resultcleanup

package
v0.2.13 Latest Latest
Warning

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

Go to latest
Published: Nov 15, 2026 License: MIT Imports: 22 Imported by: 0

Documentation

Overview

Package resultcleanup retires and collects runner evidence through direct API reads. It never deletes plans or reads/deletes Secret projections.

Index

Constants

This section is empty.

Variables

View Source
var ErrRetained = errors.New("result evidence is not eligible for collection")

Functions

func RootBinding

func RootBinding(record *api.PtahResultRecord) (resultstore.Binding, error)

Types

type Collector

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

Collector scans metadata in bounded pages outside both reconciliation workers. It keeps no payload cache. The API's required admission guard authorizes each UID-bound DELETE with current reads; the collector does not repeat that entire verdict before sending the request. Only the leader scans.

func New

func New(writer client.Client, policy Policy, registry prometheus.Registerer, recorder EventRecorder) (*Collector, error)

New builds the collector. recorder receives the Warning Events that explain a retirement a ResourceQuota refused; a nil recorder records none.

func (*Collector) NeedLeaderElection

func (*Collector) NeedLeaderElection() bool

func (*Collector) Start

func (c *Collector) Start(ctx context.Context) error

func (*Collector) Step

func (c *Collector) Step(ctx context.Context) error

Step visits at most one 64-object metadata page and attempts at most ten DELETEs, retaining the unfinished page across a deadline or write budget. Errors on one attempt do not starve later attempts. Start calls it serially; tests can advance a complete scan without sleeping.

type EventRecorder added in v0.2.11

type EventRecorder interface {
	Eventf(object runtime.Object, eventtype, reason, messageFmt string, args ...any)
}

EventRecorder is the part of client-go's record.EventRecorder the collector uses.

type Policy

type Policy struct {
	Reader client.Reader
	Window time.Duration
	// Now is injectable for tests; production uses the wall clock.
	Now func() time.Time
}

func (Policy) AuthorizeDelete

func (p Policy) AuthorizeDelete(ctx context.Context, record *api.PtahResultRecord) error

AuthorizeDelete is the admission verdict, regardless of the DELETE caller. record is the API server's stored admission object. Neither a collector's cached decision nor an administrator's identity replaces these current checks. Every API failure retains the evidence. A refusal that is about time is a resultretention.WindowError; every other refusal names no instant, because none would lift it.

func (Policy) AuthorizeForegroundRetirement

func (p Policy) AuthorizeForegroundRetirement(ctx context.Context, record *api.PtahResultRecord) error

AuthorizeForegroundRetirement permits only the start of a foreground deletion. Its API-controlled timestamp supplies the retirement clock when the API refuses a separate marker: a terminating namespace refuses every CREATE, and a full ResourceQuota on result records refuses this one. Nothing here reads why, because the verdict is the same either way: the namespace's phase and its quotas say nothing about the evidence. Admission protects the foreground finalizer until AuthorizeDelete permits destruction of bytes.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL