VictoriaMetrics/app/vmalert/remotewrite/remotewrite.go

package remotewrite

import (
	"bytes"
	"context"
	"flag"
	"fmt"
	"io"
	"net/http"
	"path"
	"strings"
	"sync"
	"time"

	"github.com/golang/snappy"

	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promauth"
	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
	"github.com/VictoriaMetrics/metrics"
)

var (
	disablePathAppend = flag.Bool("remoteWrite.disablePathAppend", false, "Whether to disable automatic appending of '/api/v1/write' path to the configured -remoteWrite.url.")
	sendTimeout       = flag.Duration("remoteWrite.sendTimeout", 30*time.Second, "Timeout for sending data to the configured -remoteWrite.url.")
)

// Client is an asynchronous HTTP client for writing
// timeseries via remote write protocol.
type Client struct {
	addr          string
	c             *http.Client
	authCfg       *promauth.Config
	input         chan prompbmarshal.TimeSeries
	flushInterval time.Duration
	maxBatchSize  int
	maxQueueSize  int

	wg     sync.WaitGroup
	doneCh chan struct{}
}

// Config is config for remote write.
type Config struct {
	// Addr of remote storage
	Addr    string
	AuthCfg *promauth.Config

	// Concurrency defines number of readers that
	// concurrently read from the queue and flush data
	Concurrency int
	// MaxBatchSize defines max number of timeseries
	// to be flushed at once
	MaxBatchSize int
	// MaxQueueSize defines max length of input queue
	// populated by Push method.
	// Push will be rejected once queue is full.
	MaxQueueSize int
	// FlushInterval defines time interval for flushing batches
	FlushInterval time.Duration
	// Transport will be used by the underlying http.Client
	Transport *http.Transport
}

const (
	defaultConcurrency   = 4
	defaultMaxBatchSize  = 1e3
	defaultMaxQueueSize  = 1e5
	defaultFlushInterval = 5 * time.Second
	defaultWriteTimeout  = 30 * time.Second
)

// NewClient returns asynchronous client for
// writing timeseries via remotewrite protocol.
func NewClient(ctx context.Context, cfg Config) (*Client, error) {
	if cfg.Addr == "" {
		return nil, fmt.Errorf("config.Addr can't be empty")
	}
	if cfg.MaxBatchSize == 0 {
		cfg.MaxBatchSize = defaultMaxBatchSize
	}
	if cfg.MaxQueueSize == 0 {
		cfg.MaxQueueSize = defaultMaxQueueSize
	}
	if cfg.FlushInterval == 0 {
		cfg.FlushInterval = defaultFlushInterval
	}
	if cfg.Transport == nil {
		cfg.Transport = http.DefaultTransport.(*http.Transport).Clone()
	}
	cc := defaultConcurrency
	if cfg.Concurrency > 0 {
		cc = cfg.Concurrency
	}
	c := &Client{
		c: &http.Client{
			Timeout:   *sendTimeout,
			Transport: cfg.Transport,
		},
		addr:          strings.TrimSuffix(cfg.Addr, "/"),
		authCfg:       cfg.AuthCfg,
		flushInterval: cfg.FlushInterval,
		maxBatchSize:  cfg.MaxBatchSize,
		maxQueueSize:  cfg.MaxQueueSize,
		doneCh:        make(chan struct{}),
		input:         make(chan prompbmarshal.TimeSeries, cfg.MaxQueueSize),
	}

	for i := 0; i < cc; i++ {
		c.run(ctx)
	}
	return c, nil
}

// Push adds timeseries into queue for writing into remote storage.
// Push returns and error if client is stopped or if queue is full.
func (c *Client) Push(s prompbmarshal.TimeSeries) error {
	select {
	case <-c.doneCh:
		return fmt.Errorf("client is closed")
	case c.input <- s:
		return nil
	default:
		return fmt.Errorf("failed to push timeseries - queue is full (%d entries). "+
			"Queue size is controlled by -remoteWrite.maxQueueSize flag",
			c.maxQueueSize)
	}
}

// Close stops the client and waits for all goroutines
// to exit.
func (c *Client) Close() error {
	if c.doneCh == nil {
		return fmt.Errorf("client is already closed")
	}
	close(c.input)
	close(c.doneCh)
	c.wg.Wait()
	return nil
}

func (c *Client) run(ctx context.Context) {
	ticker := time.NewTicker(c.flushInterval)
	wr := &prompbmarshal.WriteRequest{}
	shutdown := func() {
		for ts := range c.input {
			wr.Timeseries = append(wr.Timeseries, ts)
		}
		lastCtx, cancel := context.WithTimeout(context.Background(), defaultWriteTimeout)
		c.flush(lastCtx, wr)
		cancel()
	}
	c.wg.Add(1)
	go func() {
		defer c.wg.Done()
		defer ticker.Stop()
		for {
			select {
			case <-c.doneCh:
				shutdown()
				return
			case <-ctx.Done():
				shutdown()
				return
			case <-ticker.C:
				c.flush(ctx, wr)
			case ts, ok := <-c.input:
				if !ok {
					continue
				}
				wr.Timeseries = append(wr.Timeseries, ts)
				if len(wr.Timeseries) >= c.maxBatchSize {
					c.flush(ctx, wr)
				}
			}
		}
	}()
}

var (
	sentRows            = metrics.NewCounter(`vmalert_remotewrite_sent_rows_total`)
	sentBytes           = metrics.NewCounter(`vmalert_remotewrite_sent_bytes_total`)
	droppedRows         = metrics.NewCounter(`vmalert_remotewrite_dropped_rows_total`)
	droppedBytes        = metrics.NewCounter(`vmalert_remotewrite_dropped_bytes_total`)
	bufferFlushDuration = metrics.NewHistogram(`vmalert_remotewrite_flush_duration_seconds`)
)

// flush is a blocking function that marshals WriteRequest and sends
// it to remote-write endpoint. Flush performs limited amount of retries
// if request fails.
func (c *Client) flush(ctx context.Context, wr *prompbmarshal.WriteRequest) {
	if len(wr.Timeseries) < 1 {
		return
	}
	defer prompbmarshal.ResetWriteRequest(wr)
	defer bufferFlushDuration.UpdateDuration(time.Now())

	data, err := wr.Marshal()
	if err != nil {
		logger.Errorf("failed to marshal WriteRequest: %s", err)
		return
	}

	b := snappy.Encode(nil, data)

	const (
		retryCount   = 5
		retryBackoff = time.Second
	)

	for attempts := 0; attempts < retryCount; attempts++ {
		err := c.send(ctx, b)
		if err == nil {
			sentRows.Add(len(wr.Timeseries))
			sentBytes.Add(len(b))
			return
		}

		_, isRetriable := err.(*retriableError)
		logger.Warnf("attempt %d to send request failed: %s (retriable: %v)", attempts+1, err, isRetriable)

		if !isRetriable {
			// exit fast if error isn't retriable
			break
		}

		// check if request has been cancelled before backoff
		select {
		case <-ctx.Done():
			break
		default:
		}

		// sleeping to avoid remote db hammering
		time.Sleep(retryBackoff)
	}

	droppedRows.Add(len(wr.Timeseries))
	droppedBytes.Add(len(b))
	logger.Errorf("attempts to send remote-write request failed - dropping %d time series",
		len(wr.Timeseries))
}

func (c *Client) send(ctx context.Context, data []byte) error {
	r := bytes.NewReader(data)
	req, err := http.NewRequest(http.MethodPost, c.addr, r)
	if err != nil {
		return fmt.Errorf("failed to create new HTTP request: %w", err)
	}

	// RFC standard compliant headers
	req.Header.Set("Content-Encoding", "snappy")
	req.Header.Set("Content-Type", "application/x-protobuf")

	// Prometheus compliant headers
	req.Header.Set("X-Prometheus-Remote-Write-Version", "0.1.0")

	if c.authCfg != nil {
		c.authCfg.SetHeaders(req, true)
	}
	if !*disablePathAppend {
		req.URL.Path = path.Join(req.URL.Path, "/api/v1/write")
	}
	resp, err := c.c.Do(req.WithContext(ctx))
	if err != nil {
		return fmt.Errorf("error while sending request to %s: %w; Data len %d(%d)",
			req.URL.Redacted(), err, len(data), r.Size())
	}
	defer func() { _ = resp.Body.Close() }()

	body, _ := io.ReadAll(resp.Body)

	// according to https://prometheus.io/docs/concepts/remote_write_spec/
	// Prometheus remote Write compatible receivers MUST
	switch resp.StatusCode / 100 {
	case 2:
		// respond with a HTTP 2xx status code when the write is successful.
		return nil
	case 5:
		// respond with HTTP status code 5xx when the write fails and SHOULD be retried.
		return &retriableError{fmt.Errorf("unexpected response code %d for %s. Response body %q",
			resp.StatusCode, req.URL.Redacted(), body)}
	default:
		// respond with HTTP status code 4xx when the request is invalid, will never be able to succeed
		// and should not be retried.
		return fmt.Errorf("unexpected response code %d for %s. Response body %q",
			resp.StatusCode, req.URL.Redacted(), body)
	}
}

type retriableError struct {
	err error
}

func (e *retriableError) Error() string {
	return e.err.Error()
}