Skip to content
Docs
Release candidateThese docs are for Napwake 0.2.0-rc.9, a release candidate, and may change before it's released. Read the docs for the latest release.

Reference

API reference

Every field of the ManagedWorkload resource: its type, its default and what it does.

View .mdAsk Claude (opens claude.ai)Ask ChatGPT (opens chatgpt.com)
Group
napwake.dev
Version
v1alpha1
Kind
ManagedWorkload
Scope
Namespaced

Columns

Section titled “Columns”

What kubectl get managedworkloads shows.

Column Shows
Phase The phase
Freed What pausing freed this month
Saved What of that it saved
Would Free What dry-run measured it would free. Blank outside dry-run
Age Age

Spec (ManagedWorkloadSpec)

Section titled “Spec (ManagedWorkloadSpec)”
Field Type Default Description
#target WorkloadRef required The workload. Can’t be changed once set
#target.kind Deployment or StatefulSet Deployment Its kind
#target.name string required Its name, same namespace
#idlePolicy IdlePolicySpec Idle detection. Without it, never paused automatically
#idlePolicy.idleAfter duration, at least 1m 1h Time without activity before pausing
#idlePolicy.activity.cpuThreshold int or string, 0.1-100, to two places 10 CPU % of requests above which it’s active. A decimal is written as a string, such as "0.5"
#idlePolicy.activity.prometheus[].promQL string required in each entry PromQL instant query. Above zero is activity. At most 10
#idlePolicy.autoResume bool false Wake ahead of confident forecast demand
#idlePolicy.pauseEarly bool false Pause after 20 idle minutes, before idleAfter, if a confident forecast expects quiet until then and the doorman would wake it. See Pausing early
#dependsOn[] DependencyRef Workloads this one needs. At most 20
#dependsOn[].namespace string this namespace Its namespace
#dependsOn[].kind Deployment or StatefulSet required Its kind
#dependsOn[].name string required Its name
#dependsOn[].waitForReady bool false Hold this one’s resume until it’s Ready
#wake WakeSpec What wakes it through the doorman while paused
#wake.mode Request or Ask Request What wakes it through the doorman. Request is any request but a known probe’s. Ask is only a person’s click on Wake it, and anything else gets a 503. See Asking before waking
#wake.maxWait duration, at most 1h 2m How long a request is held while it wakes
#wake.page bool true Answer a browser’s page load with the waking-up page, instead of holding it. With mode: Ask, a browser always gets a page
#prediction PredictionSpec {} The forecast. Left out or {}, it takes the defaults
#prediction.mode eager, balanced or cautious balanced How soon the forecast acts, at confidence 60, 75 or 85. See Forecast setting
#prediction.confidence int, 50-100 the mode’s Accuracy (1 − WAPE, %) a season’s forecasts must reach to drive decisions. Overrides the mode’s
#costTracking CostTrackingSpec Your own cost rates
#costTracking.rates CostRates node list prices, then AWS defaults Custom rates, none negative
#dryRun bool false Measure without pausing. Turning it on wakes a paused workload. See Dry run

Durations

Section titled “Durations”
Duration
45s, 90m, 1h30m Go durations above zero work
1d, 2w, spaces Days, weeks and spaces don’t. Write 24h or 168h
Anything else The API server rejects it

CostRates

Section titled “CostRates”

For costTracking.rates and cost.listRates.

Field Type Default Description
#cpuPerHour quantity 0.031 USD per vCPU-hour
#memoryPerHour quantity 0.004 USD per GiB-hour
#storagePerMonth quantity 0.08 USD per GiB-month

Status (ManagedWorkloadStatus)

Section titled “Status (ManagedWorkloadStatus)”
Field Type Description
#phase WorkloadPhase Current lifecycle phase
#conditions[] Condition Standard Kubernetes conditions. See Conditions and events
#lastActedAt time Last pause or wake, by Napwake or a scale-up from zero outside it
#lastPauseRequest string Last napwake.dev/pause-requested value acted on, so each acts once. Outcome in the PauseRequest condition. See Pause now
#lastWakeRequest string Last napwake.dev/wake-requested value acted on, so each acts once. Outcome in the WakeRequest condition. See Wake now
#turnedAway TurnedAway With wake.mode: Ask, what the doorman turned away during the current pause, or the last once it’s awake. pausedAt, when that pause began, and requests. Brought up to date about once a minute. See What it turns away
#lastTransitionTime time Last phase change
#learnedDependencies LearnedDependencies Found by Napwake, held and woken like dependsOn. Fields from, at and dependencies, each with namespace, kind, name and source, which is environment (with via, address) or wake (with lastSeen). See Learned dependencies
#lastScaledUp ScaledUp Last scale-up outside Napwake while paused, which wakes it. Fields at, replicas, by (field manager, such as argocd-controller, or unknown through the scale subresource, as kubectl scale, which records none) and gitOps (Argo CD, Flux, Fleet, kapp or Kluctl, if from Git). See GitOps tools
#doorman[] DoormanRoute Service ports routed to the doorman while paused, with service, portName, doormanPort and forwarded (behind GKE NEGs, through the forwarder)
#doormanDraining[] DrainingPort Doorman ports that routed it until it woke, with port and until. No other workload gets the port before until, so a late client never reaches another. Kept in status, so an operator that restarts or becomes leader honours it

activity

Section titled “activity”
Field Type Description
#activity.lastActivityTime time Latest activity, any source
#activity.lastActivitySource string created, woke, request, cpu, rollout, annotation, prometheus, unobserved or scaled-up
#activity.pauseAt time When it pauses without more activity
#activity.lastEvaluatedTime time Last activity check
#activity.templateHash string Pod template fingerprint, to detect deploys
#activity.heldByDependents []string Awake dependents keeping it from pausing (HeldByDependents), as namespace/name. At most 10, first by name
#activity.forecastBusyFrom time Start of the hour of expected demand, while the forecast holds off the pause (IdleVetoed)
#activity.forecastQuietUntil time End of the quiet hours expected when the forecast brought the pause forward (PausedEarly). Kept until the pause, or dry-run’s would-be pause, ends

pause

Section titled “pause”
Field Type Description
#pause PauseStatus What a pause changed, from before scaling to zero until running again
#pause.previousReplicas int32 Replicas before the pause, restored on resume
#pause.startedAt time When the pause began, as it went Pausing, before anything was scaled. Or when Napwake took up a pause whose record was lost. The asleep page’s Wake it button is good for this pause alone. See The Wake it button
#pause.routesWrittenAt time When the doorman’s routes for this pause were stored in status, which every doorman pod listens on. Its EndpointSlices go in 5 seconds after. Unset for a workload the doorman doesn’t route
#pause.routedAt time When the doorman’s EndpointSlices first stood in front of its Services for this pause, beside its pods. It’s scaled to zero a few seconds after. Unset for a workload the doorman doesn’t route. See What happens in a pause
#pause.requested bool The pause was asked for, with napwake.dev/pause-requested, not decided by the idle clock or the forecast
#pause.cpuBefore #pause.cpuJustBefore string The most CPU it used, as a percentage of its requests. cpuBefore is through the stretch without activity that ran its clock out. cpuJustBefore is in its last 15 minutes, or idleAfter if shorter. Unset when none was measured. See Paused while in use
#pause.pausedAt time When scaled to zero, unset before. Paused once its pods stop
#pause.resumedAt time When the current resume scaled it up. Scaled to zero after that, not by Napwake, it’s left off
#pause.firstReadyAt time When the current resume first saw a replica Ready. The rest get 2 minutes more
#pause.scaledObject string KEDA ScaledObject held at zero while paused. See HPA and KEDA
#pause.scaledObjectPausedReplicas string Its own autoscaling.keda.sh/paused-replicas before the pause, put back after
#pause.resources ResourceSnapshot What it ran as the pause began, the base for savings. replicas, per replica cpuMillis, memoryBytes, claims’ storageBytes, and nodePools (name, "" without a pool label, and its pods)
#pause.wakeAnnotations object Activity annotations on the ManagedWorkload (workload) and target (target) as the pause began. A change wakes it
#pause.queryActiveSincePause bool A Prometheus activity query has been active at every check since a requested pause began. They don’t wake it until a check finds all quiet

idleWait

Section titled “idleWait”
Field Type Description
#idleWait IdleWaitStatus Pauses made while it was in use, which lengthen its idle clock, and its CPU use idle. See Paused while in use
#idleWait.wrongPauses int32 Pauses in a row its own use undid soon after they began. Each doubles how long the idle clock waits, up to idleAfter plus an hour. A pause that holds sets it back to zero
#idleWait.lastWrongPause WrongPause The latest. pausedAt, after (how soon it was undone), by (request, click, wakeRequest or annotation), from (the caller’s address, for a request), and inUseCPU
#idleWait.idleCPU #idleWait.idleCPUAt string, time The most CPU it used, as a percentage of its requests, before its pauses that held, over the last week. idleCPUAt is when it was last measured

dryRun

Section titled “dryRun”
Field Type Description
#dryRun DryRunStatus What dry-run measured. Only while spec.dryRun is true
#dryRun.since time When measuring started
#dryRun.pauses int32 Would-be pauses, including one under way
#dryRun.slept duration Total length of finished would-be pauses
#dryRun.freedCPUHours #dryRun.freedMemoryHours quantity vCPU-hours and GiB-hours requested during slept
#dryRun.estimatedFreed string Those hours priced, for display, counting a would-be pause under way to the last status write. What pausing would have freed. The Would Free column
#dryRun.resources ResourceSnapshot What it ran as the current would-be pause began
#dryRun.queryActiveSincePause bool As pause.queryActiveSincePause, since the would-be pause under way was asked for. Queries don’t end it until all are quiet

prediction

Section titled “prediction”
Field Type Description
#prediction.dailyPhase #prediction.weeklyPhase string Each season’s phase, Observing, Suggesting or Active
#prediction.dailyConfidence #prediction.weeklyConfidence int Accuracy % over the last 24 and 168 observed hours
#prediction.timeZone string IANA zone for hours and weekdays, such as Europe/London. The operator’s --timezone (Helm value timezone)
#prediction.threshold int Confidence it acts at, spec.prediction.confidence or the mode’s
#prediction.acting bool Whether it drives decisions, holding off pauses and, with autoResume, waking ahead of demand
#prediction.startsActing time When it should start acting, if the coming days repeat the last week. Unset while acting, or if not within four weeks
#prediction.reason #prediction.message string Why it is or isn’t acting, as Acting, Observing, EarningConfidence, Relearning or NotLearning, and a sentence. See When it acts
#prediction.history PredictionHistory What it learned from Prometheus history when management began. Fields reason (Learned, NoPrometheusURL, NoHistory, NoCPURequests, QueryFailed), hours, from, to, message. See Learning from history
#prediction.state string Learned state, gzipped and base64-encoded

cost

Section titled “cost”
Field Type Description
#cost CostStatus This calendar month’s cost, in UTC. See Cost tracking
#cost.tracked duration How much of the month the totals cover
#cost.lastAccumulatedAt time When totals were last brought up to date
#cost.awakeCPUHours #cost.awakeMemoryHours quantity vCPU-hours and GiB-hours requested while awake
#cost.storageHours quantity GiB-hours of claims its pods mount, in every phase
#cost.pausedCPUHours #cost.pausedMemoryHours quantity vCPU-hours and GiB-hours the paused replicas had requested, while paused. What pausing freed
#cost.savedCPUHours #cost.savedMemoryHours quantity Its share of the paused requests its node pools’ spare capacity can’t hold, every 5 minutes. What pausing saved, at least. See How saved is measured
#cost.costThisMonth string Awake and storage hours, priced
#cost.freedThisMonth string Paused hours, priced
#cost.savedThisMonth string Saved hours, priced. unknown while savingsUnknown is set
#cost.savingsUnknown string Why saved can’t be measured, such as pods outside watchNamespaces being invisible. Unset when measured
#cost.costWithoutNapwakeThisMonth string costThisMonth plus freedThisMonth
#cost.projectedMonthlyCost #cost.projectedMonthlyFreed #cost.projectedMonthlySavings string The month so far, carried to a full month. pending until a day is tracked
#cost.running ResourceSnapshot What it ran when last brought up to date while awake
#cost.pricedAt #cost.pricedTemplateHash time, string When requests and list rates were last read, and the pod template then
#cost.listRates CostRates List rates of the nodes its pods last ran on, spot at spot. Unset without a list price
#cost.listPrice string Which list prices listRates are, OnDemand, Spot or SpotAndOnDemand. Unset with it
#cost.resourceReduction ResourceReduction What the current pause freed, as cpuMillis, memoryBytes and replicas. Only while paused

WorkloadPhase values

Section titled “WorkloadPhase values”
Value Meaning
#Running #Idle #Pausing #Paused #Resuming The phases. Lifecycle says what each means
No phase Until the first reconcile
Creating Accepted by the schema but never set

Next