Let partition capacity use issue evaluation.

cmd/metal-api/internal/issues/asn-uniqueness.go

@@ -18,8 +18,8 @@ type (
 	}
 )
 
-func (i *IssueASNUniqueness) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueASNUniqueness) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeASNUniqueness,
 		Severity:    IssueSeverityMinor,
 		Description: "The ASN is not unique (only impact on firewalls)",

cmd/metal-api/internal/issues/bmc-info-outdated.go

@@ -37,8 +37,8 @@ func (i *IssueBMCInfoOutdated) Evaluate(m metal.Machine, ec metal.ProvisioningEv
 	return false
 }
 
-func (*IssueBMCInfoOutdated) Spec() *issueSpec {
-	return &issueSpec{
+func (*IssueBMCInfoOutdated) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeBMCInfoOutdated,
 		Severity:    IssueSeverityMajor,
 		Description: "BMC has not been updated from either metal-hammer or metal-bmc",

cmd/metal-api/internal/issues/bmc-without-ip.go

@@ -10,8 +10,8 @@ type (
 	IssueBMCWithoutIP struct{}
 )
 
-func (i *IssueBMCWithoutIP) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueBMCWithoutIP) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeBMCWithoutIP,
 		Severity:    IssueSeverityMajor,
 		Description: "BMC has no ip address",

cmd/metal-api/internal/issues/bmc-without-mac.go

@@ -10,8 +10,8 @@ type (
 	IssueBMCWithoutMAC struct{}
 )
 
-func (i *IssueBMCWithoutMAC) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueBMCWithoutMAC) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeBMCWithoutMAC,
 		Severity:    IssueSeverityMajor,
 		Description: "BMC has no mac address",

cmd/metal-api/internal/issues/crash-loop.go

@@ -13,8 +13,8 @@ type (
 	IssueCrashLoop struct{}
 )
 
-func (i *IssueCrashLoop) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueCrashLoop) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeCrashLoop,
 		Severity:    IssueSeverityMajor,
 		Description: "machine is in a provisioning crash loop (⭕)",

cmd/metal-api/internal/issues/failed-machine-reclaim.go

@@ -13,8 +13,8 @@ type (
 	IssueFailedMachineReclaim struct{}
 )
 
-func (i *IssueFailedMachineReclaim) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueFailedMachineReclaim) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeFailedMachineReclaim,
 		Severity:    IssueSeverityCritical,
 		Description: "machine phones home but not allocated",

cmd/metal-api/internal/issues/issues.go

@@ -38,20 +38,20 @@ type (
 	// MachineIssues is map of a machine response to a list of machine issues
 	MachineIssues []*MachineWithIssues
 
-	machineIssueMap map[*metal.Machine]Issues
+	MachineIssuesMap map[*metal.Machine]Issues
 
-	issueImpl interface {
+	issue interface {
 		// Evaluate decides whether a given machine has the machine issue.
 		// the third argument contains additional information that may be required for the issue evaluation
 		Evaluate(m metal.Machine, ec metal.ProvisioningEventContainer, c *IssueConfig) bool
 		// Spec returns the issue spec of this issue.
-		Spec() *issueSpec
+		Spec() *spec
 		// Details returns additional information on the issue after the evaluation.
 		Details() string
 	}
 
-	// issueSpec defines the specification of an issue.
-	issueSpec struct {
+	// spec defines the specification of an issue.
+	spec struct {
 		Type        IssueType
 		Severity    IssueSeverity
 		Description string
@@ -74,7 +74,7 @@ func AllIssues() Issues {
 	return res
 }
 
-func toIssue(i issueImpl) Issue {
+func toIssue(i issue) Issue {
 	return Issue{
 		Type:        i.Spec().Type,
 		Severity:    i.Spec().Severity,
@@ -84,8 +84,12 @@ func toIssue(i issueImpl) Issue {
 	}
 }
 
-func FindIssues(c *IssueConfig) (MachineIssues, error) {
-	res := machineIssueMap{}
+func FindIssues(c *IssueConfig) (MachineIssuesMap, error) {
+	if c.LastErrorThreshold == 0 {
+		c.LastErrorThreshold = DefaultLastErrorThreshold()
+	}
+
+	res := MachineIssuesMap{}
 
 	ecs := c.EventContainers.ByID()
 
@@ -114,7 +118,7 @@ func FindIssues(c *IssueConfig) (MachineIssues, error) {
 		}
 	}
 
-	return res.toList(), nil
+	return res, nil
 }
 
 func (mis MachineIssues) Get(id string) *MachineWithIssues {
@@ -161,7 +165,7 @@ func (c *IssueConfig) includeIssue(t IssueType) bool {
 	return true
 }
 
-func (mim machineIssueMap) add(m metal.Machine, issue Issue) {
+func (mim MachineIssuesMap) add(m metal.Machine, issue Issue) {
 	issues, ok := mim[&m]
 	if !ok {
 		issues = Issues{}
@@ -170,7 +174,7 @@ func (mim machineIssueMap) add(m metal.Machine, issue Issue) {
 	mim[&m] = issues
 }
 
-func (mim machineIssueMap) toList() MachineIssues {
+func (mim MachineIssuesMap) ToList() MachineIssues {
 	var res MachineIssues
 
 	for m, issues := range mim {

cmd/metal-api/internal/issues/issues_test.go

@@ -509,7 +509,7 @@ func TestFindIssues(t *testing.T) {
 				want = tt.want(ms)
 			}
 
-			if diff := cmp.Diff(want, got, cmp.AllowUnexported(IssueLastEventError{}, IssueASNUniqueness{}, IssueNonDistinctBMCIP{})); diff != "" {
+			if diff := cmp.Diff(want, got.ToList(), cmp.AllowUnexported(IssueLastEventError{}, IssueASNUniqueness{}, IssueNonDistinctBMCIP{})); diff != "" {
 				t.Errorf("diff (+got -want):\n %s", diff)
 			}
 		})

cmd/metal-api/internal/issues/last-event-error.go

@@ -21,8 +21,8 @@ func DefaultLastErrorThreshold() time.Duration {
 	return 7 * 24 * time.Hour
 }
 
-func (i *IssueLastEventError) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueLastEventError) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeLastEventError,
 		Severity:    IssueSeverityMinor,
 		Description: "the machine had an error during the provisioning lifecycle",

cmd/metal-api/internal/issues/liveliness-dead.go

@@ -10,8 +10,8 @@ type (
 	IssueLivelinessDead struct{}
 )
 
-func (i *IssueLivelinessDead) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueLivelinessDead) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeLivelinessDead,
 		Severity:    IssueSeverityMajor,
 		Description: "the machine is not sending events anymore",

cmd/metal-api/internal/issues/liveliness-not-available.go

@@ -10,8 +10,8 @@ type (
 	IssueLivelinessNotAvailable struct{}
 )
 
-func (i *IssueLivelinessNotAvailable) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueLivelinessNotAvailable) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeLivelinessNotAvailable,
 		Severity:    IssueSeverityMinor,
 		Description: "the machine liveliness is not available",

cmd/metal-api/internal/issues/liveliness-unknown.go

@@ -10,8 +10,8 @@ type (
 	IssueLivelinessUnknown struct{}
 )
 
-func (i *IssueLivelinessUnknown) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueLivelinessUnknown) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeLivelinessUnknown,
 		Severity:    IssueSeverityMajor,
 		Description: "the machine is not sending LLDP alive messages anymore",

cmd/metal-api/internal/issues/no-event-container.go

@@ -12,8 +12,8 @@ type (
 	IssueNoEventContainer struct{}
 )
 
-func (i *IssueNoEventContainer) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueNoEventContainer) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeNoEventContainer,
 		Severity:    IssueSeverityMajor,
 		Description: "machine has no event container",

cmd/metal-api/internal/issues/no-partition.go

@@ -10,8 +10,8 @@ type (
 	IssueNoPartition struct{}
 )
 
-func (i *IssueNoPartition) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueNoPartition) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeNoPartition,
 		Severity:    IssueSeverityMajor,
 		Description: "machine with no partition",

cmd/metal-api/internal/issues/non-distinct-bmc-ip.go

@@ -16,8 +16,8 @@ type (
 	}
 )
 
-func (i *IssueNonDistinctBMCIP) Spec() *issueSpec {
-	return &issueSpec{
+func (i *IssueNonDistinctBMCIP) Spec() *spec {
+	return &spec{
 		Type:        IssueTypeNonDistinctBMCIP,
 		Description: "BMC IP address is not distinct",
 		RefURL:      "https://docs.metal-stack.io/stable/installation/troubleshoot/#bmc-no-distinct-ip",

cmd/metal-api/internal/issues/types.go

@@ -24,7 +24,7 @@ func AllIssueTypes() []IssueType {
 	}
 }
 
-func NewIssueFromType(t IssueType) (issueImpl, error) {
+func NewIssueFromType(t IssueType) (issue, error) {
 	switch t {
 	case IssueTypeNoPartition:
 		return &IssueNoPartition{}, nil

cmd/metal-api/internal/metal/provisioning.go

@@ -48,8 +48,8 @@ var (
 type ProvisioningEvents []ProvisioningEvent
 
 // Is return true if given event is equal to specific EventType
-func (p ProvisioningEventType) Is(event string) bool {
-	return string(p) == event
+func (p ProvisioningEventType) Is(event ProvisioningEventType) bool {
+	return p == event
 }
 
 // TrimEvents trim the events to maxCount

cmd/metal-api/internal/metal/provisioning_test.go

@@ -8,7 +8,7 @@ import (
 func TestProvisioningEventType_Is(t *testing.T) {
 	tests := []struct {
 		name  string
-		event string
+		event ProvisioningEventType
 		p     ProvisioningEventType
 		want  bool
 	}{

cmd/metal-api/internal/service/machine-service.go

@@ -593,7 +593,7 @@ func (r *machineResource) issues(request *restful.Request, response *restful.Res
 		return
 	}
 
-	issues, err := issues.FindIssues(&issues.IssueConfig{
+	machinesWithIssues, err := issues.FindIssues(&issues.IssueConfig{
 		Machines:           ms,
 		EventContainers:    ecs,
 		Severity:           severity,
@@ -607,7 +607,7 @@ func (r *machineResource) issues(request *restful.Request, response *restful.Res
 	}
 
 	var issueResponse []*v1.MachineIssueResponse
-	for _, machineWithIssues := range issues {
+	for _, machineWithIssues := range machinesWithIssues.ToList() {
 		machineWithIssues := machineWithIssues
 
 		entry := &v1.MachineIssueResponse{
@@ -1939,7 +1939,7 @@ func evaluateMachineLiveliness(ds *datastore.RethinkStore, m metal.Machine) (met
 	provisioningEvents, err := ds.FindProvisioningEventContainer(m.ID)
 	if err != nil {
 		// we have no provisioning events... we cannot tell
-		return metal.MachineLivelinessUnknown, fmt.Errorf("no provisioningEvents found for ID: %s", m.ID)
+		return metal.MachineLivelinessUnknown, fmt.Errorf("no provisioning event container found for machine: %s", m.ID)
 	}
 
 	old := *provisioningEvents
@@ -1956,6 +1956,7 @@ func evaluateMachineLiveliness(ds *datastore.RethinkStore, m metal.Machine) (met
 		} else {
 			provisioningEvents.Liveliness = metal.MachineLivelinessAlive
 		}
+
 		err = ds.UpdateProvisioningEventContainer(&old, provisioningEvents)
 		if err != nil {
 			return provisioningEvents.Liveliness, err
@@ -2013,7 +2014,6 @@ func ResurrectMachines(ctx context.Context, ds *datastore.RethinkStore, publishe
 			}
 			continue
 		}
-
 	}
 
 	logger.Info("finished machine resurrection")
@@ -2238,27 +2238,6 @@ func publishMachineCmd(logger *zap.SugaredLogger, m *metal.Machine, publisher bu
 	return nil
 }
 
-func machineHasIssues(m *v1.MachineResponse) bool {
-	if m.Partition == nil {
-		return true
-	}
-	if !metal.MachineLivelinessAlive.Is(m.Liveliness) {
-		return true
-	}
-	if m.Allocation == nil && len(m.RecentProvisioningEvents.Events) > 0 && metal.ProvisioningEventPhonedHome.Is(m.RecentProvisioningEvents.Events[0].Event) {
-		// not allocated, but phones home
-		return true
-	}
-	if m.RecentProvisioningEvents.CrashLoop || m.RecentProvisioningEvents.FailedMachineReclaim {
-		// Machines in crash loop but in "Waiting" state are considered available
-		if len(m.RecentProvisioningEvents.Events) > 0 && !metal.ProvisioningEventWaiting.Is(m.RecentProvisioningEvents.Events[0].Event) {
-			return true
-		}
-	}
-
-	return false
-}
-
 func makeMachineResponse(m *metal.Machine, ds *datastore.RethinkStore) (*v1.MachineResponse, error) {
 	s, p, i, ec, err := findMachineReferencedEntities(m, ds)
 	if err != nil {