func (as *AgentState) GetReplacedUnit(j *job.Job) (string, error) {
cExists, replaced := as.hasReplace(j.Name, j.Replaces())
if !cExists {
return "", fmt.Errorf("cannot find units to be replaced for Unit(%s)", j.Name)
}
return replaced, nil
}
func mapUnitToJob(entity *schema.Unit, mm map[string]*machine.MachineState) (*job.Job, error) {
contents, err := base64.StdEncoding.DecodeString(entity.FileContents)
if err != nil {
return nil, err
}
u, err := unit.NewUnit(string(contents))
if err != nil {
return nil, err
}
js := job.JobState(entity.CurrentState)
j := job.Job{
Name: entity.Name,
State: &js,
Unit: *u,
}
// populate a UnitState object only if the entity
// is actually reporting relevant data
if entity.Systemd != nil {
j.UnitState = &unit.UnitState{
LoadState: entity.Systemd.LoadState,
ActiveState: entity.Systemd.ActiveState,
SubState: entity.Systemd.SubState,
}
if len(entity.Systemd.MachineID) > 0 {
j.UnitState.MachineID = entity.Systemd.MachineID
}
}
return &j, nil
}
// partitionCluster returns a slice of IDs from a subset of active machines
// that should be considered for scheduling the specified job.
// The returned slice is sorted by ascending lexicographical string value of machine boot id.
func (eg *Engine) partitionCluster(j *job.Job) ([]string, error) {
if machID, ok := j.RequiredTarget(); ok {
return []string{machID}, nil
}
// TODO(uwedeportivo): for now punt on jobs with requirements and offer to all machines
// because agents are decoding the requirements
if len(j.Requirements()) > 0 {
machines, _ := eg.registry.Machines()
machineIDs := make([]string, len(machines))
for i, mach := range machines {
machineIDs[i] = mach.ID
}
sort.Strings(machineIDs)
return machineIDs, nil
}
// this is usually a cheap no-op
eg.refreshCluster(false)
// as an initial heuristic, choose the k least loaded, with k = partitionSize
machineIDs := eg.clust.kLeastLoaded(partitionSize)
sort.Strings(machineIDs)
return machineIDs, nil
}
// AbleToRun determines if an Agent can run the provided Job based on
// the Agent's current state. A boolean indicating whether this is the
// case or not is returned. The following criteria is used:
// - Agent must meet the Job's machine target requirement (if any)
// - Agent must have all of the Job's required metadata (if any)
// - Agent must have all required Peers of the Job scheduled locally (if any)
// - Job must not conflict with any other Units scheduled to the agent
// - Job must specially handle replaced units to be rescheduled
func (as *AgentState) AbleToRun(j *job.Job) (jobAction job.JobAction, errstr string) {
if tgt, ok := j.RequiredTarget(); ok && !as.MState.MatchID(tgt) {
return job.JobActionUnschedule, fmt.Sprintf("agent ID %q does not match required %q", as.MState.ID, tgt)
}
metadata := j.RequiredTargetMetadata()
if len(metadata) != 0 {
if !machine.HasMetadata(as.MState, metadata) {
return job.JobActionUnschedule, "local Machine metadata insufficient"
}
}
peers := j.Peers()
if len(peers) != 0 {
for _, peer := range peers {
if !as.unitScheduled(peer) {
return job.JobActionUnschedule, fmt.Sprintf("required peer Unit(%s) is not scheduled locally", peer)
}
}
}
if cExists, cJobName := as.HasConflict(j.Name, j.Conflicts()); cExists {
return job.JobActionUnschedule, fmt.Sprintf("found conflict with locally-scheduled Unit(%s)", cJobName)
}
// Handle Replace option specially for rescheduling the unit
if cExists, cJobName := as.hasReplace(j.Name, j.Replaces()); cExists {
return job.JobActionReschedule, fmt.Sprintf("found replace with locally-scheduled Unit(%s)", cJobName)
}
return job.JobActionSchedule, ""
}
// Determine if the Agent can run the provided Job
func (a *Agent) AbleToRun(j *job.Job) bool {
if !a.VerifyJob(j) {
log.V(1).Infof("Failed to verify Job(%s)", j.Name)
return false
}
requirements := j.Requirements()
if len(requirements) == 0 {
log.V(1).Infof("Job(%s) has no requirements", j.Name)
return true
}
if log.V(1) {
var reqString string
for key, slice := range requirements {
reqString += fmt.Sprintf("%s = [", key)
for _, val := range slice {
reqString += fmt.Sprintf("%s, ", val)
}
reqString += fmt.Sprint("] ")
}
log.Infof("Job(%s) has requirements: %s", j.Name, reqString)
}
metadata := extractMachineMetadata(requirements)
log.V(1).Infof("Job(%s) requires machine metadata: %v", j.Name, metadata)
if !a.machine.HasMetadata(metadata) {
log.V(1).Infof("Unable to run Job(%s), local Machine metadata insufficient", j.Name)
return false
}
bootID, ok := requirements[unit.FleetXConditionMachineBootID]
if ok && len(bootID) > 0 && !a.machine.State().MatchBootID(bootID[0]) {
log.V(1).Infof("Agent does not pass MachineBootID condition for Job(%s)", j.Name)
return false
}
peers := j.Payload.Peers()
if len(peers) > 0 {
log.V(1).Infof("Asserting required Peers %v of Job(%s) are scheduled locally", peers, j.Name)
for _, peer := range peers {
if !a.peerScheduledHere(j.Name, peer) {
log.V(1).Infof("Required Peer(%s) of Job(%s) is not scheduled locally", peer, j.Name)
return false
}
}
} else {
log.V(2).Infof("Job(%s) has no peers to worry about", j.Name)
}
if conflicted, conflictedJobName := a.state.HasConflict(j.Name, j.Payload.Conflicts()); conflicted {
log.V(1).Infof("Job(%s) has conflict with Job(%s)", j.Name, conflictedJobName)
return false
}
return true
}
// AbleToRun determines if an Agent can run the provided Job based on
// the Agent's current state. A boolean indicating whether this is the
// case or not is returned. The following criteria is used:
// - Agent must meet the Job's machine target requirement (if any)
// - Agent must have all of the Job's required metadata (if any)
// - Agent must have all required Peers of the Job scheduled locally (if any)
// - Job must not conflict with any other Jobs scheduled to the agent
func (as *AgentState) AbleToRun(j *job.Job) (bool, string) {
if tgt, ok := j.RequiredTarget(); ok && !as.MState.MatchID(tgt) {
return false, fmt.Sprintf("agent ID %q does not match required %q", as.MState.ID, tgt)
}
metadata := j.RequiredTargetMetadata()
if len(metadata) != 0 {
if !machine.HasMetadata(as.MState, metadata) {
return false, "local Machine metadata insufficient"
}
}
peers := j.Peers()
if len(peers) != 0 {
for _, peer := range peers {
if !as.jobScheduled(peer) {
return false, fmt.Sprintf("required peer Job(%s) is not scheduled locally", peer)
}
}
}
if cExists, cJobName := as.hasConflict(j.Name, j.Conflicts()); cExists {
return false, fmt.Sprintf("found conflict with locally-scheduled Job(%s)", cJobName)
}
return true, ""
}
func (r *EtcdRegistry) hydrateJob(j *job.Job) (err error) {
j.TargetState, err = r.jobTargetState(j.Name)
if err != nil {
return
}
j.TargetMachineID, err = r.jobTargetMachine(j.Name)
if err != nil {
return
}
j.UnitState = r.getUnitState(j.Name)
j.State = r.determineJobState(j.Name)
return
}
// Determine if the Agent can run the provided Job
func (a *Agent) ableToRun(j *job.Job) bool {
if !a.verifyJob(j) {
log.V(1).Infof("Failed to verify Job(%s)", j.Name)
return false
}
requirements := j.Requirements()
if len(requirements) == 0 {
log.V(1).Infof("Job(%s) has no requirements", j.Name)
}
log.Infof("Job(%s) has requirements: %s", j.Name, requirements)
metadata := j.RequiredTargetMetadata()
log.V(1).Infof("Job(%s) requires machine metadata: %v", j.Name, metadata)
ms := a.Machine.State()
if !machine.HasMetadata(&ms, metadata) {
log.Infof("Unable to run Job(%s), local Machine metadata insufficient", j.Name)
return false
}
if tgt, ok := j.RequiredTarget(); ok && !a.Machine.State().MatchID(tgt) {
log.Infof("Agent does not meet machine target requirement for Job(%s)", j.Name)
return false
}
peers := j.Peers()
if len(peers) > 0 {
log.V(1).Infof("Asserting required Peers %v of Job(%s) are scheduled locally", peers, j.Name)
for _, peer := range peers {
if !a.peerScheduledHere(j.Name, peer) {
log.Infof("Required Peer(%s) of Job(%s) is not scheduled locally", peer, j.Name)
return false
}
}
} else {
log.V(1).Infof("Job(%s) has no peers to worry about", j.Name)
}
if conflicted, conflictedJobName := a.HasConflict(j.Name, j.Conflicts()); conflicted {
log.Infof("Job(%s) has conflict with Job(%s)", j.Name, conflictedJobName)
return false
}
return true
}
// ableToRun determines if the Agent can run the provided Job, and returns a boolean indicating
// whether this is the case. There are five criteria for an Agent to be eligible to run a Job:
// - Job must pass signature verification
// - agent must have all of the Job's required metadata (if any)
// - agent must meet the Job's machine target requirement (if any)
// - agent must have all required Peers of the Job scheduled locally (if any)
// - Job must not conflict with any other Jobs scheduled to the agent
func (a *Agent) ableToRun(j *job.Job) bool {
if !a.verifyJobSignature(j) {
log.V(1).Infof("Failed to verify Job(%s)", j.Name)
return false
}
log.Infof("Job(%s) has requirements: %s", j.Name, j.Requirements())
metadata := j.RequiredTargetMetadata()
if len(metadata) == 0 {
log.V(1).Infof("Job(%s) has no required machine metadata", j.Name)
} else {
log.V(1).Infof("Job(%s) requires machine metadata: %v", j.Name, metadata)
ms := a.Machine.State()
if !machine.HasMetadata(&ms, metadata) {
log.Infof("Unable to run Job(%s): local Machine metadata insufficient", j.Name)
return false
}
}
if tgt, ok := j.RequiredTarget(); ok && !a.Machine.State().MatchID(tgt) {
log.Infof("Unable to run Job(%s): agent does not meet machine target requirement (%s)", j.Name, tgt)
return false
}
peers := j.Peers()
if len(peers) == 0 {
log.V(1).Infof("Job(%s) has no required peers", j.Name)
} else {
log.V(1).Infof("Job(%s) requires peers: %v", j.Name, peers)
for _, peer := range peers {
if !a.peerScheduledHere(j.Name, peer) {
log.Infof("Unable to run Job(%s): required Peer(%s) is not scheduled locally", j.Name, peer)
return false
}
}
}
if conflicted, conflictedJobName := a.HasConflict(j.Name, j.Conflicts()); conflicted {
log.Infof("Unable to run Job(%s): conflict with Job(%s)", j.Name, conflictedJobName)
return false
}
return true
}
func (r *EtcdRegistry) hydrateJob(j *job.Job) error {
tgt, err := r.jobTargetState(j.Name)
if err != nil {
return err
}
j.TargetState = tgt
j.TargetMachineID, err = r.jobTargetMachine(j.Name)
if err != nil {
return err
}
j.UnitState = r.getUnitState(j.Name)
j.State = r.determineJobState(j.Name)
return nil
}
// ableToRun determines if the Agent can run the provided Job based on
// the Agent's desired state. A boolean indicating whether this is the
// case or not is returned. The following criteria is used:
// - Agent must meet the Job's machine target requirement (if any)
// - Job must pass signature verification
// - Agent must have all of the Job's required metadata (if any)
// - Agent must have all required Peers of the Job scheduled locally (if any)
// - Job must not conflict with any other Jobs scheduled to the agent
func (ar *AgentReconciler) ableToRun(as *agentState, ms *machine.MachineState, j *job.Job) (bool, string) {
log.V(1).Infof("Attempting to determine if able to run Job(%s)", j.Name)
if tgt, ok := j.RequiredTarget(); ok && !ms.MatchID(tgt) {
return false, fmt.Sprintf("Agent ID %q does not match required %q", ms.ID, tgt)
}
if !ar.verifyJobSignature(j) {
return false, "unable to verify signature"
}
log.V(1).Infof("Job(%s) has requirements: %s", j.Name, j.Requirements())
metadata := j.RequiredTargetMetadata()
if len(metadata) == 0 {
log.V(1).Infof("Job(%s) has no required machine metadata", j.Name)
} else {
log.V(1).Infof("Job(%s) requires machine metadata: %v", j.Name, metadata)
if !machine.HasMetadata(ms, metadata) {
return false, "local Machine metadata insufficient"
}
}
peers := j.Peers()
if len(peers) == 0 {
log.V(1).Infof("Job(%s) has no required peers", j.Name)
} else {
log.V(1).Infof("Job(%s) requires peers: %v", j.Name, peers)
for _, peer := range peers {
if !as.jobScheduled(peer) {
return false, fmt.Sprintf("required peer Job(%s) is not scheduled locally", peer)
}
}
}
if cExists, cJobName := as.hasConflict(j.Name, j.Conflicts()); cExists {
return false, fmt.Sprintf("found conflict with locally-scheduled Job(%s)", cJobName)
}
log.V(1).Infof("Determined local Agent is able to run Job(%s)", j.Name)
return true, ""
}
func newClusterState(units []job.Unit, sUnits []job.ScheduledUnit, machines []machine.MachineState) *clusterState {
sUnitMap := make(map[string]*job.ScheduledUnit)
for _, sUnit := range sUnits {
sUnit := sUnit
sUnitMap[sUnit.Name] = &sUnit
}
jMap := make(map[string]*job.Job)
guMap := make(map[string]*job.Unit)
for _, u := range units {
if u.IsGlobal() {
u := u
guMap[u.Name] = &u
} else {
j := job.Job{
Name: u.Name,
Unit: u.Unit,
TargetState: u.TargetState,
}
if sUnit, ok := sUnitMap[u.Name]; ok {
j.TargetMachineID = sUnit.TargetMachineID
j.State = sUnit.State
}
jMap[j.Name] = &j
}
}
mMap := make(map[string]*machine.MachineState, len(machines))
for _, ms := range machines {
ms := ms
mMap[ms.ID] = &ms
}
return &clusterState{
jobs: jMap,
gUnits: guMap,
machines: mMap,
}
}
func (r *EtcdRegistry) parseJobDir(j *job.Job, dir *etcd.Node) (err error) {
var heartbeat string
for _, node := range dir.Nodes {
switch node.Key {
case r.jobTargetStatePath(j.Name):
j.TargetState, err = job.ParseJobState(node.Value)
if err != nil {
return
}
case r.jobTargetAgentPath(j.Name):
j.TargetMachineID = node.Value
case r.jobHeartbeatPath(j.Name):
heartbeat = node.Value
}
}
j.UnitState = r.getUnitState(j.Name)
js := determineJobState(j, heartbeat)
j.State = &js
return
}
// TrackJob extracts and stores information about the given job for later reference
func (as *AgentState) TrackJob(j *job.Job) {
as.trackJobPeers(j.Name, j.Peers())
as.trackJobConflicts(j.Name, j.Conflicts())
}
// requiresMachine returns whether specified job requires a specific machine.
func (eg *Engine) requiresMachine(j *job.Job) ([]string, bool) {
requirements := j.Requirements()
bootID, ok := requirements[unit.FleetXConditionMachineBootID]
return bootID, ok && len(bootID) > 0
}
