Files
ThothII/tools/tht/internal/pi/restart.go
T

361 lines
13 KiB
Go

package pi
import (
"context"
"errors"
"fmt"
"os"
"path/filepath"
)
var (
ErrInterruptedRestart = errors.New("a previous Pi restart is incomplete; recover lifecycle maintenance before another Pi lifecycle operation")
errRestartImageDrift = errors.New("core image changed during Pi restart")
errRestartConfigurationDrift = errors.New("external endpoint configuration changed during Pi restart")
errRestartMountDrift = errors.New("core persistence mount contract changed during Pi restart")
errRestartConfirmation = &restartDiagnosticError{
message: "restart requires --yes after reviewing the planned Pi core recreation",
cause: ErrConfirmationRequired,
}
errRestartActiveSessions = &restartDiagnosticError{
message: "active sessions must be drained before restarting Pi; use --drain only after they are complete",
cause: ErrActiveSessions,
}
)
type restartDiagnosticError struct {
message string
cause error
}
func (e *restartDiagnosticError) Error() string { return e.message }
func (e *restartDiagnosticError) Unwrap() error { return e.cause }
type RestartRequest struct {
StatePath string
UpdateStatePath string
Confirm bool
Drain bool
}
type RestartResult struct {
StatePath string
Version string
}
func Restart(ctx context.Context, runner Runner, request RestartRequest) (RestartResult, error) {
return restartWithHooks(ctx, runner, request, defaultLifecycleHooks)
}
func restartWithHooks(
ctx context.Context,
runner Runner,
request RestartRequest,
hooks lifecycleHooks,
) (result RestartResult, retErr error) {
if err := validateRestartStatePaths(request.StatePath, request.UpdateStatePath); err != nil {
return RestartResult{}, err
}
lock, err := acquireLock(request.StatePath)
if err != nil {
return RestartResult{StatePath: request.StatePath}, err
}
defer lock.Release()
if !request.Confirm {
return RestartResult{StatePath: request.StatePath}, errRestartConfirmation
}
if state, err := readState(request.UpdateStatePath); err == nil && stateNeedsRecovery(state) {
return RestartResult{StatePath: request.StatePath}, ErrInterruptedUpdate
} else if err != nil && !errors.Is(err, os.ErrNotExist) {
return RestartResult{StatePath: request.StatePath}, err
}
if err := prepareLifecycleMutation(request.StatePath, hooks.removeFile); err != nil {
return RestartResult{StatePath: request.StatePath}, err
}
clearMaintenance := true
mutationStarted := false
var state State
defer func() {
if !clearMaintenance {
return
}
if clearErr := setMaintenance(context.Background(), runner, false); clearErr != nil {
result.StatePath = request.StatePath
if mutationStarted {
clearMaintenance = false
cause := clearErr
if writeErr := hooks.writeState(request.StatePath, state); writeErr != nil {
cause = errors.Join(cause, fmt.Errorf("restart recovery state could not be restored: %w", writeErr))
}
retErr = errors.Join(retErr, recoveryRequired("Pi restart maintenance cleanup failed", cause))
return
}
retErr = errors.Join(retErr, fmt.Errorf("maintenance admission gate could not be cleared: %w", clearErr))
}
}()
if err := setMaintenance(ctx, runner, true); err != nil {
return RestartResult{StatePath: request.StatePath}, err
}
if err := waitForInactiveSessions(ctx, runner, request.Drain, hooks.sleep); err != nil {
return RestartResult{StatePath: request.StatePath}, restartDiagnostic(err)
}
if err := Doctor(ctx, runner); err != nil {
return RestartResult{StatePath: request.StatePath}, err
}
version, err := Status(ctx, runner)
if err != nil {
return RestartResult{StatePath: request.StatePath}, err
}
configured, err := renderedCore(ctx, runner)
if err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
previous, err := runningImage(ctx, runner, configured.Reference)
if err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
previous.ConfigurationSHA = configured.ConfigurationSHA
transaction := lifecycleTransaction(request.StatePath)
previous.Reference = lifecycleImageTag(transaction, "restart")
if err := tagImage(ctx, runner, previous.ID, previous.Reference, "restart image pin"); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
state = State{
Transaction: transaction,
Phase: PhasePreflight,
Target: Target{Version: version, Source: "restart"},
Previous: previous,
}
if err := hooks.writeState(request.StatePath, state); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
overridePath := lifecycleOverridePath(request.StatePath, transaction)
if err := writeLifecycleOverride(overridePath, previous.Reference); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
lifecycle := composeOverrideRunner{Runner: runner, path: overridePath}
if running, err := activeSessions(ctx, runner); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
} else if running {
return RestartResult{StatePath: request.StatePath, Version: version}, errRestartActiveSessions
}
state.MutationStarted = true
if err := hooks.writeState(request.StatePath, state); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, err
}
mutationStarted = true
clearMaintenance = false
if err := recreateCoreWithoutImageChanges(ctx, lifecycle); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart core recreation failed", err)
}
if err := ensureMaintenance(ctx, lifecycle); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart maintenance proof failed", err)
}
state.Phase = PhaseRecreated
if err := hooks.writeState(request.StatePath, state); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart recreation state could not be recorded", err)
}
if err := verifyRestart(ctx, lifecycle, version, previous); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart verification failed", err)
}
state.Phase = PhaseVerified
if err := hooks.writeState(request.StatePath, state); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart verification state could not be recorded", err)
}
if err := hooks.removeFile(overridePath); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart image override could not be removed", err)
}
if err := hooks.removeFile(request.StatePath); err != nil {
return RestartResult{StatePath: request.StatePath, Version: version}, recoveryRequired("Pi restart recovery state could not be removed", err)
}
clearMaintenance = true
return RestartResult{StatePath: request.StatePath, Version: version}, nil
}
func recreateCoreWithoutImageChanges(ctx context.Context, runner Runner) error {
result, err := runCompose(
ctx,
runner,
"up",
"--detach",
"--wait",
"--wait-timeout",
"45",
"--no-deps",
"--force-recreate",
"--no-build",
"--pull",
"never",
"core",
)
if err != nil {
return commandError("core recreation", result, err)
}
return nil
}
func verifyRestart(ctx context.Context, runner Runner, wanted string, previous Image) error {
if err := Doctor(ctx, runner); err != nil {
return err
}
version, err := Status(ctx, runner)
if err != nil {
return err
}
if version != wanted {
return errors.New("Pi version changed during core restart")
}
configured, err := renderedCore(ctx, runner)
if err != nil {
return err
}
after, err := runningImage(ctx, runner, configured.Reference)
if err != nil {
return err
}
if after.ID != previous.ID {
return errRestartImageDrift
}
if configured.ConfigurationSHA != previous.ConfigurationSHA {
return errRestartConfigurationDrift
}
if !sameMounts(previous.Mounts, after.Mounts) {
return errRestartMountDrift
}
return nil
}
func RecoverLifecycleMaintenance(
ctx context.Context,
runner Runner,
updateStatePath string,
restartStatePath string,
confirm bool,
) error {
if err := validateRestartStatePaths(restartStatePath, updateStatePath); err != nil {
return err
}
if !confirm {
return errRestartConfirmation
}
lock, err := acquireLock(restartStatePath)
if err != nil {
return err
}
defer lock.Release()
restartState, restartErr := readState(restartStatePath)
if restartErr == nil {
if err := validateRestartRecoveryState(restartState); err != nil {
return err
}
restartOverride := lifecycleOverridePath(restartStatePath, restartState.Transaction)
if restartState.MutationStarted {
if err := tagImage(ctx, runner, restartState.Previous.ID, restartState.Previous.Reference, "restart recovery image pin"); err != nil {
return recoveryRequired("Pi restart recovery image pin could not be restored", err)
}
if err := writeLifecycleOverride(restartOverride, restartState.Previous.Reference); err != nil {
return recoveryRequired("Pi restart recovery image override could not be restored", err)
}
lifecycle := composeOverrideRunner{Runner: runner, path: restartOverride}
if err := ensureMaintenance(ctx, lifecycle); err != nil {
return recoveryRequired("Pi restart maintenance recovery failed", err)
}
if err := verifyRestart(ctx, lifecycle, restartState.Target.Version, restartState.Previous); err != nil {
return recoveryRequired("Pi restart recovery verification failed", err)
}
}
if err := durableRemove(restartOverride); err != nil {
return recoveryRequired("Pi restart recovery image override could not be removed", err)
}
if err := durableRemove(restartStatePath); err != nil {
return recoveryRequired("Pi restart recovery state could not be removed", err)
}
} else if !errors.Is(restartErr, os.ErrNotExist) {
return restartErr
}
return recoverMaintenanceLocked(ctx, runner, updateStatePath)
}
func restartDiagnostic(err error) error {
if errors.Is(err, ErrActiveSessions) {
return errRestartActiveSessions
}
return err
}
func validateRestartRecoveryState(state State) error {
invalid := func(reason string) error {
return fmt.Errorf("%w: restart recovery state is invalid: %s", ErrInvalidRequest, reason)
}
if state.Transaction == "" {
return invalid("transaction is missing")
}
if state.Target.Source != "restart" || state.Target.Version == "" {
return invalid("target is not a restart with a recorded version")
}
if state.Previous.ConfigurationSHA == "" {
return invalid("previous external configuration identity is missing")
}
if state.Candidate.ID != "" || state.Candidate.Reference != "" || len(state.Candidate.Mounts) != 0 ||
state.Candidate.MountFingerprint != "" || state.Candidate.ConfigurationSHA != "" {
return invalid("restart state contains image candidate metadata")
}
switch state.Phase {
case PhasePreflight:
return nil
case PhaseRecreated, PhaseVerified:
if state.MutationStarted {
return nil
}
return invalid("post-recreation phase has no mutation marker")
default:
return invalid("phase is not valid for restart")
}
}
func validateRestartStatePaths(restartStatePath, updateStatePath string) error {
if restartStatePath == "" {
return errors.New("restart state path is required")
}
if updateStatePath == "" {
return errors.New("update state path is required")
}
if filepath.Clean(restartStatePath) == filepath.Clean(updateStatePath) {
return fmt.Errorf("%w: restart and update state paths must remain separate", ErrInvalidRequest)
}
if lifecycleLockPath(restartStatePath) != lifecycleLockPath(updateStatePath) {
return fmt.Errorf("%w: restart and update state paths must share one lifecycle control directory", ErrInvalidRequest)
}
return nil
}
func pairedRestartStatePath(updateStatePath string) string {
return filepath.Join(filepath.Dir(updateStatePath), "restart-state.json")
}
func prepareLifecycleMutation(restartStatePath string, removeFile func(string) error) error {
state, err := readState(restartStatePath)
if errors.Is(err, os.ErrNotExist) {
return nil
}
if err != nil {
return fmt.Errorf("restart recovery state could not be validated: %w", err)
}
if err := validateRestartRecoveryState(state); err != nil {
return err
}
if state.Phase != PhaseVerified || !state.MutationStarted {
return ErrInterruptedRestart
}
if err := removeFile(lifecycleOverridePath(restartStatePath, state.Transaction)); err != nil {
return recoveryRequired("verified restart override could not be cleaned up", err)
}
if err := removeFile(restartStatePath); err != nil {
return recoveryRequired("verified restart state could not be cleaned up", err)
}
return nil
}