Skip to content

Commit

Permalink
Support CanaryAutoPauseMaxSlowStartDuration option (#997)
Browse files Browse the repository at this point in the history 
* Add support for max slow start duration config param

* Update max slow start arg name to include auto pause

* fixup! Update max slow start arg name to include auto pause
  • Loading branch information
@jennchenn
jennchenn authored Nov 22, 2023
1 parent 492afcc commit 04d86b1
Show file tree
Hide file tree
Showing 3 changed files with 60 additions and 49 deletions.
17 changes: 11 additions & 6 deletions controllers/datadogagent/component/agent/new.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,12 +64,13 @@ type ExtendedDaemonsetOptions struct {
MaxPodUnavailable string
MaxPodSchedulerFailure string

CanaryDuration time.Duration
CanaryReplicas string
CanaryAutoPauseEnabled bool
CanaryAutoPauseMaxRestarts int32
CanaryAutoFailEnabled bool
CanaryAutoFailMaxRestarts int32
CanaryDuration time.Duration
CanaryReplicas string
CanaryAutoPauseEnabled bool
CanaryAutoPauseMaxRestarts int32
CanaryAutoFailEnabled bool
CanaryAutoFailMaxRestarts int32
CanaryAutoPauseMaxSlowStartDuration time.Duration
}

func defaultEDSSpec(options *ExtendedDaemonsetOptions) edsv1alpha1.ExtendedDaemonSetSpec {
Expand Down Expand Up @@ -101,6 +102,10 @@ func defaultEDSSpec(options *ExtendedDaemonsetOptions) edsv1alpha1.ExtendedDaemo
spec.Strategy.Canary.AutoFail.MaxRestarts = edsv1alpha1.NewInt32(options.CanaryAutoFailMaxRestarts)
}

if options.CanaryAutoPauseMaxSlowStartDuration != 0 {
spec.Strategy.Canary.AutoPause.MaxSlowStartDuration = &metav1.Duration{Duration: options.CanaryAutoPauseMaxSlowStartDuration}
}

spec.Strategy.Canary.AutoPause.Enabled = edsv1alpha1.NewBool(options.CanaryAutoPauseEnabled)
if options.CanaryAutoPauseMaxRestarts > 0 {
spec.Strategy.Canary.AutoPause.MaxRestarts = edsv1alpha1.NewInt32(options.CanaryAutoPauseMaxRestarts)
Expand Down
32 changes: 17 additions & 15 deletions controllers/setup.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -49,12 +49,13 @@ type ExtendedDaemonsetOptions struct {
MaxPodUnavailable string
MaxPodSchedulerFailure string

CanaryDuration time.Duration
CanaryReplicas string
CanaryAutoPauseEnabled bool
CanaryAutoPauseMaxRestarts int
CanaryAutoFailEnabled bool
CanaryAutoFailMaxRestarts int
CanaryDuration time.Duration
CanaryReplicas string
CanaryAutoPauseEnabled bool
CanaryAutoPauseMaxRestarts int
CanaryAutoFailEnabled bool
CanaryAutoFailMaxRestarts int
CanaryAutoPauseMaxSlowStartDuration time.Duration
}

type starterFunc func(logr.Logger, manager.Manager, *version.Info, kubernetes.PlatformInfo, SetupOptions) error
Expand Down Expand Up @@ -122,15 +123,16 @@ func startDatadogAgent(logger logr.Logger, mgr manager.Manager, vInfo *version.I
Recorder: mgr.GetEventRecorderFor(agentControllerName),
Options: datadogagent.ReconcilerOptions{
ExtendedDaemonsetOptions: componentagent.ExtendedDaemonsetOptions{
Enabled: options.SupportExtendedDaemonset.Enabled,
MaxPodUnavailable: options.SupportExtendedDaemonset.MaxPodUnavailable,
MaxPodSchedulerFailure: options.SupportExtendedDaemonset.MaxPodSchedulerFailure,
CanaryDuration: options.SupportExtendedDaemonset.CanaryDuration,
CanaryReplicas: options.SupportExtendedDaemonset.CanaryReplicas,
CanaryAutoPauseEnabled: options.SupportExtendedDaemonset.CanaryAutoPauseEnabled,
CanaryAutoPauseMaxRestarts: int32(options.SupportExtendedDaemonset.CanaryAutoPauseMaxRestarts),
CanaryAutoFailEnabled: options.SupportExtendedDaemonset.CanaryAutoFailEnabled,
CanaryAutoFailMaxRestarts: int32(options.SupportExtendedDaemonset.CanaryAutoFailMaxRestarts),
Enabled: options.SupportExtendedDaemonset.Enabled,
MaxPodUnavailable: options.SupportExtendedDaemonset.MaxPodUnavailable,
MaxPodSchedulerFailure: options.SupportExtendedDaemonset.MaxPodSchedulerFailure,
CanaryDuration: options.SupportExtendedDaemonset.CanaryDuration,
CanaryReplicas: options.SupportExtendedDaemonset.CanaryReplicas,
CanaryAutoPauseEnabled: options.SupportExtendedDaemonset.CanaryAutoPauseEnabled,
CanaryAutoPauseMaxRestarts: int32(options.SupportExtendedDaemonset.CanaryAutoPauseMaxRestarts),
CanaryAutoPauseMaxSlowStartDuration: options.SupportExtendedDaemonset.CanaryAutoPauseMaxSlowStartDuration,
CanaryAutoFailEnabled: options.SupportExtendedDaemonset.CanaryAutoFailEnabled,
CanaryAutoFailMaxRestarts: int32(options.SupportExtendedDaemonset.CanaryAutoFailMaxRestarts),
},
SupportCilium: options.SupportCilium,
OperatorMetricsEnabled: options.OperatorMetricsEnabled,
Expand Down
60 changes: 32 additions & 28 deletions main.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -82,8 +82,9 @@ const (
defaultCanaryAutoPauseEnabled = true
defaultCanaryAutoFailEnabled = true
// default to 0, to use default value from EDS.
defaultCanaryAutoPauseMaxRestarts = 0
defaultCanaryAutoFailMaxRestarts = 0
defaultCanaryAutoPauseMaxRestarts = 0
defaultCanaryAutoFailMaxRestarts = 0
defaultCanaryAutoPauseMaxSlowStartDuration = 0
)

type options struct {
Expand All @@ -101,23 +102,24 @@ type options struct {
leaderElectionLeaseDuration time.Duration

// Controllers options
supportExtendedDaemonset bool
edsMaxPodUnavailable string
edsMaxPodSchedulerFailure string
edsCanaryDuration time.Duration
edsCanaryReplicas string
edsCanaryAutoPauseEnabled bool
edsCanaryAutoPauseMaxRestarts int
edsCanaryAutoFailEnabled bool
edsCanaryAutoFailMaxRestarts int
supportCilium bool
datadogAgentEnabled bool
datadogMonitorEnabled bool
datadogSLOEnabled bool
operatorMetricsEnabled bool
webhookEnabled bool
v2APIEnabled bool
maximumGoroutines int
supportExtendedDaemonset bool
edsMaxPodUnavailable string
edsMaxPodSchedulerFailure string
edsCanaryDuration time.Duration
edsCanaryReplicas string
edsCanaryAutoPauseEnabled bool
edsCanaryAutoPauseMaxRestarts int
edsCanaryAutoFailEnabled bool
edsCanaryAutoFailMaxRestarts int
edsCanaryAutoPauseMaxSlowStartDuration time.Duration
supportCilium bool
datadogAgentEnabled bool
datadogMonitorEnabled bool
datadogSLOEnabled bool
operatorMetricsEnabled bool
webhookEnabled bool
v2APIEnabled bool
maximumGoroutines int

// Secret Backend options
secretBackendCommand string
Expand Down Expand Up @@ -161,6 +163,7 @@ func (opts *options) Parse() {
flag.IntVar(&opts.edsCanaryAutoPauseMaxRestarts, "edsCanaryAutoPauseMaxRestarts", defaultCanaryAutoPauseMaxRestarts, "ExtendedDaemonset canary auto pause max restart count")
flag.BoolVar(&opts.edsCanaryAutoFailEnabled, "edsCanaryAutoFailEnabled", defaultCanaryAutoFailEnabled, "ExtendedDaemonset canary auto fail enabled")
flag.IntVar(&opts.edsCanaryAutoFailMaxRestarts, "edsCanaryAutoFailMaxRestarts", defaultCanaryAutoFailMaxRestarts, "ExtendedDaemonset canary auto fail max restart count")
flag.DurationVar(&opts.edsCanaryAutoPauseMaxSlowStartDuration, "edsCanaryAutoPauseMaxSlowStartDuration", defaultCanaryAutoPauseMaxSlowStartDuration*time.Minute, "ExtendedDaemonset canary max slow start duration")

// Parsing flags
flag.Parse()
Expand Down Expand Up @@ -247,15 +250,16 @@ func run(opts *options) error {

options := controllers.SetupOptions{
SupportExtendedDaemonset: controllers.ExtendedDaemonsetOptions{
Enabled: opts.supportExtendedDaemonset,
MaxPodUnavailable: opts.edsMaxPodUnavailable,
CanaryDuration: opts.edsCanaryDuration,
CanaryReplicas: opts.edsCanaryReplicas,
CanaryAutoPauseEnabled: opts.edsCanaryAutoPauseEnabled,
CanaryAutoPauseMaxRestarts: opts.edsCanaryAutoPauseMaxRestarts,
CanaryAutoFailEnabled: opts.edsCanaryAutoFailEnabled,
CanaryAutoFailMaxRestarts: opts.edsCanaryAutoFailMaxRestarts,
MaxPodSchedulerFailure: opts.edsMaxPodSchedulerFailure,
Enabled: opts.supportExtendedDaemonset,
MaxPodUnavailable: opts.edsMaxPodUnavailable,
CanaryDuration: opts.edsCanaryDuration,
CanaryReplicas: opts.edsCanaryReplicas,
CanaryAutoPauseEnabled: opts.edsCanaryAutoPauseEnabled,
CanaryAutoPauseMaxRestarts: opts.edsCanaryAutoPauseMaxRestarts,
CanaryAutoFailEnabled: opts.edsCanaryAutoFailEnabled,
CanaryAutoFailMaxRestarts: opts.edsCanaryAutoFailMaxRestarts,
CanaryAutoPauseMaxSlowStartDuration: opts.edsCanaryAutoPauseMaxSlowStartDuration,
MaxPodSchedulerFailure: opts.edsMaxPodSchedulerFailure,
},
SupportCilium: opts.supportCilium,
Creds: creds,
Expand Down

0 comments on commit 04d86b1

Please sign in to comment.