soft_quay/core/installer/switcher.go

package installer

import (
	"errors"
	"fmt"
	"os"
)

var (
	ErrHealthCheckRequired = errors.New("health check is required")
	ErrHealthCheckFailed   = errors.New("installed version failed health check")
	ErrRollbackFailed      = errors.New("install rollback failed")
)

type HealthCheck func(currentPath string) error

// PreSwitchCheck runs immediately before replacing an existing current
// directory. It is intentionally core-only so callers can inject process
// state without bringing platform APIs into the transaction.
type PreSwitchCheck func() error

type switchStep string

const (
	stepPrepared         switchStep = "prepared"
	stepCurrentRenamed   switchStep = "current_renamed"
	stepCurrentBackedUp  switchStep = "current_backed_up"
	stepStagingRenamed   switchStep = "staging_renamed"
	stepStagingActivated switchStep = "staging_activated"
	stepRollbackRequired switchStep = "rollback_required"
	stepCommitted        switchStep = "committed"
)

// RollbackError reports both the health failure and rollback failure.
type RollbackError struct {
	Health   error
	Rollback error
}

func (err *RollbackError) Error() string {
	return fmt.Sprintf("%s: health=%v; rollback=%v", ErrRollbackFailed, err.Health, err.Rollback)
}

func (err *RollbackError) Unwrap() []error {
	return []error{ErrRollbackFailed, err.Health, err.Rollback}
}

// Switcher activates a verified staging directory and runs an injected check.
type Switcher struct {
	health         HealthCheck
	preSwitchCheck PreSwitchCheck
	afterStep      func(switchStep) error
	durability     durabilityFence
}

func NewSwitcher(health HealthCheck) *Switcher {
	return NewSwitcherWithPreSwitchCheck(health, nil)
}

// NewSwitcherWithPreSwitchCheck creates a switcher that performs the optional
// check only for updates with an existing current directory.
func NewSwitcherWithPreSwitchCheck(
	health HealthCheck,
	preSwitchCheck PreSwitchCheck,
) *Switcher {
	return &Switcher{
		health:         health,
		preSwitchCheck: preSwitchCheck,
		durability:     defaultDurability(),
	}
}

func (switcher *Switcher) Switch(root string) error {
	if switcher.health == nil {
		return ErrHealthCheckRequired
	}
	fence := effectiveDurability(switcher.durability)
	layout, err := inspectAppLayout(root)
	if err != nil {
		return err
	}
	if _, exists, err := loadTransaction(layout); err != nil {
		return err
	} else if exists {
		return ErrRecoveryRequired
	}
	state, err := inspectDirectories(layout)
	if err != nil {
		return err
	}
	if !state.staging {
		return ErrStagingMissing
	}
	if state.backup {
		return ErrBackupExists
	}
	if state.current && switcher.preSwitchCheck != nil {
		if err := switcher.preSwitchCheck(); err != nil {
			if cleanupErr := removeManagedDirectoryWithFence(layout, layout.staging, fence); cleanupErr != nil {
				return errors.Join(err, cleanupErr)
			}
			return err
		}
	}

	record := newTransaction(phasePrepared, state.current)
	if err := writeTransactionWithFence(layout, record, fence); err != nil {
		return err
	}
	if err := switcher.runStep(stepPrepared); err != nil {
		return err
	}

	if state.current {
		if err := renameManagedDirectory(
			layout,
			layout.current,
			layout.backup,
			fence,
			"app root after current backup rename",
		); err != nil {
			return fmt.Errorf("backup current directory: %w", err)
		}
		if err := switcher.runStep(stepCurrentRenamed); err != nil {
			return err
		}
	}
	record.Phase = phaseCurrentBackedUp
	if err := writeTransactionWithFence(layout, record, fence); err != nil {
		return err
	}
	if err := switcher.runStep(stepCurrentBackedUp); err != nil {
		return err
	}

	if err := renameManagedDirectory(
		layout,
		layout.staging,
		layout.current,
		fence,
		"app root after staging activation rename",
	); err != nil {
		return fmt.Errorf("activate staging directory: %w", err)
	}
	if err := switcher.runStep(stepStagingRenamed); err != nil {
		return err
	}
	record.Phase = phaseStagingActivated
	if err := writeTransactionWithFence(layout, record, fence); err != nil {
		return err
	}
	if err := switcher.runStep(stepStagingActivated); err != nil {
		return err
	}

	healthErr := switcher.health(layout.current)
	if healthErr != nil {
		record.Phase = phaseRollbackRequired
		if err := writeTransactionWithFence(layout, record, fence); err != nil {
			return &RollbackError{Health: healthErr, Rollback: err}
		}
		if err := switcher.runStep(stepRollbackRequired); err != nil {
			return err
		}
		if err := rollbackActivated(layout, record.HadCurrent, fence); err != nil {
			return &RollbackError{Health: healthErr, Rollback: err}
		}
		if err := removeTransactionWithFence(layout, fence); err != nil {
			return &RollbackError{Health: healthErr, Rollback: err}
		}
		return fmt.Errorf("%w: %w", ErrHealthCheckFailed, healthErr)
	}

	record.Phase = phaseCommitted
	if err := writeTransactionWithFence(layout, record, fence); err != nil {
		return err
	}
	if err := switcher.runStep(stepCommitted); err != nil {
		return err
	}
	if record.HadCurrent {
		if err := removeManagedDirectoryWithFence(layout, layout.backup, fence); err != nil {
			return fmt.Errorf("%w: cleanup committed backup: %w", ErrRecoveryRequired, err)
		}
	}
	if err := removeTransactionWithFence(layout, fence); err != nil {
		return fmt.Errorf("%w: %w", ErrRecoveryRequired, err)
	}
	return nil
}

func (switcher *Switcher) runStep(step switchStep) error {
	if switcher.afterStep == nil {
		return nil
	}
	return switcher.afterStep(step)
}

func rollbackActivated(layout appLayout, hadCurrent bool, fence durabilityFence) error {
	state, err := inspectDirectories(layout)
	if err != nil {
		return err
	}
	if hadCurrent {
		if !state.backup {
			return fmt.Errorf("%w: previous current backup is missing", ErrRollbackFailed)
		}
		if state.current {
			if state.staging {
				return fmt.Errorf("%w: current and staging both exist", ErrRollbackFailed)
			}
			if err := renameManagedDirectory(
				layout,
				layout.current,
				layout.staging,
				fence,
				"app root after rollback staging rename",
			); err != nil {
				return fmt.Errorf("move failed current aside: %w", err)
			}
		}
		if err := renameManagedDirectory(
			layout,
			layout.backup,
			layout.current,
			fence,
			"app root after rollback current restore",
		); err != nil {
			if _, statErr := os.Stat(layout.staging); statErr == nil {
				_ = renameManagedDirectory(
					layout,
					layout.staging,
					layout.current,
					fence,
					"app root after rollback restore",
				)
			}
			return fmt.Errorf("restore previous current: %w", err)
		}
		return removeManagedDirectoryWithFence(layout, layout.staging, fence)
	}

	if state.backup {
		return fmt.Errorf("%w: unexpected backup without previous current", ErrRollbackFailed)
	}
	if state.current {
		if state.staging {
			return fmt.Errorf("%w: current and staging both exist", ErrRollbackFailed)
		}
		if err := renameManagedDirectory(
			layout,
			layout.current,
			layout.staging,
			fence,
			"app root after initial rollback rename",
		); err != nil {
			return fmt.Errorf("move failed initial install aside: %w", err)
		}
	}
	return removeManagedDirectoryWithFence(layout, layout.staging, fence)
}