costmodel

package
v0.2.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: May 1, 2026 License: Apache-2.0 Imports: 24 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	Hourly  Period = 1
	Daily          = 24
	Weekly         = 24 * 7
	Monthly        = 24 * 30
	Yearly         = 24 * 365
)
View Source 
const CommentPrefix = "<!-- kost -->"

CommentPrefix is used to identify and hide old messages on GitHub.

Variables

View Source 
var (
	ErrNoResults          = errors.New("no cost results")
	ErrBadQuery           = errors.New("bad query")
	ErrNilConfig          = errors.New("client config is nil")
	ErrEmptyAddress       = errors.New("client address can't be empty")
	ErrProdConfigMissing  = errors.New("prod config is missing")
	ErrHPADetectionFailed = errors.New("hpa detection failed")
	ErrUnsupportedKind    = errors.New("unsupported workload kind")
)

ErrNoResults is the error returned when querying for costs returns no results.

View Source 
var ErrNoReports = errors.New("nothing to report")
View Source 
var ErrUnknownKind = errors.New("unknown kind")

ErrUnknownKind is the error throw when the kind of the resource in the manifest is unknown to the parser.

Functions

This section is empty.

Types

type Client 

type Client struct {
	// contains filtered or unexported fields
}

Client is a client for the cost model.

func NewClient 

func NewClient(config *ClientConfig) (*Client, error)

NewClient creates a new cost model client with the given configuration.

func (*Client) GetCostForPersistentVolume 

func (c *Client) GetCostForPersistentVolume(ctx context.Context, cluster string) (Cost, error)

GetCostForPersistentVolume returns the average cost per persistent volume for a given cluster

func (*Client) GetCostPerCPU 

func (c *Client) GetCostPerCPU(ctx context.Context, cluster string) (Cost, error)

GetCostPerCPU returns the average cost per CPU for a given cluster.

func (*Client) GetMemoryCost 

func (c *Client) GetMemoryCost(ctx context.Context, cluster string) (Cost, error)

GetMemoryCost returns the cost per memory for a given cluster

func (*Client) GetNodeCount 

func (c *Client) GetNodeCount(ctx context.Context, cluster string) (int, error)

GetNodeCount returns the average number of nodes over 30 days for a given cluster

func (*Client) GetObservedReplicas added in v0.2.0 

func (c *Client) GetObservedReplicas(ctx context.Context, cluster, namespace, kind, name string) (float64, error)

GetObservedReplicas returns the 7-day average replica count for the given workload from kube-state-metrics, regardless of who scales it (HPA, KEDA, manual). Used to substitute manifest replicas with reality for HPA-managed workloads. Returns ErrUnsupportedKind for kinds without a replica-count metric (DaemonSet, Pod, Job). Returns ErrNoResults if the query succeeds but has no samples (e.g., brand-new workload with no history) — callers should surface this rather than silently fall back.

func (*Client) HPATargeting added in v0.2.0 

func (c *Client) HPATargeting(ctx context.Context, cluster, namespace, kind, name string) (string, error)

HPATargeting returns the name of the HorizontalPodAutoscaler targeting the given (cluster, namespace, kind, name) workload, or an empty string if no HPA targets it. Works for vanilla HPAs and KEDA-managed HPAs (KEDA creates a regular HPA underneath). Transport or unexpected response shape errors are wrapped with ErrHPADetectionFailed so callers can distinguish "definitely not HPA-managed" from "we couldn't tell."

type ClientConfig 

type ClientConfig struct {
	Address        string
	HTTPConfigFile string
	Username       string
	Password       string
}

ClientConfig is the configuration for the cost model client.

type Clients 

type Clients struct {
	Prod *Client
	Dev  *Client
}

Clients bundles the dev and prod client in one struct.

func NewClients 

func NewClients(prodConfig, devConfig *ClientConfig) (*Clients, error)

NewClients creates a new cost model clients with the given configuration.

func (*Clients) GetClusterCosts 

func (c *Clients) GetClusterCosts(ctx context.Context, cluster string) (*CostModel, error)

GetClusterCosts returns the cost for a cluster and differentiate for dev and prod clusters

func (*Clients) GetObservedReplicas added in v0.2.0 

func (c *Clients) GetObservedReplicas(ctx context.Context, cluster, namespace, kind, name string) (float64, error)

GetObservedReplicas routes to the appropriate prod/dev client based on cluster prefix.

func (*Clients) HPATargeting added in v0.2.0 

func (c *Clients) HPATargeting(ctx context.Context, cluster, namespace, kind, name string) (string, error)

HPATargeting routes to the appropriate prod/dev client based on cluster prefix.

type Cluster 

type Cluster struct {
	Name      string
	NodeCount int
}

type Cost 

type Cost struct {
	Dollars float64
	Spot    float64
	NonSpot float64
}

Cost represents the _hourly_ cost of a resource in USD. If the cluster does not have pricing data for spot nodes, then Dollars will be set.

func (Cost) DollarsForPeriod 

func (c Cost) DollarsForPeriod(p Period, r int64) float64

DollarsForPeriod returns the cost of a resource in USD for a given period. Primarily used by PersistentVolumeClaims which do not have spot/non spot pricing.

func (Cost) DollarsYearly 

func (c Cost) DollarsYearly(memReq int64) float64

func (Cost) NonSpotCPUForPeriod 

func (c Cost) NonSpotCPUForPeriod(p Period, r int64) float64

func (Cost) NonSpotMemoryForPeriod 

func (c Cost) NonSpotMemoryForPeriod(p Period, r int64) float64

func (Cost) NonSpotYearly 

func (c Cost) NonSpotYearly(cpuReq int64) float64

func (Cost) SpotCPUForPeriod 

func (c Cost) SpotCPUForPeriod(p Period, r int64) float64

func (Cost) SpotMemoryForPeriod 

func (c Cost) SpotMemoryForPeriod(p Period, r int64) float64

func (Cost) SpotYearly 

func (c Cost) SpotYearly(cpuReq int64) float64

type CostModel 

type CostModel struct {
	Cluster          *Cluster
	CPU              Cost
	RAM              Cost
	PersistentVolume Cost
}

CostModel represents the cost of each resource for a specific cluster

func GetCostModelForCluster 

func GetCostModelForCluster(ctx context.Context, client *Client, cluster string) (*CostModel, error)

func (*CostModel) TotalCostForPeriod 

func (c *CostModel) TotalCostForPeriod(p Period, r Requirements) float64

TotalCostForPeriod calculates the costs of each resource on the CostModel and returns the sum of the costs

type HPAResolver added in v0.2.0 

type HPAResolver interface {
	HPATargeting(ctx context.Context, cluster, namespace, kind, name string) (string, error)
	GetObservedReplicas(ctx context.Context, cluster, namespace, kind, name string) (float64, error)
}

HPAResolver is the subset of *Client behavior ResolveReplicas needs. Lets policy be tested without httptest.

type Period 

type Period float64

func (Period) Keys 

func (p Period) Keys() PeriodKeys

func (Period) String 

func (p Period) String() string

type PeriodKeys 

type PeriodKeys struct {
	To    string
	Delta string
	From  string
}

type ReplicaSource added in v0.2.0 

type ReplicaSource int

ReplicaSource describes which signal ResolveReplicas used. 

const (
	// SourceManifest indicates the manifest's spec.replicas was authoritative
	// (no HPA targets the workload).
	SourceManifest ReplicaSource = iota
	// SourceObservedHPA indicates an HPA targets the workload and the manifest
	// replicas were replaced with the observed average from kube-state-metrics.
	SourceObservedHPA
)

func ResolveReplicas added in v0.2.0 

func ResolveReplicas(ctx context.Context, r HPAResolver, cluster, namespace, kind, name string, manifestReplicas int) (int, ReplicaSource, error)

ResolveReplicas decides the right replica count for a workload at PR-cost-estimation time. If no HPA targets the workload, the manifest count is authoritative. If an HPA is detected, the observed 7d-average from kube-state-metrics is substituted (rounded to nearest int) — manifest replicas are a lie under HPA control. Errors from either query are returned so the caller can surface them rather than silently fall back to a wrong number.

type ReportType 

type ReportType string
const (
	Table    ReportType = "table"
	Summary  ReportType = "summary"
	Markdown ReportType = "markdown"
)

type Reporter 

type Reporter struct {
	Writer io.Writer
	// contains filtered or unexported fields
}

func New 

func New(w io.Writer, reportType string) *Reporter

func (*Reporter) AddError added in v0.2.0 

func (r *Reporter) AddError(msg string)

AddError records a message about an unexpected event that may have led to inaccurate cost numbers. Surfaced under the Errors section in the markdown report.

func (*Reporter) AddReport 

func (r *Reporter) AddReport(costModel *CostModel, from, to Requirements)

AddReport adds a costmodel and associated from, to resources to the reporter.

func (*Reporter) AddReportWithResolvedReplicas added in v0.2.0 

func (r *Reporter) AddReportWithResolvedReplicas(
	ctx context.Context,
	resolver HPAResolver,
	cm *CostModel,
	from, to Requirements,
)

AddReportWithResolvedReplicas adds a cost report after substituting the manifest replica count with the observed 7d-average for HPA-managed Deployment/StatefulSet workloads. For unsupported kinds (DaemonSet, Pod, Job, CronJob) or when the CostModel cannot be used to identify a cluster, falls back to AddReport with the manifest replica count.

On substitution, an audit-trail Warning is added to the reporter so reviewers can see that observed replicas were used. On resolver errors, the manifest count is preserved and an Error is added so misleading numbers are flagged rather than silently presented.

func (*Reporter) AddWarning added in v0.2.0 

func (r *Reporter) AddWarning(msg string)

AddWarning records a non-fatal message that will be surfaced in the markdown report under the Warnings section. Use for expected events or known limitations (e.g., observed-replicas substitution applied for an HPA-managed workload).

func (*Reporter) Write 

func (r *Reporter) Write() error

type Requirements 

type Requirements struct {
	CPUPerPod              int64
	MemoryPerPod           int64
	PersistentVolumePerPod int64
	Replicas               int
	Kind                   string
	Namespace              string
	Name                   string
}

Requirements holds the per-pod resource requirements parsed from a manifest, plus the replica count needed to compute aggregate cost. CPUPerPod is in millicores; MemoryPerPod and PersistentVolumePerPod are in bytes. Each PerPod field is the sum across all containers (or PVC templates) in a single pod. Use TotalCPU / TotalMemory / TotalPersistentVolume to get aggregate values across replicas.

func Delta 

func Delta(from, to Requirements) Requirements

Delta returns the field-wise difference between two resources. A positive value signals that the resource has increased. A negative value signals that the resource has decreased.

func ParseManifest 

func ParseManifest(src []byte, costModel *CostModel) (Requirements, error)

ParseManifest will parse a manifest file and return the aggregated amount of resources requested. The manifest can be a Deployment, StatefulSet, DaemonSet, Cronjob, Job, or Pod. If the manifest has the number of Replicas, the total resources will be multiplied by the number of replicas.

func (*Requirements) AddRequirements 

func (r *Requirements) AddRequirements(reqs corev1.ResourceRequirements)

AddRequirements increments the per-pod resources by the amount specified.

func (Requirements) TotalCPU added in v0.2.0 

func (r Requirements) TotalCPU() int64

TotalCPU returns aggregate CPU (millicores) across all replicas.

func (Requirements) TotalMemory added in v0.2.0 

func (r Requirements) TotalMemory() int64

TotalMemory returns aggregate memory (bytes) across all replicas.

func (Requirements) TotalPersistentVolume added in v0.2.0 

func (r Requirements) TotalPersistentVolume() int64

TotalPersistentVolume returns aggregate persistent volume (bytes) across all replicas.

Source Files

View all Source files

Directories

Path Synopsis

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.