galaxy-game/mail/internal/adapters/redisstate/attempt_execution_store.go

package redisstate

import (
	"context"
	"errors"
	"fmt"
	"time"

	"galaxy/mail/internal/domain/attempt"
	"galaxy/mail/internal/domain/common"
	deliverydomain "galaxy/mail/internal/domain/delivery"
	"galaxy/mail/internal/service/acceptgenericdelivery"
	"galaxy/mail/internal/service/executeattempt"
	"galaxy/mail/internal/telemetry"

	"github.com/redis/go-redis/v9"
)

var errNotClaimable = errors.New("attempt is not claimable")

// AttemptExecutionStore provides the Redis-backed durable storage used by the
// attempt scheduler and attempt execution service.
type AttemptExecutionStore struct {
	client *redis.Client
	keys   Keyspace
}

// NewAttemptExecutionStore constructs one Redis-backed attempt execution
// store.
func NewAttemptExecutionStore(client *redis.Client) (*AttemptExecutionStore, error) {
	if client == nil {
		return nil, errors.New("new attempt execution store: nil redis client")
	}

	return &AttemptExecutionStore{
		client: client,
		keys:   Keyspace{},
	}, nil
}

// NextDueDeliveryIDs returns up to limit due delivery identifiers ordered by
// the attempt schedule score.
func (store *AttemptExecutionStore) NextDueDeliveryIDs(ctx context.Context, now time.Time, limit int64) ([]common.DeliveryID, error) {
	if store == nil || store.client == nil {
		return nil, errors.New("next due delivery ids: nil store")
	}
	if ctx == nil {
		return nil, errors.New("next due delivery ids: nil context")
	}
	if limit <= 0 {
		return nil, errors.New("next due delivery ids: non-positive limit")
	}

	values, err := store.client.ZRangeByScore(ctx, store.keys.AttemptSchedule(), &redis.ZRangeBy{
		Min:   "-inf",
		Max:   fmt.Sprintf("%d", now.UTC().UnixMilli()),
		Count: limit,
	}).Result()
	if err != nil {
		return nil, fmt.Errorf("next due delivery ids: %w", err)
	}

	ids := make([]common.DeliveryID, len(values))
	for index, value := range values {
		ids[index] = common.DeliveryID(value)
	}

	return ids, nil
}

// ReadAttemptScheduleSnapshot returns the current depth of the durable attempt
// schedule together with its oldest scheduled timestamp when one exists.
func (store *AttemptExecutionStore) ReadAttemptScheduleSnapshot(ctx context.Context) (telemetry.AttemptScheduleSnapshot, error) {
	if store == nil || store.client == nil {
		return telemetry.AttemptScheduleSnapshot{}, errors.New("read attempt schedule snapshot: nil store")
	}
	if ctx == nil {
		return telemetry.AttemptScheduleSnapshot{}, errors.New("read attempt schedule snapshot: nil context")
	}

	depth, err := store.client.ZCard(ctx, store.keys.AttemptSchedule()).Result()
	if err != nil {
		return telemetry.AttemptScheduleSnapshot{}, fmt.Errorf("read attempt schedule snapshot: depth: %w", err)
	}

	snapshot := telemetry.AttemptScheduleSnapshot{
		Depth: depth,
	}
	if depth == 0 {
		return snapshot, nil
	}

	values, err := store.client.ZRangeWithScores(ctx, store.keys.AttemptSchedule(), 0, 0).Result()
	if err != nil {
		return telemetry.AttemptScheduleSnapshot{}, fmt.Errorf("read attempt schedule snapshot: oldest scheduled entry: %w", err)
	}
	if len(values) == 0 {
		return snapshot, nil
	}

	oldestScheduledFor := time.UnixMilli(int64(values[0].Score)).UTC()
	snapshot.OldestScheduledFor = &oldestScheduledFor
	return snapshot, nil
}

// SendingDeliveryIDs returns every delivery id currently indexed as
// `mail_delivery.status=sending`.
func (store *AttemptExecutionStore) SendingDeliveryIDs(ctx context.Context) ([]common.DeliveryID, error) {
	if store == nil || store.client == nil {
		return nil, errors.New("sending delivery ids: nil store")
	}
	if ctx == nil {
		return nil, errors.New("sending delivery ids: nil context")
	}

	values, err := store.client.ZRange(ctx, store.keys.StatusIndex(deliverydomain.StatusSending), 0, -1).Result()
	if err != nil {
		return nil, fmt.Errorf("sending delivery ids: %w", err)
	}

	ids := make([]common.DeliveryID, len(values))
	for index, value := range values {
		ids[index] = common.DeliveryID(value)
	}

	return ids, nil
}

// RemoveScheduledDelivery removes deliveryID from the attempt schedule set.
func (store *AttemptExecutionStore) RemoveScheduledDelivery(ctx context.Context, deliveryID common.DeliveryID) error {
	if store == nil || store.client == nil {
		return errors.New("remove scheduled delivery: nil store")
	}
	if ctx == nil {
		return errors.New("remove scheduled delivery: nil context")
	}
	if err := deliveryID.Validate(); err != nil {
		return fmt.Errorf("remove scheduled delivery: %w", err)
	}

	if err := store.client.ZRem(ctx, store.keys.AttemptSchedule(), deliveryID.String()).Err(); err != nil {
		return fmt.Errorf("remove scheduled delivery: %w", err)
	}

	return nil
}

// LoadWorkItem loads the current delivery and its latest attempt when both are
// present.
func (store *AttemptExecutionStore) LoadWorkItem(ctx context.Context, deliveryID common.DeliveryID) (executeattempt.WorkItem, bool, error) {
	if store == nil || store.client == nil {
		return executeattempt.WorkItem{}, false, errors.New("load attempt work item: nil store")
	}
	if ctx == nil {
		return executeattempt.WorkItem{}, false, errors.New("load attempt work item: nil context")
	}
	if err := deliveryID.Validate(); err != nil {
		return executeattempt.WorkItem{}, false, fmt.Errorf("load attempt work item: %w", err)
	}

	deliveryRecord, found, err := store.loadDelivery(ctx, deliveryID)
	if err != nil || !found {
		return executeattempt.WorkItem{}, found, err
	}
	if deliveryRecord.AttemptCount < 1 {
		return executeattempt.WorkItem{}, false, nil
	}

	attemptRecord, found, err := store.loadAttempt(ctx, deliveryID, deliveryRecord.AttemptCount)
	if err != nil || !found {
		return executeattempt.WorkItem{}, found, err
	}

	return executeattempt.WorkItem{
		Delivery: deliveryRecord,
		Attempt:  attemptRecord,
	}, true, nil
}

// LoadPayload loads one stored raw attachment payload bundle.
func (store *AttemptExecutionStore) LoadPayload(ctx context.Context, deliveryID common.DeliveryID) (acceptgenericdelivery.DeliveryPayload, bool, error) {
	if store == nil || store.client == nil {
		return acceptgenericdelivery.DeliveryPayload{}, false, errors.New("load attempt payload: nil store")
	}
	if ctx == nil {
		return acceptgenericdelivery.DeliveryPayload{}, false, errors.New("load attempt payload: nil context")
	}
	if err := deliveryID.Validate(); err != nil {
		return acceptgenericdelivery.DeliveryPayload{}, false, fmt.Errorf("load attempt payload: %w", err)
	}

	payload, err := store.client.Get(ctx, store.keys.DeliveryPayload(deliveryID)).Bytes()
	switch {
	case errors.Is(err, redis.Nil):
		return acceptgenericdelivery.DeliveryPayload{}, false, nil
	case err != nil:
		return acceptgenericdelivery.DeliveryPayload{}, false, fmt.Errorf("load attempt payload: %w", err)
	}

	record, err := UnmarshalDeliveryPayload(payload)
	if err != nil {
		return acceptgenericdelivery.DeliveryPayload{}, false, fmt.Errorf("load attempt payload: %w", err)
	}

	return record, true, nil
}

// ClaimDueAttempt transitions one due scheduled attempt into `in_progress`
// ownership and returns the claimed work item.
func (store *AttemptExecutionStore) ClaimDueAttempt(ctx context.Context, deliveryID common.DeliveryID, now time.Time) (executeattempt.WorkItem, bool, error) {
	if store == nil || store.client == nil {
		return executeattempt.WorkItem{}, false, errors.New("claim due attempt: nil store")
	}
	if ctx == nil {
		return executeattempt.WorkItem{}, false, errors.New("claim due attempt: nil context")
	}
	if err := deliveryID.Validate(); err != nil {
		return executeattempt.WorkItem{}, false, fmt.Errorf("claim due attempt: %w", err)
	}

	claimedAt := now.UTC().Truncate(time.Millisecond)
	if claimedAt.IsZero() {
		return executeattempt.WorkItem{}, false, errors.New("claim due attempt: zero claim time")
	}

	deliveryKey := store.keys.Delivery(deliveryID)

	var claimed executeattempt.WorkItem

	watchErr := store.client.Watch(ctx, func(tx *redis.Tx) error {
		deliveryRecord, err := loadDeliveryFromTx(ctx, tx, deliveryKey)
		switch {
		case errors.Is(err, ErrConflict):
			return errNotClaimable
		case err != nil:
			return fmt.Errorf("claim due attempt: %w", err)
		}
		if deliveryRecord.AttemptCount < 1 {
			return errNotClaimable
		}

		attemptKey := store.keys.Attempt(deliveryID, deliveryRecord.AttemptCount)
		attemptRecord, err := loadAttemptFromTx(ctx, tx, attemptKey)
		switch {
		case errors.Is(err, ErrConflict):
			return errNotClaimable
		case err != nil:
			return fmt.Errorf("claim due attempt: %w", err)
		}

		score, err := tx.ZScore(ctx, store.keys.AttemptSchedule(), deliveryID.String()).Result()
		switch {
		case errors.Is(err, redis.Nil):
			return errNotClaimable
		case err != nil:
			return fmt.Errorf("claim due attempt: read attempt schedule: %w", err)
		}

		switch deliveryRecord.Status {
		case deliverydomain.StatusQueued, deliverydomain.StatusRendered:
		default:
			return errNotClaimable
		}
		if attemptRecord.Status != attempt.StatusScheduled {
			return errNotClaimable
		}
		if score > ScheduledForScore(claimedAt) || attemptRecord.ScheduledFor.After(claimedAt) {
			return errNotClaimable
		}

		claimedDelivery := deliveryRecord
		claimedDelivery.Status = deliverydomain.StatusSending
		claimedDelivery.UpdatedAt = claimedAt
		if err := claimedDelivery.Validate(); err != nil {
			return fmt.Errorf("claim due attempt: build claimed delivery: %w", err)
		}

		claimedAttempt := attemptRecord
		claimedAttempt.Status = attempt.StatusInProgress
		claimedAttempt.StartedAt = ptrTime(claimedAt)
		if err := claimedAttempt.Validate(); err != nil {
			return fmt.Errorf("claim due attempt: build claimed attempt: %w", err)
		}

		deliveryPayload, err := MarshalDelivery(claimedDelivery)
		if err != nil {
			return fmt.Errorf("claim due attempt: %w", err)
		}
		attemptPayload, err := MarshalAttempt(claimedAttempt)
		if err != nil {
			return fmt.Errorf("claim due attempt: %w", err)
		}

		deliveryTTL, err := ttlForExistingKey(ctx, tx, deliveryKey, DeliveryTTL)
		if err != nil {
			return fmt.Errorf("claim due attempt: delivery ttl: %w", err)
		}
		attemptTTL, err := ttlForExistingKey(ctx, tx, attemptKey, AttemptTTL)
		if err != nil {
			return fmt.Errorf("claim due attempt: attempt ttl: %w", err)
		}

		createdAtScore := CreatedAtScore(deliveryRecord.CreatedAt)

		_, err = tx.TxPipelined(ctx, func(pipe redis.Pipeliner) error {
			pipe.Set(ctx, deliveryKey, deliveryPayload, deliveryTTL)
			pipe.Set(ctx, attemptKey, attemptPayload, attemptTTL)
			pipe.ZRem(ctx, store.keys.StatusIndex(deliveryRecord.Status), deliveryID.String())
			pipe.ZAdd(ctx, store.keys.StatusIndex(deliverydomain.StatusSending), redis.Z{
				Score:  createdAtScore,
				Member: deliveryID.String(),
			})
			pipe.ZRem(ctx, store.keys.AttemptSchedule(), deliveryID.String())
			return nil
		})
		if err != nil {
			return fmt.Errorf("claim due attempt: %w", err)
		}

		claimed = executeattempt.WorkItem{
			Delivery: claimedDelivery,
			Attempt:  claimedAttempt,
		}
		return nil
	}, deliveryKey)

	switch {
	case errors.Is(watchErr, errNotClaimable), errors.Is(watchErr, redis.TxFailedErr):
		return executeattempt.WorkItem{}, false, nil
	case watchErr != nil:
		return executeattempt.WorkItem{}, false, watchErr
	default:
		return claimed, true, nil
	}
}

// Commit atomically stores one complete attempt execution outcome.
func (store *AttemptExecutionStore) Commit(ctx context.Context, input executeattempt.CommitStateInput) error {
	if store == nil || store.client == nil {
		return errors.New("commit attempt outcome: nil store")
	}
	if ctx == nil {
		return errors.New("commit attempt outcome: nil context")
	}
	if err := input.Validate(); err != nil {
		return fmt.Errorf("commit attempt outcome: %w", err)
	}

	deliveryKey := store.keys.Delivery(input.Delivery.DeliveryID)
	currentAttemptKey := store.keys.Attempt(input.Attempt.DeliveryID, input.Attempt.AttemptNo)

	deliveryPayload, err := MarshalDelivery(input.Delivery)
	if err != nil {
		return fmt.Errorf("commit attempt outcome: %w", err)
	}
	attemptPayload, err := MarshalAttempt(input.Attempt)
	if err != nil {
		return fmt.Errorf("commit attempt outcome: %w", err)
	}

	var (
		nextAttemptKey     string
		nextAttemptPayload []byte
		nextAttemptScore   float64
		deadLetterKey      string
		deadLetterPayload  []byte
	)
	if input.NextAttempt != nil {
		nextAttemptKey = store.keys.Attempt(input.NextAttempt.DeliveryID, input.NextAttempt.AttemptNo)
		nextAttemptPayload, err = MarshalAttempt(*input.NextAttempt)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: %w", err)
		}
		nextAttemptScore = ScheduledForScore(input.NextAttempt.ScheduledFor)
	}
	if input.DeadLetter != nil {
		deadLetterKey = store.keys.DeadLetter(input.DeadLetter.DeliveryID)
		deadLetterPayload, err = MarshalDeadLetter(*input.DeadLetter)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: %w", err)
		}
	}

	watchKeys := []string{deliveryKey, currentAttemptKey}
	if nextAttemptKey != "" {
		watchKeys = append(watchKeys, nextAttemptKey)
	}
	if deadLetterKey != "" {
		watchKeys = append(watchKeys, deadLetterKey)
	}

	watchErr := store.client.Watch(ctx, func(tx *redis.Tx) error {
		currentDelivery, err := loadDeliveryFromTx(ctx, tx, deliveryKey)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: %w", err)
		}
		currentAttempt, err := loadAttemptFromTx(ctx, tx, currentAttemptKey)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: %w", err)
		}
		if currentDelivery.Status != deliverydomain.StatusSending {
			return fmt.Errorf("commit attempt outcome: %w", ErrConflict)
		}
		if currentAttempt.Status != attempt.StatusInProgress {
			return fmt.Errorf("commit attempt outcome: %w", ErrConflict)
		}
		if nextAttemptKey != "" {
			if err := ensureKeyAbsent(ctx, tx, nextAttemptKey); err != nil {
				return fmt.Errorf("commit attempt outcome: %w", err)
			}
		}
		if deadLetterKey != "" {
			if err := ensureKeyAbsent(ctx, tx, deadLetterKey); err != nil {
				return fmt.Errorf("commit attempt outcome: %w", err)
			}
		}

		deliveryTTL, err := ttlForExistingKey(ctx, tx, deliveryKey, DeliveryTTL)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: delivery ttl: %w", err)
		}
		attemptTTL, err := ttlForExistingKey(ctx, tx, currentAttemptKey, AttemptTTL)
		if err != nil {
			return fmt.Errorf("commit attempt outcome: attempt ttl: %w", err)
		}
		createdAtScore := CreatedAtScore(currentDelivery.CreatedAt)

		_, err = tx.TxPipelined(ctx, func(pipe redis.Pipeliner) error {
			pipe.Set(ctx, deliveryKey, deliveryPayload, deliveryTTL)
			pipe.Set(ctx, currentAttemptKey, attemptPayload, attemptTTL)
			pipe.ZRem(ctx, store.keys.StatusIndex(currentDelivery.Status), input.Delivery.DeliveryID.String())
			pipe.ZAdd(ctx, store.keys.StatusIndex(input.Delivery.Status), redis.Z{
				Score:  createdAtScore,
				Member: input.Delivery.DeliveryID.String(),
			})
			pipe.ZRem(ctx, store.keys.AttemptSchedule(), input.Delivery.DeliveryID.String())
			if nextAttemptKey != "" {
				pipe.Set(ctx, nextAttemptKey, nextAttemptPayload, AttemptTTL)
				pipe.ZAdd(ctx, store.keys.AttemptSchedule(), redis.Z{
					Score:  nextAttemptScore,
					Member: input.Delivery.DeliveryID.String(),
				})
			}
			if deadLetterKey != "" {
				pipe.Set(ctx, deadLetterKey, deadLetterPayload, DeadLetterTTL)
			}
			return nil
		})
		if err != nil {
			return fmt.Errorf("commit attempt outcome: %w", err)
		}

		return nil
	}, watchKeys...)

	switch {
	case errors.Is(watchErr, redis.TxFailedErr):
		return fmt.Errorf("commit attempt outcome: %w", ErrConflict)
	case watchErr != nil:
		return watchErr
	default:
		return nil
	}
}

func (store *AttemptExecutionStore) loadDelivery(ctx context.Context, deliveryID common.DeliveryID) (deliverydomain.Delivery, bool, error) {
	payload, err := store.client.Get(ctx, store.keys.Delivery(deliveryID)).Bytes()
	switch {
	case errors.Is(err, redis.Nil):
		return deliverydomain.Delivery{}, false, nil
	case err != nil:
		return deliverydomain.Delivery{}, false, fmt.Errorf("load attempt delivery: %w", err)
	}

	record, err := UnmarshalDelivery(payload)
	if err != nil {
		return deliverydomain.Delivery{}, false, fmt.Errorf("load attempt delivery: %w", err)
	}

	return record, true, nil
}

func (store *AttemptExecutionStore) loadAttempt(ctx context.Context, deliveryID common.DeliveryID, attemptNo int) (attempt.Attempt, bool, error) {
	payload, err := store.client.Get(ctx, store.keys.Attempt(deliveryID, attemptNo)).Bytes()
	switch {
	case errors.Is(err, redis.Nil):
		return attempt.Attempt{}, false, nil
	case err != nil:
		return attempt.Attempt{}, false, fmt.Errorf("load attempt record: %w", err)
	}

	record, err := UnmarshalAttempt(payload)
	if err != nil {
		return attempt.Attempt{}, false, fmt.Errorf("load attempt record: %w", err)
	}

	return record, true, nil
}

func ptrTime(value time.Time) *time.Time {
	return &value
}