSpecificRequest struct { NodeID string QueryOptions } // JobRegisterRequest is used for Job.Register endpoint // to register a job as being a schedulable entity. type JobRegisterRequest struct { Job *Job WriteRequest } // JobDeregisterRequest is used for Job.Deregister endpoint // to deregister a job as being a schedulable entity. type JobDeregisterRequest struct { JobID string WriteRequest } // JobEvaluateRequest is used when we just need to re-evaluate a target job type JobEvaluateRequest struct { JobID string WriteRequest } // JobSpecificRequest is used when we just need to specify a target job type JobSpecificRequest struct { JobID string QueryOptions } // EvalUpdateRequest is used for upserting evaluations. type EvalUpdateRequest struct { Evals []*Evaluation EvalToken string WriteRequest } // EvalDeleteRequest is used for deleting an evaluation. type EvalDeleteRequest struct { Evals []string Allocs []string WriteRequest } // EvalSpecificRequest is used when we just need to specify a target evaluation type EvalSpecificRequest struct { EvalID string QueryOptions } // EvalAckRequest is used to Ack/Nack a specific evaluation type EvalAckRequest struct { EvalID string Token string WriteRequest } // EvalDequeueRequest is used when we want to dequeue an evaluation type EvalDequeueRequest struct { Schedulers []string Timeout time.Duration WriteRequest } // PlanRequest is used to submit an allocation plan to the leader type PlanRequest struct { Plan *Plan WriteRequest } // AllocUpdateRequest is used to submit changes to allocations, either // to cause evictions or to assign new allocaitons. Both can be done // within a single transaction type AllocUpdateRequest struct { // Evict is the list of allocation IDs to evict Evict []string // Alloc is the list of new allocations to assign Alloc []*Allocation } // GenericRequest is used to request where no // specific information is needed. type GenericRequest struct { QueryOptions } // GenericResponse is used to respond to a request where no // specific response information is needed. type GenericResponse struct { WriteMeta } const ( ProtocolVersion = "protocol" APIMajorVersion = "api.major" APIMinorVersion = "api.minor" ) // VersionResponse is used for the Status.Version reseponse type VersionResponse struct { Build string Versions map[string]int QueryMeta } // JobRegisterResponse is used to respond to a job registration type JobRegisterResponse struct { EvalID string EvalCreateIndex uint64 JobModifyIndex uint64 QueryMeta } // JobDeregisterResponse is used to respond to a job deregistration type JobDeregisterResponse struct { EvalID string EvalCreateIndex uint64 JobModifyIndex uint64 QueryMeta } // NodeUpdateResponse is used to respond to a node update type NodeUpdateResponse struct { EvalIDs []string EvalCreateIndex uint64 NodeModifyIndex uint64 QueryMeta } // SingleNodeResponse is used to return a single node type SingleNodeResponse struct { Node *Node QueryMeta } // SingleJobResponse is used to return a single job type SingleJobResponse struct { Job *Job QueryMeta } // SingleEvalResponse is used to return a single evaluation type SingleEvalResponse struct { Eval *Evaluation QueryMeta } // EvalDequeueResponse is used to return from a dequeue type EvalDequeueResponse struct { Eval *Evaluation Token string QueryMeta } // PlanResponse is used to return from a PlanRequest type PlanResponse struct { Result *PlanResult WriteMeta } const ( NodeStatusInit = "initializing" NodeStatusReady = "ready" NodeStatusMaint = "maintenance" NodeStatusDrain = "drain" NodeStatusDown = "down" ) // ShouldDrainNode checks if a given node status should trigger an // evaluation. Some states don't require any further action. func ShouldDrainNode(status string) bool { switch status { case NodeStatusInit, NodeStatusReady, NodeStatusMaint: return false case NodeStatusDrain, NodeStatusDown: return true default: panic(fmt.Sprintf("unhandled node status %s", status)) } } // ValidNodeStatus is used to check if a node status is valid func ValidNodeStatus(status string) bool { switch status { case NodeStatusInit, NodeStatusReady, NodeStatusMaint, NodeStatusDrain, NodeStatusDown: return true default: return false } } // Node is a representation of a schedulable client node type Node struct { // ID is a unique identifier for the node. It can be constructed // by doing a concatenation of the Name and Datacenter as a simple // approach. Alternatively a UUID may be used. ID string // Datacenter for this node Datacenter string // Node name Name string // Attributes is an arbitrary set of key/value // data that can be used for constraints. Examples // include "os=linux", "arch=386", "driver.docker=1", // "docker.runtime=1.8.3" Attributes map[string]string // Resources is the available resources on the client. // For example 'cpu=2' 'memory=2048' Resources *Resources // Reserved is the set of resources that are reserved, // and should be subtracted from the total resources for // the purposes of scheduling. This may be provide certain // high-watermark tolerances or because of external schedulers // consuming resources. Reserved *Resources // Links are used to 'link' this client to external // systems. For example 'consul=foo.dc1' 'aws=i-83212' // 'ami=ami-123' Links map[string]string // Meta is used to associate arbitrary metadata with this // client. This is opaque to Nomad. Meta map[string]string // NodeClass is an opaque identifier used to group nodes // together for the purpose of determining scheduling pressure. NodeClass string // Status of this node Status string // StatusDescription is meant to provide more human useful information StatusDescription string // Raft Indexes CreateIndex uint64 ModifyIndex uint64 } // Resources is used to define the resources available // on a client type Resources struct { CPU float64 MemoryMB int DiskMB int IOPS int Networks []*NetworkResource } // NetIndexByCIDR scans the list of networks for a matching // CIDR, returning the index. This currently ONLY handles // an exact match and not a subset CIDR. func (r *Resources) NetIndexByCIDR(cidr string) int { for idx, net := range r.Networks { if net.CIDR == cidr { return idx } } return -1 } // Superset checks if one set of resources is a superset // of another. func (r *Resources) Superset(other *Resources) bool { if r.CPU < other.CPU { return false } if r.MemoryMB < other.MemoryMB { return false } if r.DiskMB < other.DiskMB { return false } if r.IOPS < other.IOPS { return false } for _, net := range r.Networks { idx := other.NetIndexByCIDR(net.CIDR) if idx >= 0 { if net.MBits < other.Networks[idx].MBits { return false } } } // Check that other does not have a network we are missing for _, net := range other.Networks { idx := r.NetIndexByCIDR(net.CIDR) if idx == -1 { return false } } return true } // Add adds the resources of the delta to this, potentially // returning an error if not possible. func (r *Resources) Add(delta *Resources) error { if delta == nil { return nil } r.CPU += delta.CPU r.MemoryMB += delta.MemoryMB r.DiskMB += delta.DiskMB r.IOPS += delta.IOPS for _, net := range delta.Networks { idx := r.NetIndexByCIDR(net.CIDR) if idx == -1 { return fmt.Errorf("missing network for CIDR %s", net.CIDR) } r.Networks[idx].Add(net) } return nil } // NetworkResource is used to represesent available network // resources type NetworkResource struct { Public bool // Is this a public address? CIDR string // CIDR block of addresses ReservedPorts []int // Reserved ports MBits int // Throughput } // Add adds the resources of the delta to this, potentially // returning an error if not possible. func (n *NetworkResource) Add(delta *NetworkResource) { if len(delta.ReservedPorts) > 0 { n.ReservedPorts = append(n.ReservedPorts, delta.ReservedPorts...) } n.MBits += delta.MBits } const ( // JobTypeNomad is reserved for internal system tasks and is // always handled by the CoreScheduler. JobTypeCore = "_core" JobTypeSystem = "system" JobTypeService = "service" JobTypeBatch = "batch" ) const ( JobStatusPending = "pending" // Pending means the job is waiting on scheduling JobStatusRunning = "running" // Running means the entire job is running JobStatusComplete = "complete" // Complete means there was a clean termination JobStatusDead = "dead" // Dead means there was abnormal termination ) const ( // JobMinPriority is the minimum allowed priority JobMinPriority = 1 // JobDefaultPriority is the default priority if not // not specified. JobDefaultPriority = 50 // JobMaxPriority is the maximum allowed priority JobMaxPriority = 100 // Ensure CoreJobPriority is higher than any user // specified job so that it gets priority. This is important // for the system to remain healthy. CoreJobPriority = JobMaxPriority * 2 ) // Job is the scope of a scheduling request to Nomad. It is the largest // scoped object, and is a named collection of task groups. Each task group // is further composed of tasks. A task group (TG) is the unit of scheduling // however. type Job struct { // ID is a unique identifier for the job. It can be the same as // the job name, or alternatively a UUID may be used. ID string // Name is the logical name of the job used to refer to it. This is unique // per region, but not unique globally. Name string // Type is used to control various behaviors about the job. Most jobs // are service jobs, meaning they are expected to be long lived. // Some jobs are batch oriented meaning they run and then terminate. // This can be extended in the future to support custom schedulers. Type string // Priority is used to control scheduling importance and if this job // can preempt other jobs. Priority int // AllAtOnce is used to control if incremental scheduling of task groups // is allowed or if we must do a gang scheduling of the entire job. This // can slow down larger jobs if resources are not available. AllAtOnce bool // Datacenters contains all the datacenters this job is allowed to span Datacenters []string // Constraints can be specified at a job level and apply to // all the task groups and tasks. Constraints []*Constraint // TaskGroups are the collections of task groups that this job needs // to run. Each task group is an atomic unit of scheduling and placement. TaskGroups []*TaskGroup // Meta is used to associate arbitrary metadata with this // job. This is opaque to Nomad. Meta map[string]string // Job status Status string // StatusDescription is meant to provide more human useful information StatusDescription string // Raft Indexes CreateIndex uint64 ModifyIndex uint64 } // TaskGroup is an atomic unit of placement. Each task group belongs to // a job and may contain any number of tasks. A task group support running // in many replicas using the same configuration.. type TaskGroup struct { // Name of the task group Name string // Count is the number of replicas of this task group that should // be scheduled. Count int // Constraints can be specified at a task group level and apply to // all the tasks contained. Constraints []*Constraint // Tasks are the collection of tasks that this task group needs to run Tasks []*Task // Meta is used to associate arbitrary metadata with this // task group. This is opaque to Nomad. Meta map[string]string } // Task is a single process typically that is executed as part of a task group. type Task struct { // Name of the task Name string // Driver is used to control which driver is used Driver string // Config is provided to the driver to initialize Config map[string]string // Constraints can be specified at a task level and apply only to // the particular task. Constraints []*Constraint // Resources is the resources needed by this task Resources *Resources // Meta is used to associate arbitrary metadata with this // task. This is opaque to Nomad. Meta map[string]string } // Constraints are used to restrict placement options in the case of // a hard constraint, and used to prefer a placement in the case of // a soft constraint. type Constraint struct { Hard bool // Hard or soft constraint LTarget string // Left-hand target RTarget string // Right-hand target Operand string // Constraint operand (<=, <, =, !=, >, >=), contains, near Weight int // Soft constraints can vary the weight } func (c *Constraint) String() string { return fmt.Sprintf("%s %s %s", c.LTarget, c.Operand, c.RTarget) } const ( AllocStatusPending = "pending" AllocStatusInit = "initializing" AllocStatusRunning = "running" AllocStatusComplete = "complete" AllocStatusDead = "dead" AllocStatusFailed = "failed" ) // Allocation is used to allocate the placement of a task group to a node. type Allocation struct { // ID of the allocation (UUID) ID string // ID of the evaluation that generated this allocation EvalID string // Name is a logical name of the allocation. Name string // NodeID is the node this is being placed on NodeID string // Job is the parent job of the task group being allocated. // This is copied at allocation time to avoid issues if the job // definition is updated. JobID string Job *Job // Resources is the set of resources allocated as part // of this allocation of the task group. Resources *Resources // Metrics associated with this allocation Metrics *AllocMetric // Status of the allocation Status string // StatusDescription is meant to provide more human useful information StatusDescription string // Raft Indexes CreateIndex uint64 ModifyIndex uint64 } // TerminalStatus returns if the current status is terminal and // will no longer transition. func (a *Allocation) TerminalStatus() bool { switch a.Status { case AllocStatusComplete, AllocStatusDead, AllocStatusFailed: return true default: return false } } // AllocMetric is used to track various metrics while attempting // to make an allocation. These are used to debug a job, or to better // understand the pressure within the system. type AllocMetric struct { // NodesEvaluated is the number of nodes that were evaluated NodesEvaluated int // NodesFiltered is the number of nodes filtered due to // a hard constraint NodesFiltered int // ClassFiltered is the number of nodes filtered by class ClassFiltered map[string]int // ConstraintFiltered is the number of failures caused by constraint ConstraintFiltered map[string]int // NodesExhausted is the nubmer of nodes skipped due to being // exhausted of at least one resource NodesExhausted int // ClassExhausted is the number of nodes exhausted by class ClassExhausted map[string]int // Scores is the scores of the final few nodes remaining // for placement. The top score is typically selected. Scores map[string]float64 // AllocationTime is a measure of how long the allocation // attempt took. This can affect performance and SLAs. AllocationTime time.Duration // CoalescedFailures indicates the number of other // allocations that were coalesced into this failed allocation. // This is to prevent creating many failed allocations for a // single task group. CoalescedFailures int } func (a *AllocMetric) EvaluateNode() { a.NodesEvaluated += 1 } func (a *AllocMetric) FilterNode(node *Node, constraint string) { a.NodesFiltered += 1 if node != nil && node.NodeClass != "" { if a.ClassFiltered == nil { a.ClassFiltered = make(map[string]int) } a.ClassFiltered[node.NodeClass] += 1 } if constraint != "" { if a.ConstraintFiltered == nil { a.ConstraintFiltered = make(map[string]int) } a.ConstraintFiltered[constraint] += 1 } } func (a *AllocMetric) ExhaustedNode(node *Node) { a.NodesExhausted += 1 if node != nil && node.NodeClass != "" { if a.ClassExhausted == nil { a.ClassExhausted = make(map[string]int) } a.ClassExhausted[node.NodeClass] += 1 } } func (a *AllocMetric) ScoreNode(node *Node, name string, score float64) { if a.Scores == nil { a.Scores = make(map[string]float64) } key := fmt.Sprintf("%s.%s", node.ID, name) a.Scores[key] = score } const ( EvalStatusPending = "pending" EvalStatusComplete = "complete" EvalStatusFailed = "failed" ) const ( EvalTriggerJobRegister = "job-register" EvalTriggerJobDeregister = "job-deregister" EvalTriggerNodeUpdate = "node-update" EvalTriggerScheduled = "scheduled" ) const ( CoreJobEvalGC = "eval-gc" ) // Evaluation is used anytime we need to apply business logic as a result // of a change to our desired state (job specification) or the emergent state // (registered nodes). When the inputs change, we need to "evaluate" them, // potentially taking action (allocation of work) or doing nothing if the state // of the world does not require it. type Evaluation struct { // ID is a randonly generated UUID used for this evaluation. This // is assigned upon the creation of the evaluation. ID string // Priority is used to control scheduling importance and if this job // can preempt other jobs. Priority int // Type is used to control which schedulers are available to handle // this evaluation. Type string // TriggeredBy is used to give some insight into why this Eval // was created. (Job change, node failure, alloc failure, etc). TriggeredBy string // JobID is the job this evaluation is scoped to. Evalutions cannot // be run in parallel for a given JobID, so we serialize on this. JobID string // JobModifyIndex is the modify index of the job at the time // the evaluation was created JobModifyIndex uint64 // NodeID is the node that was affected triggering the evaluation. NodeID string // NodeModifyIndex is the modify index of the node at the time // the evaluation was created NodeModifyIndex uint64 // Status of the evaluation Status string // StatusDescription is meant to provide more human useful information StatusDescription string // Raft Indexes CreateIndex uint64 ModifyIndex uint64 } // TerminalStatus returns if the current status is terminal and // will no longer transition. func (e *Evaluation) TerminalStatus() bool { switch e.Status { case EvalStatusComplete, EvalStatusFailed: return true default: return false } } func (e *Evaluation) GoString() string { return fmt.Sprintf("", e.ID, e.JobID) } func (e *Evaluation) Copy() *Evaluation { ne := new(Evaluation) *ne = *e return ne } // ShouldEnqueue checks if a given evaluation should be enqueued func (e *Evaluation) ShouldEnqueue() bool { switch e.Status { case EvalStatusPending: return true case EvalStatusComplete, EvalStatusFailed: return false default: panic(fmt.Sprintf("unhandled evaluation (%s) status %s", e.ID, e.Status)) } } // MakePlan is used to make a plan from the given evaluation // for a given Job func (e *Evaluation) MakePlan(j *Job) *Plan { p := &Plan{ EvalID: e.ID, Priority: e.Priority, NodeEvict: make(map[string][]string), NodeAllocation: make(map[string][]*Allocation), } if j != nil { p.AllAtOnce = j.AllAtOnce } return p } // Plan is used to submit a commit plan for task allocations. These // are submitted to the leader which verifies that resources have // not been overcommitted before admiting the plan. type Plan struct { // EvalID is the evaluation ID this plan is associated with EvalID string // EvalToken is used to prevent a split-brain processing of // an evaluation. There should only be a single scheduler running // an Eval at a time, but this could be violated after a leadership // transition. This unique token is used to reject plans that are // being submitted from a different leader. EvalToken string // Priority is the priority of the upstream job Priority int // AllAtOnce is used to control if incremental scheduling of task groups // is allowed or if we must do a gang scheduling of the entire job. // If this is false, a plan may be partially applied. Otherwise, the // entire plan must be able to make progress. AllAtOnce bool // NodeEvict contains all the evictions for each node. For each node, // this is a list of the allocation IDs to evict. NodeEvict map[string][]string // NodeAllocation contains all the allocations for each node. // The evicts must be considered prior to the allocations. NodeAllocation map[string][]*Allocation // FailedAllocs are allocations that could not be made, // but are persisted so that the user can use the feedback // to determine the cause. FailedAllocs []*Allocation } func (p *Plan) AppendEvict(alloc *Allocation) { node := alloc.NodeID existing := p.NodeEvict[node] p.NodeEvict[node] = append(existing, alloc.ID) } func (p *Plan) AppendAlloc(alloc *Allocation) { node := alloc.NodeID existing := p.NodeAllocation[node] p.NodeAllocation[node] = append(existing, alloc) } func (p *Plan) AppendFailed(alloc *Allocation) { p.FailedAllocs = append(p.FailedAllocs, alloc) } // IsNoOp checks if this plan would do nothing func (p *Plan) IsNoOp() bool { return len(p.NodeEvict) == 0 && len(p.NodeAllocation) == 0 && len(p.FailedAllocs) == 0 } // PlanResult is the result of a plan submitted to the leader. type PlanResult struct { // NodeEvict contains all the evictions that were committed. NodeEvict map[string][]string // NodeAllocation contains all the allocations that were committed. NodeAllocation map[string][]*Allocation // FailedAllocs are allocations that could not be made, // but are persisted so that the user can use the feedback // to determine the cause. FailedAllocs []*Allocation // RefreshIndex is the index the worker should refresh state up to. // This allows all evictions and allocations to be materialized. // If any allocations were rejected due to stale data (node state, // over committed) this can be used to force a worker refresh. RefreshIndex uint64 // AllocIndex is the Raft index in which the evictions and // allocations took place. This is used for the write index. AllocIndex uint64 } // FullCommit is used to check if all the allocations in a plan // were committed as part of the result. Returns if there was // a match, and the number of expected and actual allocations. func (p *PlanResult) FullCommit(plan *Plan) (bool, int, int) { expected := 0 actual := 0 for name, allocList := range plan.NodeAllocation { didAlloc, _ := p.NodeAllocation[name] expected += len(allocList) actual += len(didAlloc) } return actual == expected, expected, actual } // msgpackHandle is a shared handle for encoding/decoding of structs var msgpackHandle = &codec.MsgpackHandle{} // Decode is used to decode a MsgPack encoded object func Decode(buf []byte, out interface{}) error { return codec.NewDecoder(bytes.NewReader(buf), msgpackHandle).Decode(out) } // Encode is used to encode a MsgPack object with type prefix func Encode(t MessageType, msg interface{}) ([]byte, error) { var buf bytes.Buffer buf.WriteByte(uint8(t)) err := codec.NewEncoder(&buf, msgpackHandle).Encode(msg) return buf.Bytes(), err }