NixMsg/internal/store/queue.go

package store

import (
	"context"
	"database/sql"
	"errors"
	"fmt"
	"sync"
	"sync/atomic"
	"time"
)

// ErrQueueClosed 表示写入队列已关闭。
var ErrQueueClosed = errors.New("store: write queue closed")

// ErrBusy 表示写库基础设施失败（可映射为协议 busy；/readyz 应失败）。
var ErrBusy = errors.New("store: busy")

const (
	maxBatchOps   = 256
	batchWait     = 2 * time.Millisecond
	queueBuffSize = 1024
)

// WriteFunc 在单个写事务中执行的操作。
type WriteFunc func(tx *sql.Tx) error

type writeJob struct {
	ctx context.Context
	fn  WriteFunc
	res chan error
}

// Queue 写入队列：写 goroutine 合并提交（最多 256 个或凑满 2ms），每操作用 SAVEPOINT 隔离。
type Queue struct {
	db *sql.DB

	ch     chan writeJob
	done   chan struct{}
	closed atomic.Bool

	mu           sync.Mutex
	ready        bool
	lastWriteErr error
	pending      int
}

// NewQueue 创建合并写入队列并启动写 goroutine。
func NewQueue(db *sql.DB) *Queue {
	q := &Queue{
		db:    db,
		ch:    make(chan writeJob, queueBuffSize),
		done:  make(chan struct{}),
		ready: true,
	}
	go q.loop()
	return q
}

// Do 提交写操作并等待提交结果。
func (q *Queue) Do(ctx context.Context, fn WriteFunc) error {
	if fn == nil {
		return errors.New("store: nil write func")
	}
	if q.closed.Load() {
		return ErrQueueClosed
	}
	if err := ctx.Err(); err != nil {
		return err
	}
	job := writeJob{ctx: ctx, fn: fn, res: make(chan error, 1)}
	q.mu.Lock()
	q.pending++
	q.mu.Unlock()
	select {
	case q.ch <- job:
	case <-ctx.Done():
		q.mu.Lock()
		q.pending--
		q.mu.Unlock()
		return ctx.Err()
	case <-q.done:
		q.mu.Lock()
		q.pending--
		q.mu.Unlock()
		return ErrQueueClosed
	}
	select {
	case err := <-job.res:
		return err
	case <-ctx.Done():
		// 操作可能仍在队列中执行；结果通道仍会被写端关闭式填入。
		select {
		case err := <-job.res:
			if err != nil {
				return err
			}
			return ctx.Err()
		case <-time.After(30 * time.Second):
			return ctx.Err()
		}
	}
}

func (q *Queue) loop() {
	defer close(q.done)
	for {
		job, ok := <-q.ch
		if !ok {
			return
		}
		batch := []writeJob{job}
		timer := time.NewTimer(batchWait)
	collect:
		for len(batch) < maxBatchOps {
			select {
			case j, ok := <-q.ch:
				if !ok {
					break collect
				}
				batch = append(batch, j)
			case <-timer.C:
				break collect
			}
		}
		timer.Stop()
		q.runBatch(batch)
	}
}

func (q *Queue) runBatch(batch []writeJob) {
	defer func() {
		q.mu.Lock()
		q.pending -= len(batch)
		if q.pending < 0 {
			q.pending = 0
		}
		q.mu.Unlock()
	}()

	// 过滤已取消的任务。
	active := make([]writeJob, 0, len(batch))
	for _, j := range batch {
		if err := j.ctx.Err(); err != nil {
			j.res <- err
			continue
		}
		active = append(active, j)
	}
	if len(active) == 0 {
		return
	}

	tx, err := q.db.BeginTx(context.Background(), nil)
	if err != nil {
		q.markBusy(err)
		for _, j := range active {
			j.res <- errors.Join(ErrBusy, err)
		}
		return
	}

	type outcome struct {
		job     writeJob
		opErr   error
		success bool
	}
	outcomes := make([]outcome, 0, len(active))
	for i, j := range active {
		sp := fmt.Sprintf("sp_%d", i)
		if _, err := tx.Exec("SAVEPOINT " + sp); err != nil {
			_ = tx.Rollback()
			q.markBusy(err)
			for _, o := range outcomes {
				if o.success {
					o.job.res <- errors.Join(ErrBusy, err)
				}
			}
			for _, rest := range active[i:] {
				rest.res <- errors.Join(ErrBusy, err)
			}
			return
		}
		opErr := j.fn(tx)
		if opErr != nil {
			if _, rbErr := tx.Exec("ROLLBACK TO " + sp); rbErr != nil {
				_ = tx.Rollback()
				q.markBusy(rbErr)
				for _, o := range outcomes {
					if o.success {
						o.job.res <- errors.Join(ErrBusy, rbErr)
					}
				}
				j.res <- opErr
				for _, rest := range active[i+1:] {
					rest.res <- errors.Join(ErrBusy, rbErr)
				}
				return
			}
			_, _ = tx.Exec("RELEASE " + sp)
			outcomes = append(outcomes, outcome{job: j, opErr: opErr, success: false})
			continue
		}
		if _, err := tx.Exec("RELEASE " + sp); err != nil {
			_ = tx.Rollback()
			q.markBusy(err)
			for _, o := range outcomes {
				if o.success {
					o.job.res <- errors.Join(ErrBusy, err)
				}
			}
			j.res <- errors.Join(ErrBusy, err)
			for _, rest := range active[i+1:] {
				rest.res <- errors.Join(ErrBusy, err)
			}
			return
		}
		outcomes = append(outcomes, outcome{job: j, success: true})
	}

	if err := tx.Commit(); err != nil {
		q.markBusy(err)
		for _, o := range outcomes {
			if o.success {
				o.job.res <- errors.Join(ErrBusy, err)
			} else {
				o.job.res <- o.opErr
			}
		}
		return
	}
	for _, o := range outcomes {
		if o.success {
			o.job.res <- nil
		} else {
			o.job.res <- o.opErr
		}
	}
}

func (q *Queue) markBusy(err error) {
	q.mu.Lock()
	defer q.mu.Unlock()
	q.ready = false
	q.lastWriteErr = err
}

// IsReady 写库是否仍可用（写失败后为 false）。
func (q *Queue) IsReady() bool {
	q.mu.Lock()
	defer q.mu.Unlock()
	return q.ready
}

// LastWriteError 返回最近一次基础设施写失败。
func (q *Queue) LastWriteError() error {
	q.mu.Lock()
	defer q.mu.Unlock()
	return q.lastWriteErr
}

// Len 返回排队与批次中的写操作数。
func (q *Queue) Len() int {
	q.mu.Lock()
	defer q.mu.Unlock()
	return q.pending
}

// Drain 等待已入队操作完成，或 ctx 取消。关闭后也会等到 loop 退出。
func (q *Queue) Drain(ctx context.Context) error {
	ticker := time.NewTicker(5 * time.Millisecond)
	defer ticker.Stop()
	for {
		q.mu.Lock()
		n := q.pending
		q.mu.Unlock()
		if n == 0 {
			return nil
		}
		select {
		case <-ctx.Done():
			return ctx.Err()
		case <-ticker.C:
		}
	}
}

// Close 关闭队列：不再接受新任务，并等待写 goroutine 处理完已入队任务后退出。
func (q *Queue) Close() error {
	if q.closed.Swap(true) {
		return nil
	}
	close(q.ch)
	<-q.done
	return nil
}