// Package restartruntime implements the `restart` lifecycle operation
// owned by Runtime Manager. Restart is a recreate: under one outer
// per-game lease the service runs the stop service, removes the
// container with `docker rm`, and runs the start service with the
// runtime's current `image_ref`. The hostname / engine endpoint stays
// stable across the recreate; `container_id` changes.
//
// Lifecycle and failure-mode semantics follow `rtmanager/README.md
// §Lifecycles → Restart`. Design rationale is captured in
// `rtmanager/docs/services.md`, in particular the lease-sharing
// pattern with `startruntime.Service.Run` / `stopruntime.Service.Run`,
// the correlation-id reuse on `source_ref`, and the
// inner-stop-then-rm-failure recovery rule.
package restartruntime

import (
	"context"
	"crypto/rand"
	"encoding/base64"
	"errors"
	"fmt"
	"log/slog"
	"strings"
	"time"

	"galaxy/rtmanager/internal/config"
	"galaxy/rtmanager/internal/domain/operation"
	"galaxy/rtmanager/internal/domain/runtime"
	"galaxy/rtmanager/internal/logging"
	"galaxy/rtmanager/internal/ports"
	"galaxy/rtmanager/internal/service/startruntime"
	"galaxy/rtmanager/internal/service/stopruntime"
	"galaxy/rtmanager/internal/telemetry"
)

// leaseReleaseTimeout bounds the deferred lease-release call.
const leaseReleaseTimeout = 5 * time.Second

// Input stores the per-call arguments for one restart operation.
type Input struct {
	// GameID identifies the platform game to restart.
	GameID string

	// OpSource classifies how the request entered Runtime Manager.
	// Required: every operation_log entry carries an op_source.
	OpSource operation.OpSource

	// SourceRef stores the optional opaque per-source reference (REST
	// request id, admin user id). When non-empty it is reused as the
	// correlation id linking the outer restart entry to the inner stop
	// and start log entries.
	SourceRef string
}

// Validate reports whether input carries the structural invariants the
// service requires.
func (input Input) Validate() error {
	if strings.TrimSpace(input.GameID) == "" {
		return fmt.Errorf("game id must not be empty")
	}
	if !input.OpSource.IsKnown() {
		return fmt.Errorf("op source %q is unsupported", input.OpSource)
	}
	return nil
}

// Result stores the deterministic outcome of one Handle call.
type Result struct {
	// Record carries the runtime record installed by the inner start on
	// success; zero on failure.
	Record runtime.RuntimeRecord

	// Outcome reports whether the operation completed (success) or
	// produced a stable failure code.
	Outcome operation.Outcome

	// ErrorCode stores the stable error code on failure. Empty for
	// success.
	ErrorCode string

	// ErrorMessage stores the operator-readable detail on failure.
	// Empty for success.
	ErrorMessage string
}

// Dependencies groups the collaborators required by Service.
type Dependencies struct {
	// RuntimeRecords reads the runtime record at the start of restart
	// to capture the current image_ref and container_id.
	RuntimeRecords ports.RuntimeRecordStore

	// OperationLogs records the outer restart audit entry. Inner stop
	// and start services append their own entries through their own
	// stores.
	OperationLogs ports.OperationLogStore

	// Docker drives the docker rm step between the inner stop and
	// inner start.
	Docker ports.DockerClient

	// Leases serialises operations against the same game id. The outer
	// lease is held for the entire stop + rm + start sequence.
	Leases ports.GameLeaseStore

	// StopService runs the inner stop step under the outer lease.
	StopService *stopruntime.Service

	// StartService runs the inner start step under the outer lease.
	StartService *startruntime.Service

	// Coordination supplies the per-game lease TTL.
	Coordination config.CoordinationConfig

	// Telemetry records restart outcomes and lease latency. Required.
	Telemetry *telemetry.Runtime

	// Logger records structured service-level events. Defaults to
	// `slog.Default()` when nil.
	Logger *slog.Logger

	// Clock supplies the wall-clock used for operation timestamps.
	// Defaults to `time.Now` when nil.
	Clock func() time.Time

	// NewToken supplies a unique opaque token. Used both for the lease
	// and for the correlation id when Input.SourceRef is empty.
	// Defaults to a 32-byte random base64url string when nil.
	NewToken func() string
}

// Service executes the restart lifecycle operation.
type Service struct {
	runtimeRecords ports.RuntimeRecordStore
	operationLogs  ports.OperationLogStore
	docker         ports.DockerClient
	leases         ports.GameLeaseStore
	stopService    *stopruntime.Service
	startService   *startruntime.Service

	leaseTTL time.Duration

	telemetry *telemetry.Runtime
	logger    *slog.Logger

	clock    func() time.Time
	newToken func() string
}

// NewService constructs one Service from deps.
func NewService(deps Dependencies) (*Service, error) {
	switch {
	case deps.RuntimeRecords == nil:
		return nil, errors.New("new restart runtime service: nil runtime records")
	case deps.OperationLogs == nil:
		return nil, errors.New("new restart runtime service: nil operation logs")
	case deps.Docker == nil:
		return nil, errors.New("new restart runtime service: nil docker client")
	case deps.Leases == nil:
		return nil, errors.New("new restart runtime service: nil lease store")
	case deps.StopService == nil:
		return nil, errors.New("new restart runtime service: nil stop service")
	case deps.StartService == nil:
		return nil, errors.New("new restart runtime service: nil start service")
	case deps.Telemetry == nil:
		return nil, errors.New("new restart runtime service: nil telemetry runtime")
	}
	if err := deps.Coordination.Validate(); err != nil {
		return nil, fmt.Errorf("new restart runtime service: coordination config: %w", err)
	}

	clock := deps.Clock
	if clock == nil {
		clock = time.Now
	}
	logger := deps.Logger
	if logger == nil {
		logger = slog.Default()
	}
	logger = logger.With("service", "rtmanager.restartruntime")

	newToken := deps.NewToken
	if newToken == nil {
		newToken = defaultTokenGenerator()
	}

	return &Service{
		runtimeRecords: deps.RuntimeRecords,
		operationLogs:  deps.OperationLogs,
		docker:         deps.Docker,
		leases:         deps.Leases,
		stopService:    deps.StopService,
		startService:   deps.StartService,
		leaseTTL:       deps.Coordination.GameLeaseTTL,
		telemetry:      deps.Telemetry,
		logger:         logger,
		clock:          clock,
		newToken:       newToken,
	}, nil
}

// Handle executes one restart operation end-to-end. The Go-level error
// return is reserved for non-business failures (nil context, nil
// receiver). Every business outcome — success or any of the stable
// failure codes — flows through Result.
func (service *Service) Handle(ctx context.Context, input Input) (Result, error) {
	if service == nil {
		return Result{}, errors.New("restart runtime: nil service")
	}
	if ctx == nil {
		return Result{}, errors.New("restart runtime: nil context")
	}

	opStartedAt := service.clock().UTC()

	if err := input.Validate(); err != nil {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeInvalidRequest,
			errorMessage: err.Error(),
		}), nil
	}

	token := service.newToken()
	leaseStart := service.clock()
	acquired, err := service.leases.TryAcquire(ctx, input.GameID, token, service.leaseTTL)
	service.telemetry.RecordLeaseAcquireLatency(ctx, service.clock().Sub(leaseStart))
	if err != nil {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeServiceUnavailable,
			errorMessage: fmt.Sprintf("acquire game lease: %s", err.Error()),
		}), nil
	}
	if !acquired {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeConflict,
			errorMessage: "another lifecycle operation is in progress for this game",
		}), nil
	}
	defer service.releaseLease(ctx, input.GameID, token)

	return service.runUnderLease(ctx, input, opStartedAt)
}

// runUnderLease executes the lease-protected restart sequence. Loads
// the runtime record, runs inner stop, removes the container, runs
// inner start.
func (service *Service) runUnderLease(ctx context.Context, input Input, opStartedAt time.Time) (Result, error) {
	existing, err := service.runtimeRecords.Get(ctx, input.GameID)
	if errors.Is(err, runtime.ErrNotFound) {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeNotFound,
			errorMessage: fmt.Sprintf("runtime record for game %q does not exist", input.GameID),
		}), nil
	}
	if err != nil {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeInternal,
			errorMessage: fmt.Sprintf("load runtime record: %s", err.Error()),
		}), nil
	}
	if existing.Status == runtime.StatusRemoved {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeConflict,
			errorMessage: fmt.Sprintf("runtime for game %q is removed; cannot restart", input.GameID),
			imageRef:     existing.CurrentImageRef,
		}), nil
	}
	if strings.TrimSpace(existing.CurrentImageRef) == "" {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeInternal,
			errorMessage: fmt.Sprintf("runtime record for game %q has no image_ref to restart with", input.GameID),
		}), nil
	}

	correlationRef := input.SourceRef
	if correlationRef == "" {
		correlationRef = service.newToken()
	}
	containerID := existing.CurrentContainerID
	imageRef := existing.CurrentImageRef

	stopResult, err := service.stopService.Run(ctx, stopruntime.Input{
		GameID:    input.GameID,
		Reason:    stopruntime.StopReasonAdminRequest,
		OpSource:  input.OpSource,
		SourceRef: correlationRef,
	})
	if err != nil {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeInternal,
			errorMessage: fmt.Sprintf("inner stop: %s", err.Error()),
			imageRef:     imageRef,
			containerID:  containerID,
		}), nil
	}
	if stopResult.Outcome == operation.OutcomeFailure {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    stopResult.ErrorCode,
			errorMessage: fmt.Sprintf("inner stop failed: %s", stopResult.ErrorMessage),
			imageRef:     imageRef,
			containerID:  containerID,
		}), nil
	}

	if containerID != "" {
		if err := service.docker.Remove(ctx, containerID); err != nil {
			return service.recordFailure(ctx, failureCtx{
				opStartedAt:  opStartedAt,
				input:        input,
				errorCode:    startruntime.ErrorCodeServiceUnavailable,
				errorMessage: fmt.Sprintf("docker remove: %s", err.Error()),
				imageRef:     imageRef,
				containerID:  containerID,
			}), nil
		}
	}

	startResult, err := service.startService.Run(ctx, startruntime.Input{
		GameID:    input.GameID,
		ImageRef:  imageRef,
		OpSource:  input.OpSource,
		SourceRef: correlationRef,
	})
	if err != nil {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startruntime.ErrorCodeInternal,
			errorMessage: fmt.Sprintf("inner start: %s", err.Error()),
			imageRef:     imageRef,
		}), nil
	}
	if startResult.Outcome == operation.OutcomeFailure {
		return service.recordFailure(ctx, failureCtx{
			opStartedAt:  opStartedAt,
			input:        input,
			errorCode:    startResult.ErrorCode,
			errorMessage: fmt.Sprintf("inner start failed: %s", startResult.ErrorMessage),
			imageRef:     imageRef,
		}), nil
	}

	finishedAt := service.clock().UTC()
	service.bestEffortAppend(ctx, operation.OperationEntry{
		GameID:      input.GameID,
		OpKind:      operation.OpKindRestart,
		OpSource:    input.OpSource,
		SourceRef:   correlationRef,
		ImageRef:    imageRef,
		ContainerID: startResult.Record.CurrentContainerID,
		Outcome:     operation.OutcomeSuccess,
		StartedAt:   opStartedAt,
		FinishedAt:  &finishedAt,
	})
	service.telemetry.RecordRestartOutcome(ctx, string(operation.OutcomeSuccess), "")

	logArgs := []any{
		"game_id", input.GameID,
		"prev_container_id", containerID,
		"new_container_id", startResult.Record.CurrentContainerID,
		"image_ref", imageRef,
		"op_source", string(input.OpSource),
	}
	logArgs = append(logArgs, logging.ContextAttrs(ctx)...)
	service.logger.InfoContext(ctx, "runtime restarted", logArgs...)

	return Result{
		Record:  startResult.Record,
		Outcome: operation.OutcomeSuccess,
	}, nil
}

// failureCtx groups the inputs to recordFailure.
type failureCtx struct {
	opStartedAt  time.Time
	input        Input
	errorCode    string
	errorMessage string
	imageRef     string
	containerID  string
}

// recordFailure records the outer failure operation_log entry and emits
// telemetry. Inner stop / start services have already recorded their
// own entries; this is the outer summary.
func (service *Service) recordFailure(ctx context.Context, fc failureCtx) Result {
	finishedAt := service.clock().UTC()
	service.bestEffortAppend(ctx, operation.OperationEntry{
		GameID:       fc.input.GameID,
		OpKind:       operation.OpKindRestart,
		OpSource:     fc.input.OpSource,
		SourceRef:    correlationRefOrEmpty(fc.input),
		ImageRef:     fc.imageRef,
		ContainerID:  fc.containerID,
		Outcome:      operation.OutcomeFailure,
		ErrorCode:    fc.errorCode,
		ErrorMessage: fc.errorMessage,
		StartedAt:    fc.opStartedAt,
		FinishedAt:   &finishedAt,
	})
	service.telemetry.RecordRestartOutcome(ctx, string(operation.OutcomeFailure), fc.errorCode)

	logArgs := []any{
		"game_id", fc.input.GameID,
		"image_ref", fc.imageRef,
		"op_source", string(fc.input.OpSource),
		"error_code", fc.errorCode,
		"error_message", fc.errorMessage,
	}
	logArgs = append(logArgs, logging.ContextAttrs(ctx)...)
	service.logger.WarnContext(ctx, "runtime restart failed", logArgs...)

	return Result{
		Outcome:      operation.OutcomeFailure,
		ErrorCode:    fc.errorCode,
		ErrorMessage: fc.errorMessage,
	}
}

// correlationRefOrEmpty returns the original Input.SourceRef for the
// outer entry. Outer-failure paths that did not yet generate a
// correlation id (input validation, lease busy) keep the original
// `source_ref` which is the actor ref.
func correlationRefOrEmpty(input Input) string {
	return input.SourceRef
}

// releaseLease releases the per-game lease in a fresh background context.
func (service *Service) releaseLease(ctx context.Context, gameID, token string) {
	cleanupCtx, cancel := context.WithTimeout(context.Background(), leaseReleaseTimeout)
	defer cancel()
	if err := service.leases.Release(cleanupCtx, gameID, token); err != nil {
		service.logger.WarnContext(ctx, "release game lease",
			"game_id", gameID,
			"err", err.Error(),
		)
	}
}

// bestEffortAppend writes one outer operation_log entry. Inner ops have
// already appended their own; a failure here only loses the outer
// summary, which is acceptable.
func (service *Service) bestEffortAppend(ctx context.Context, entry operation.OperationEntry) {
	if _, err := service.operationLogs.Append(ctx, entry); err != nil {
		service.logger.ErrorContext(ctx, "append operation log",
			"game_id", entry.GameID,
			"op_kind", string(entry.OpKind),
			"outcome", string(entry.Outcome),
			"error_code", entry.ErrorCode,
			"err", err.Error(),
		)
	}
}

// defaultTokenGenerator returns a function that produces 32-byte
// base64url-encoded tokens.
func defaultTokenGenerator() func() string {
	return func() string {
		var buf [32]byte
		if _, err := rand.Read(buf[:]); err != nil {
			return "rtmanager-fallback-token"
		}
		return base64.RawURLEncoding.EncodeToString(buf[:])
	}
}