fix: harden Pi lifecycle recovery

This commit is contained in:
2026-08-04 20:28:09 +02:00
parent 5b3ce93e31
commit 5ba2821a1b
24 changed files with 1764 additions and 384 deletions
+329 -98
View File
@@ -2,15 +2,19 @@ package pi
import (
"context"
"crypto/sha256"
"encoding/json"
"errors"
"fmt"
"io"
"os"
"path/filepath"
"regexp"
"sort"
"strings"
"time"
"github.com/aritmolab/thothii/tools/thothctl/internal/compose"
"github.com/distribution/reference"
)
@@ -46,29 +50,49 @@ type Result struct {
StatePath string
}
type lifecycleHooks struct {
writeState func(string, State) error
removeFile func(string) error
sleep func(time.Duration)
}
var defaultLifecycleHooks = lifecycleHooks{
writeState: writeState,
removeFile: durableRemove,
sleep: time.Sleep,
}
// Update performs a recoverable core-only Pi update using the default Compose command layout.
func Update(ctx context.Context, runner Runner, request Request) (result Result, retErr error) {
return updateWithHooks(ctx, runner, request, defaultLifecycleHooks)
}
func updateWithHooks(ctx context.Context, runner Runner, request Request, hooks lifecycleHooks) (result Result, retErr error) {
if request.StatePath == "" {
return Result{}, errors.New("update state path is required")
}
lock, err := acquireLock(request.StatePath)
if err != nil {
return Result{StatePath: request.StatePath}, err
}
defer lock.Release()
if request.StatePath == "" {
return Result{}, errors.New("update state path is required")
}
if !request.Confirm {
return Result{StatePath: request.StatePath}, ErrConfirmationRequired
}
if !versionPattern.MatchString(request.Version) {
return Result{StatePath: request.StatePath}, fmt.Errorf("%w: Pi version must be an explicit pinned version", ErrInvalidRequest)
}
if request.Source == "" { return Result{StatePath: request.StatePath}, fmt.Errorf("%w: Pi update requires an explicit source: build or pull", ErrInvalidRequest) }
if request.Source == "" {
return Result{StatePath: request.StatePath}, fmt.Errorf("%w: Pi update requires an explicit source: build or pull", ErrInvalidRequest)
}
if request.Source != BuildSource && request.Source != PullSource {
return Result{StatePath: request.StatePath}, fmt.Errorf("%w: Pi update source must be build or pull", ErrInvalidRequest)
}
if request.Source == PullSource {
canonical, err := canonicalDigestReference(request.Image)
if err != nil { return Result{StatePath: request.StatePath}, fmt.Errorf("%w: %v", ErrInvalidRequest, err) }
if err != nil {
return Result{StatePath: request.StatePath}, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
}
request.Image = canonical
}
if old, err := readState(request.StatePath); err == nil && old.Phase != PhaseVerified && old.Phase != PhaseRolledBack && old.Phase != PhaseNoop {
@@ -76,12 +100,21 @@ func Update(ctx context.Context, runner Runner, request Request) (result Result,
} else if err != nil && !errors.Is(err, os.ErrNotExist) {
return Result{StatePath: request.StatePath}, err
}
if err := setMaintenance(ctx, runner, true); err != nil { return Result{StatePath: request.StatePath}, err }
if err := setMaintenance(ctx, runner, true); err != nil {
return Result{StatePath: request.StatePath}, err
}
clearMaintenance := true
defer func() {
if !clearMaintenance {
return
}
if clearErr := setMaintenance(context.Background(), runner, false); clearErr != nil {
result = Result{Phase: PhaseFailed, StatePath: request.StatePath}
if retErr == nil { retErr = errors.New("maintenance admission gate could not be cleared: recovery required")
} else { retErr = fmt.Errorf("%w; maintenance admission gate could not be cleared: recovery required", retErr) }
if retErr == nil {
retErr = errors.New("maintenance admission gate could not be cleared: recovery required")
} else {
retErr = fmt.Errorf("%w; maintenance admission gate could not be cleared: recovery required", retErr)
}
}
}()
@@ -101,7 +134,7 @@ func Update(ctx context.Context, runner Runner, request Request) (result Result,
if !running {
break
}
time.Sleep(time.Second)
hooks.sleep(time.Second)
}
if running {
return Result{StatePath: request.StatePath}, ErrActiveSessions
@@ -127,46 +160,87 @@ func Update(ctx context.Context, runner Runner, request Request) (result Result,
return Result{StatePath: request.StatePath}, err
}
previous.ConfigurationSHA = configured.ConfigurationSHA
state := State{Phase: PhasePreflight, Target: Target{Version: request.Version, Source: sourceValue(request)}, Previous: previous}
if err := writeState(request.StatePath, state); err != nil {
transaction := lifecycleTransaction(request.StatePath)
previous.Reference = lifecycleImageTag(transaction, "previous")
candidateReference := lifecycleImageTag(transaction, "candidate")
if err := tagImage(ctx, runner, previous.ID, previous.Reference, "previous Pi image pin"); err != nil {
return Result{StatePath: request.StatePath}, err
}
state := State{
Transaction: transaction,
Phase: PhasePreflight,
Target: Target{Version: request.Version, Source: sourceValue(request)},
Previous: previous,
Candidate: Image{Reference: candidateReference},
}
if err := hooks.writeState(request.StatePath, state); err != nil {
return Result{StatePath: request.StatePath}, err
}
overridePath := lifecycleOverridePath(request.StatePath, transaction)
if err := writeLifecycleOverride(overridePath, candidateReference); err != nil {
return Result{StatePath: request.StatePath}, err
}
lifecycle := composeOverrideRunner{Runner: runner, path: overridePath}
state.Phase = PhaseBuilding
if err := writeState(request.StatePath, state); err != nil { return Result{Phase: PhaseFailed, StatePath: request.StatePath}, err }
if err := prepareCandidate(ctx, runner, request, previous.Reference); err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
clearMaintenance = false
if err := hooks.writeState(request.StatePath, state); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if request.Drain {
running, err = activeSessions(ctx, runner)
if err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
}
if running {
return compensate(ctx, runner, request.StatePath, state, ErrActiveSessions)
}
if err := prepareCandidate(ctx, lifecycle, request, candidateReference); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if err := recreateCore(ctx, runner); err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
running, err = activeSessions(ctx, runner)
if err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if running {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, ErrActiveSessions, hooks)
return result, retErr
}
if err := recreateCore(ctx, lifecycle); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if err := ensureMaintenance(ctx, lifecycle); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
state.Phase = PhaseRecreated
state.Candidate, err = runningImage(ctx, runner, previous.Reference)
if err != nil { return compensate(ctx, runner, request.StatePath, state, err) }
if err := writeState(request.StatePath, state); err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
state.Candidate, err = runningImage(ctx, lifecycle, candidateReference)
if err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if err := verifyCandidate(ctx, runner, request.Version, previous); err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
if err := hooks.writeState(request.StatePath, state); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if err := verifyCandidate(ctx, lifecycle, request.Version, previous); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
state.Phase, state.Error = PhaseVerified, ""
if err := writeState(request.StatePath, state); err != nil {
return compensate(ctx, runner, request.StatePath, state, err)
if err := hooks.writeState(request.StatePath, state); err != nil {
result, retErr, clearMaintenance = compensate(ctx, runner, request.StatePath, overridePath, state, err, hooks)
return result, retErr
}
if err := hooks.removeFile(overridePath); err != nil {
return Result{Phase: PhaseFailed, StatePath: request.StatePath}, errors.New("verified update override cleanup failed: maintenance recovery required")
}
clearMaintenance = true
return Result{Phase: PhaseVerified, StatePath: request.StatePath}, nil
}
// Rollback restores the image recorded in durable update state. It is safe for interrupted runs.
func Rollback(ctx context.Context, runner Runner, statePath string, confirm bool) (result Result, retErr error) {
return rollbackWithHooks(ctx, runner, statePath, confirm, defaultLifecycleHooks)
}
func rollbackWithHooks(ctx context.Context, runner Runner, statePath string, confirm bool, hooks lifecycleHooks) (result Result, retErr error) {
lock, err := acquireLock(statePath)
if err != nil {
return Result{StatePath: statePath}, err
@@ -175,51 +249,88 @@ func Rollback(ctx context.Context, runner Runner, statePath string, confirm bool
if !confirm {
return Result{StatePath: statePath}, ErrConfirmationRequired
}
if err := setMaintenance(ctx, runner, true); err != nil { return Result{StatePath: statePath}, err }
defer func() {
if clearErr := setMaintenance(context.Background(), runner, false); clearErr != nil {
result = Result{Phase: PhaseFailed, StatePath: statePath}
if retErr == nil { retErr = errors.New("maintenance admission gate could not be cleared: recovery required")
} else { retErr = fmt.Errorf("%w; maintenance admission gate could not be cleared: recovery required", retErr) }
}
}()
if active, err := activeSessions(ctx, runner); err != nil { return Result{StatePath: statePath}, err
} else if active { return Result{StatePath: statePath}, ErrActiveSessions }
state, err := readState(statePath)
if err != nil {
if err := setMaintenance(ctx, runner, true); err != nil {
return Result{StatePath: statePath}, err
}
if err := restore(ctx, runner, state.Previous); err != nil {
clearMaintenance := true
defer func() {
if !clearMaintenance {
return
}
if clearErr := setMaintenance(context.Background(), runner, false); clearErr != nil {
result = Result{Phase: PhaseFailed, StatePath: statePath}
if retErr == nil {
retErr = errors.New("maintenance admission gate could not be cleared: recovery required")
} else {
retErr = fmt.Errorf("%w; maintenance admission gate could not be cleared: recovery required", retErr)
}
}
}()
if active, err := activeSessions(ctx, runner); err != nil {
return Result{StatePath: statePath}, err
} else if active {
return Result{StatePath: statePath}, ErrActiveSessions
}
state, err := readState(statePath)
if err != nil {
clearMaintenance = false
return Result{StatePath: statePath}, err
}
overridePath := lifecycleOverridePath(statePath, state.Transaction)
if err := writeLifecycleOverride(overridePath, state.Previous.Reference); err != nil {
clearMaintenance = false
return Result{StatePath: statePath}, err
}
clearMaintenance = false
lifecycle := composeOverrideRunner{Runner: runner, path: overridePath}
if err := restore(ctx, lifecycle, state.Previous); err != nil {
state.Phase, state.Error = PhaseFailed, "rollback failed"
if writeErr := writeState(statePath, state); writeErr != nil { return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("rollback failed and recovery state could not be persisted") }
if writeErr := hooks.writeState(statePath, state); writeErr != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("rollback failed and recovery state could not be persisted")
}
return Result{Phase: PhaseFailed, StatePath: statePath}, err
}
state.Phase, state.Error = PhaseRolledBack, ""
if err := writeState(statePath, state); err != nil {
if err := hooks.writeState(statePath, state); err != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("rollback restored the core but recovery state could not be persisted")
}
if err := hooks.removeFile(overridePath); err != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("rollback override cleanup failed: maintenance recovery required")
}
clearMaintenance = true
return Result{Phase: PhaseRolledBack, StatePath: statePath}, nil
}
func compensate(ctx context.Context, runner Runner, statePath string, state State, cause error) (Result, error) {
if restoreErr := restore(ctx, runner, state.Previous); restoreErr != nil {
func compensate(ctx context.Context, runner Runner, statePath, overridePath string, state State, cause error, hooks lifecycleHooks) (Result, error, bool) {
if err := ensureMaintenance(context.Background(), runner); err != nil {
state.Phase, state.Error = PhaseFailed, "maintenance recovery failed"
_ = hooks.writeState(statePath, state)
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("update failed and maintenance could not be reactivated: recovery required"), false
}
if active, err := activeSessions(context.Background(), runner); err != nil || active {
state.Phase, state.Error = PhaseFailed, "rollback inventory failed"
_ = hooks.writeState(statePath, state)
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("update failed and rollback inventory is not quiescent: recovery required"), false
}
if err := writeLifecycleOverride(overridePath, state.Previous.Reference); err != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("update failed and rollback override could not be prepared: recovery required"), false
}
lifecycle := composeOverrideRunner{Runner: runner, path: overridePath}
if restoreErr := restore(ctx, lifecycle, state.Previous); restoreErr != nil {
state.Phase, state.Error = PhaseFailed, "candidate verification and automatic rollback failed"
if writeErr := writeState(statePath, state); writeErr != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("update failed and rollback proof failed; recovery state could not be persisted")
if writeErr := hooks.writeState(statePath, state); writeErr != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("update failed and rollback proof failed; recovery state could not be persisted"), false
}
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("update failed; automatic rollback also failed: recovery required")
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("update failed; automatic rollback also failed: recovery required"), false
}
state.Phase, state.Error = PhaseRolledBack, ""
if writeErr := writeState(statePath, state); writeErr != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("previous core image was restored but recovery state write failed: recovery required")
if writeErr := hooks.writeState(statePath, state); writeErr != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, fmt.Errorf("previous core image was restored but recovery state write failed: recovery required"), false
}
return Result{Phase: PhaseRolledBack, StatePath: statePath}, fmt.Errorf("update failed; previous core image was restored")
}
func recordFailure(path string, state State, label string, cause error) error {
state.Phase, state.Error = PhaseFailed, label
if err := writeState(path, state); err != nil { return fmt.Errorf("%w; recovery state write failed", cause) }
return cause
if err := hooks.removeFile(overridePath); err != nil {
return Result{Phase: PhaseFailed, StatePath: statePath}, errors.New("previous core image was restored but override cleanup failed: recovery required"), false
}
return Result{Phase: PhaseRolledBack, StatePath: statePath}, fmt.Errorf("update failed; previous core image was restored"), true
}
func sourceValue(request Request) string {
@@ -234,7 +345,9 @@ func canonicalDigestReference(value string) (string, error) {
return "", errors.New("pulled Pi image must be a credential-free canonical sha256 digest reference")
}
parsed, err := reference.ParseAnyReference(value)
if err != nil { return "", errors.New("pulled Pi image must be a valid canonical sha256 digest reference") }
if err != nil {
return "", errors.New("pulled Pi image must be a valid canonical sha256 digest reference")
}
canonical, ok := parsed.(reference.Canonical)
if !ok || canonical.Digest().Algorithm().String() != "sha256" || len(canonical.Digest().Encoded()) != 64 {
return "", errors.New("pulled Pi image must use an immutable sha256 digest")
@@ -244,14 +357,56 @@ func canonicalDigestReference(value string) (string, error) {
func setMaintenance(ctx context.Context, runner Runner, enabled bool) error {
path := "deactivate"
if enabled { path = "activate" }
args := append([]string{"exec", "-T", "core", "curl", "-fsS", "-X", "POST"}, internalIdentityHeaders...)
args = append(args, "http://127.0.0.1:8787/internal/maintenance/"+path)
if enabled {
path = "activate"
}
args := []string{"exec", "-T", "core", "curl", "-fsS", "-X", "POST", "http://127.0.0.1:8787/internal/maintenance/" + path}
result, err := runCompose(ctx, runner, args...)
if err != nil { return commandError("maintenance admission gate", result, err) }
var status struct { Active bool `json:"active"`; Admissions int `json:"admissions"` }
if json.Unmarshal([]byte(result.Stdout), &status) != nil || status.Active != enabled || status.Admissions != 0 { return errors.New("maintenance admission gate did not acknowledge a quiescent state") }
return nil
status, valid := parseMaintenanceStatus(result.Stdout)
if err == nil && valid && status.Active == enabled && status.Admissions == 0 {
return nil
}
// A core recreate or transport interruption may lose only the response. Resolve ambiguity by
// reading the durable gate state before deciding that operator recovery is required.
observed, statusErr := MaintenanceStatus(ctx, runner)
if statusErr == nil && observed.Active == enabled && observed.Admissions == 0 {
return nil
}
if err != nil {
return commandError("maintenance admission gate", result, err)
}
return errors.New("maintenance admission gate did not acknowledge a quiescent state")
}
type MaintenanceState struct {
Active bool `json:"active"`
Admissions int `json:"admissions"`
}
func parseMaintenanceStatus(value string) (MaintenanceState, bool) {
var status MaintenanceState
err := json.Unmarshal([]byte(value), &status)
return status, err == nil && status.Admissions >= 0
}
func MaintenanceStatus(ctx context.Context, runner Runner) (MaintenanceState, error) {
result, err := runCompose(ctx, runner, "exec", "-T", "core", "curl", "-fsS", "http://127.0.0.1:8787/internal/maintenance/status")
if err != nil {
return MaintenanceState{}, commandError("maintenance status check", result, err)
}
status, valid := parseMaintenanceStatus(result.Stdout)
if !valid {
return MaintenanceState{}, errors.New("maintenance status check returned invalid data")
}
return status, nil
}
func ensureMaintenance(ctx context.Context, runner Runner) error {
status, err := MaintenanceStatus(ctx, runner)
if err == nil && status.Active && status.Admissions == 0 {
return nil
}
return setMaintenance(ctx, runner, true)
}
func activeSessions(ctx context.Context, runner Runner) (bool, error) {
@@ -261,14 +416,14 @@ func activeSessions(ctx context.Context, runner Runner) (bool, error) {
if err != nil {
return false, commandError("active-session check", result, err)
}
var payload struct { Sessions []struct {
var payload []struct {
Status string `json:"status"`
Archived bool `json:"archived"`
} `json:"sessions"` }
}
if err := json.Unmarshal([]byte(result.Stdout), &payload); err != nil {
return false, errors.New("active-session check returned invalid session data")
}
for _, session := range payload.Sessions {
for _, session := range payload {
if !session.Archived && session.Status != "finalized" && session.Status != "closed" {
return true, nil
}
@@ -291,9 +446,14 @@ func runningImage(ctx context.Context, runner Runner, reference string) (Image,
return Image{}, commandError("core volume check", mounts, err)
}
var raw []struct {
Type string `json:"Type"`; Name string `json:"Name"`; Source string `json:"Source"`
Destination string `json:"Destination"`; RW bool `json:"RW"`; Mode string `json:"Mode"`
Propagation string `json:"Propagation"`; Driver string `json:"Driver"`
Type string `json:"Type"`
Name string `json:"Name"`
Source string `json:"Source"`
Destination string `json:"Destination"`
RW bool `json:"RW"`
Mode string `json:"Mode"`
Propagation string `json:"Propagation"`
Driver string `json:"Driver"`
}
if err := json.Unmarshal([]byte(mounts.Stdout), &raw); err != nil {
return Image{}, errors.New("core returned invalid persistence mount data")
@@ -302,18 +462,16 @@ func runningImage(ctx context.Context, runner Runner, reference string) (Image,
return Image{}, errors.New("core has no persistence mounts to preserve")
}
contract := make([]Mount, 0, len(raw))
volumes := make([]string, 0, len(raw))
for _, mount := range raw {
if mount.Type == "" || mount.Source == "" || mount.Destination == "" { return Image{}, errors.New("core returned incomplete persistence mount data") }
contract = append(contract, Mount{Type: mount.Type, Name: mount.Name, SourceSHA256: mountSourceHash(mount.Source), Destination: mount.Destination, RW: mount.RW, Options: strings.Join([]string{mount.Mode, mount.Propagation, mount.Driver}, "\x00")})
if mount.Type == "volume" && mount.Name != "" {
volumes = append(volumes, mount.Name)
if mount.Type == "" || mount.Source == "" || mount.Destination == "" {
return Image{}, errors.New("core returned incomplete persistence mount data")
}
contract = append(contract, Mount{Type: mount.Type, Name: mount.Name, SourceSHA256: mountSourceHash(mount.Source), Destination: mount.Destination, RW: mount.RW, Options: strings.Join([]string{mount.Mode, mount.Propagation, mount.Driver}, "\x00")})
}
return Image{ID: strings.TrimSpace(image.Stdout), Reference: reference, Volumes: volumes, Mounts: contract, MountFingerprint: mountFingerprint(contract)}, nil
return Image{ID: strings.TrimSpace(image.Stdout), Reference: reference, Mounts: contract, MountFingerprint: mountFingerprint(contract)}, nil
}
func prepareCandidate(ctx context.Context, runner Runner, request Request, reference string) error {
func prepareCandidate(ctx context.Context, runner Runner, request Request, candidateReference string) error {
if request.Source == BuildSource {
result, err := runCompose(ctx, runner, "build", "--pull", "--build-arg", "PI_VERSION="+request.Version, "core")
if err != nil {
@@ -325,11 +483,7 @@ func prepareCandidate(ctx context.Context, runner Runner, request Request, refer
if err != nil {
return commandError("Pi image pull", pull, err)
}
tag, err := runner.Run(ctx, []string{"image", "tag", request.Image, reference}, nil)
if err != nil {
return commandError("Pi image tag", tag, err)
}
return nil
return tagImage(ctx, runner, request.Image, candidateReference, "Pi image tag")
}
func recreateCore(ctx context.Context, runner Runner) error {
@@ -373,13 +527,15 @@ func verifyCandidate(ctx context.Context, runner Runner, wanted string, previous
}
func restore(ctx context.Context, runner Runner, previous Image) error {
tag, err := runner.Run(ctx, []string{"image", "tag", previous.ID, previous.Reference}, nil)
if err != nil {
return commandError("rollback image restore", tag, err)
if err := tagImage(ctx, runner, previous.ID, previous.Reference, "rollback image restore"); err != nil {
return err
}
if err := recreateCore(ctx, runner); err != nil {
return err
}
if err := ensureMaintenance(ctx, runner); err != nil {
return err
}
configured, err := renderedCore(ctx, runner)
if err != nil {
return err
@@ -406,16 +562,91 @@ func restore(ctx context.Context, runner Runner, previous Image) error {
return nil
}
func nonEmptyLines(text string) []string {
var values []string
for _, value := range strings.Split(text, "\n") {
if value = strings.TrimSpace(value); value != "" {
values = append(values, value)
}
func tagImage(ctx context.Context, runner Runner, source, target, label string) error {
result, err := runner.Run(ctx, []string{"image", "tag", source, target}, nil)
if err != nil {
return commandError(label, result, err)
}
sort.Strings(values)
return values
return nil
}
type composeOverrideRunner struct {
Runner
path string
}
func (r composeOverrideRunner) Run(ctx context.Context, args []string, stdin io.Reader) (compose.Result, error) {
if len(args) > 0 && args[0] == "compose" {
withOverride := append([]string{"compose", "-f", r.path}, args[1:]...)
return r.Runner.Run(ctx, withOverride, stdin)
}
return r.Runner.Run(ctx, args, stdin)
}
func lifecycleTransaction(statePath string) string {
value := fmt.Sprintf("%s\x00%d\x00%d", filepath.Clean(statePath), os.Getpid(), time.Now().UnixNano())
sum := sha256.Sum256([]byte(value))
return fmt.Sprintf("%x", sum[:8])
}
func lifecycleImageTag(transaction, role string) string {
return "thothii-core:thothctl-" + transaction + "-" + role
}
func lifecycleOverridePath(statePath, transaction string) string {
if transaction == "" {
transaction = "recovery"
}
return filepath.Join(filepath.Dir(statePath), "pi-lifecycle-"+transaction+".yaml")
}
func writeLifecycleOverride(path, image string) error {
quoted, err := json.Marshal(image)
if err != nil {
return errors.New("lifecycle image override could not be encoded")
}
contents := []byte("services:\n core:\n image: " + string(quoted) + "\n")
if err := writeFileDurably(path, ".pi-lifecycle-", contents); err != nil {
return errors.New("lifecycle image override could not be written durably")
}
return nil
}
// RecoverMaintenance clears a stale durable gate only after the running core and terminal
// recovery metadata prove that no rollback is still required.
func RecoverMaintenance(ctx context.Context, runner Runner, statePath string, confirm bool) error {
if !confirm {
return ErrConfirmationRequired
}
lock, err := acquireLock(statePath)
if err != nil {
return err
}
defer lock.Release()
state, stateErr := readState(statePath)
if stateErr == nil {
if state.Phase != PhaseVerified && state.Phase != PhaseRolledBack && state.Phase != PhaseNoop {
return ErrInterruptedUpdate
}
if err := durableRemove(lifecycleOverridePath(statePath, state.Transaction)); err != nil {
return errors.New("maintenance recovery could not remove the lifecycle override")
}
} else if !errors.Is(stateErr, os.ErrNotExist) {
return stateErr
}
status, err := MaintenanceStatus(ctx, runner)
if err != nil {
return err
}
if !status.Active {
return nil
}
if err := Doctor(ctx, runner); err != nil {
return err
}
return setMaintenance(ctx, runner, false)
}
func sameStrings(left, right []string) bool {
left, right = append([]string(nil), left...), append([]string(nil), right...)
sort.Strings(left)