Add asynchron send of encoder metrics

This commit is contained in:
Holger Obermaier 2023-10-11 14:55:52 +02:00
parent 8837400bf2
commit b618e81cbb
3 changed files with 112 additions and 109 deletions

View File

@ -141,8 +141,7 @@ func (s *HttpSink) Write(m lp.CCMetric) error {
// Check that encoding worked
if err != nil {
cclog.ComponentError(s.name, "Write(): Encoding failed:", err)
return err
return fmt.Errorf("Encoding failed: %v", err)
}
if s.config.flushDelay == 0 {
@ -177,6 +176,7 @@ func (s *HttpSink) Write(m lp.CCMetric) error {
return nil
}
// Flush sends all metrics stored in encoder to HTTP server
func (s *HttpSink) Flush() error {
// Lock for encoder usage
@ -193,6 +193,8 @@ func (s *HttpSink) Flush() error {
return nil
}
cclog.ComponentDebug(s.name, "Flush(): Flushing")
var res *http.Response
for i := 0; i < s.config.MaxRetries; i++ {
// Create new request to send buffer
@ -239,12 +241,16 @@ func (s *HttpSink) Flush() error {
}
func (s *HttpSink) Close() {
cclog.ComponentDebug(s.name, "Closing HTTP connection")
// Stop existing timer and immediately flush
if s.flushTimer != nil {
if ok := s.flushTimer.Stop(); ok {
s.timerLock.Unlock()
}
}
// Flush
if err := s.Flush(); err != nil {
cclog.ComponentError(s.name, "Close(): Flush failed:", err)
}

View File

@ -6,7 +6,6 @@ import (
"encoding/json"
"errors"
"fmt"
"strings"
"sync"
"time"
@ -34,14 +33,13 @@ type InfluxSink struct {
// Maximum number of points sent to server in single request.
// Default: 1000
BatchSize int `json:"batch_size,omitempty"`
// Time interval for delayed sending of metrics.
// If the buffers are already filled before the end of this interval,
// the metrics are sent without further delay.
// Default: 1s
FlushInterval string `json:"flush_delay,omitempty"`
// Number of metrics that are dropped when buffer is full
// Default: 100
DropRate int `json:"drop_rate,omitempty"`
flushDelay time.Duration
// Influx client options:
@ -56,15 +54,24 @@ type InfluxSink struct {
// Specify whether to use GZip compression in write requests
InfluxUseGzip bool `json:"use_gzip"`
}
batch []string
flushTimer *time.Timer
flushDelay time.Duration
batchMutex sync.Mutex // Flush() runs in another goroutine, so this lock has to protect the buffer
flushTimerMutex sync.Mutex // Ensure only one flush timer is running
// influx line protocol encoder
encoder influx.Encoder
// number of records stored in the encoder
numRecordsInEncoder int
// List of tags and meta data tags which should be used as tags
extended_tag_list []key_value_pair
// Flush() runs in another goroutine and accesses the influx line protocol encoder,
// so this encoderLock has to protect the encoder and numRecordsInEncoder
encoderLock sync.Mutex
// timer to run Flush()
flushTimer *time.Timer
// Lock to assure that only one timer is running at a time
timerLock sync.Mutex
// WaitGroup to ensure only one send operation is running at a time
sendWaitGroup sync.WaitGroup
}
// connect connects to the InfluxDB server
@ -159,49 +166,11 @@ func (s *InfluxSink) connect() error {
return nil
}
// Write sends metric m in influxDB line protocol
func (s *InfluxSink) Write(m lp.CCMetric) error {
if s.flushDelay != 0 && s.flushTimerMutex.TryLock() {
// Setup flush timer when flush delay is configured
// and no other timer is already running
if s.flushTimer != nil {
// Restarting existing flush timer
cclog.ComponentDebug(s.name, "Write():", "Restarting flush timer")
s.flushTimer.Reset(s.flushDelay)
} else {
// Creating and starting flush timer
cclog.ComponentDebug(s.name, "Write():", "Starting new flush timer")
s.flushTimer = time.AfterFunc(
s.flushDelay,
func() {
defer s.flushTimerMutex.Unlock()
cclog.ComponentDebug(s.name, "Starting flush in flush timer")
if err := s.Flush(); err != nil {
cclog.ComponentError(s.name, "Flush timer: flush failed:", err)
}
})
}
}
// Protect access to batch slice
s.batchMutex.Lock()
// batch slice full, dropping oldest metric(s)
// e.g. when previous flushes failed and batch slice was not cleared
if len(s.batch) == s.config.BatchSize {
newSize := s.config.BatchSize - s.config.DropRate
for i := 0; i < newSize; i++ {
s.batch[i] = s.batch[i+s.config.DropRate]
}
for i := newSize; i < s.config.BatchSize; i++ {
s.batch[i] = ""
}
s.batch = s.batch[:newSize]
cclog.ComponentError(s.name, "Write():", "Batch slice full, dropping", s.config.DropRate, "oldest metric(s)")
}
// Lock for encoder usage
s.encoderLock.Lock()
// Encode measurement name
s.encoder.StartLine(m.Name())
@ -259,70 +228,106 @@ func (s *InfluxSink) Write(m lp.CCMetric) error {
// Encode time stamp
s.encoder.EndLine(m.Time())
// Check that encoding worked
// Check for encoder errors
if err := s.encoder.Err(); err != nil {
cclog.ComponentError(s.name, "Write():", "Encoding failed:", err)
// Unlock encoder usage
s.encoderLock.Unlock()
// Unlock access to batch slice
s.batchMutex.Unlock()
return err
return fmt.Errorf("Encoding failed: %v", err)
}
s.numRecordsInEncoder++
// Append metric to batch slice
s.batch = append(s.batch,
string(
slices.Clone(
s.encoder.Bytes())))
s.encoder.Reset()
if s.config.flushDelay == 0 {
// Unlock encoder usage
s.encoderLock.Unlock()
// Flush synchronously if "flush_delay" is zero
// or
// Flush if batch size is reached
if s.flushDelay == 0 ||
len(s.batch) == s.config.BatchSize {
// Directly flush if no flush delay is configured
return s.Flush()
} else if s.numRecordsInEncoder == s.config.BatchSize {
// Unlock encoder usage
s.encoderLock.Unlock()
// Stop flush timer
if s.flushTimer != nil {
if ok := s.flushTimer.Stop(); ok {
s.flushTimerMutex.Unlock()
cclog.ComponentDebug(s.name, "Write(): Stopped flush timer. Batch size limit reached before flush delay")
s.timerLock.Unlock()
}
}
// Unlock access to batch slice
s.batchMutex.Unlock()
// Flush if batch size is reached
return s.Flush()
} else if s.timerLock.TryLock() {
// Setup flush timer when flush delay is configured
// and no other timer is already running
if s.flushTimer != nil {
// Restarting existing flush timer
cclog.ComponentDebug(s.name, "Write(): Restarting flush timer")
s.flushTimer.Reset(s.config.flushDelay)
} else {
// Creating and starting flush timer
cclog.ComponentDebug(s.name, "Write(): Starting new flush timer")
s.flushTimer = time.AfterFunc(
s.config.flushDelay,
func() {
defer s.timerLock.Unlock()
cclog.ComponentDebug(s.name, "Starting flush triggered by flush timer")
if err := s.Flush(); err != nil {
cclog.ComponentError(s.name, "Flush triggered by flush timer: flush failed:", err)
}
})
}
}
// Unlock access to batch slice
s.batchMutex.Unlock()
// Unlock encoder usage
s.encoderLock.Unlock()
return nil
}
// Flush sends all metrics buffered in batch slice to InfluxDB server
// Flush sends all metrics stored in encoder to InfluxDB server
func (s *InfluxSink) Flush() error {
cclog.ComponentDebug(s.name, "Flushing")
// Lock access to batch slice
s.batchMutex.Lock()
defer s.batchMutex.Unlock()
// Lock for encoder usage
// Own lock for as short as possible: the time it takes to clone the buffer.
s.encoderLock.Lock()
// Nothing to do, batch slice is empty
if len(s.batch) == 0 {
buf := slices.Clone(s.encoder.Bytes())
numRecordsInBuf := s.numRecordsInEncoder
s.encoder.Reset()
s.numRecordsInEncoder = 0
// Unlock encoder usage
s.encoderLock.Unlock()
if len(buf) == 0 {
return nil
}
// Send metrics from batch slice
err := s.writeApi.WriteRecord(context.Background(), strings.Join(s.batch, ""))
if err != nil {
cclog.ComponentError(s.name, "Flush():", "Flush of", len(s.batch), "metrics failed:", err)
return err
}
cclog.ComponentDebug(s.name, "Flush(): Flushing", numRecordsInBuf, "metrics")
// Clear batch slice
for i := range s.batch {
s.batch[i] = ""
// Asynchron send of encoder metrics
s.sendWaitGroup.Wait()
s.sendWaitGroup.Add(1)
go func() {
defer s.sendWaitGroup.Done()
startTime := time.Now()
err := s.writeApi.WriteRecord(context.Background(), string(buf))
if err != nil {
cclog.ComponentError(
s.name,
"Flush():",
"Flush failed:", err,
"(number of records =", numRecordsInBuf,
", buffer size =", len(buf),
", send duration =", time.Since(startTime),
")",
)
return
}
s.batch = s.batch[:0]
}()
return nil
}
@ -333,14 +338,18 @@ func (s *InfluxSink) Close() {
// Stop existing timer and immediately flush
if s.flushTimer != nil {
if ok := s.flushTimer.Stop(); ok {
s.flushTimerMutex.Unlock()
s.timerLock.Unlock()
}
}
s.Flush()
// Flush
if err := s.Flush(); err != nil {
cclog.ComponentError(s.name, "Close():", "Flush failed:", err)
}
// Wait for send operation to finish
s.sendWaitGroup.Wait()
s.client.Close()
}
@ -352,7 +361,6 @@ func NewInfluxSink(name string, config json.RawMessage) (Sink, error) {
// Set config default values
s.config.BatchSize = 1000
s.config.FlushInterval = "1s"
s.config.DropRate = 100
// Read config
if len(config) > 0 {
@ -388,24 +396,13 @@ func NewInfluxSink(name string, config json.RawMessage) (Sink, error) {
if len(s.config.FlushInterval) > 0 {
t, err := time.ParseDuration(s.config.FlushInterval)
if err == nil {
s.flushDelay = t
s.config.flushDelay = t
}
}
if !(s.config.BatchSize > 0) {
return s, fmt.Errorf("batch_size=%d in InfluxDB config must be > 0", s.config.BatchSize)
}
if !(s.config.DropRate > 0) {
return s, fmt.Errorf("drop_rate=%d in InfluxDB config must be > 0", s.config.DropRate)
}
if !(s.config.BatchSize > s.config.DropRate) {
return s, fmt.Errorf(
"batch_size=%d must be greater then drop_rate=%d in InfluxDB config",
s.config.BatchSize, s.config.DropRate)
}
// allocate batch slice
s.batch = make([]string, 0, s.config.BatchSize)
// Connect to InfluxDB server
if err := s.connect(); err != nil {

View File

@ -27,9 +27,9 @@ The `influxdb` sink uses the official [InfluxDB golang client](https://pkg.go.de
- `meta_as_tags`: print all meta information as tags in the output (optional)
- `database`: All metrics are written to this bucket
- `host`: Hostname of the InfluxDB database server
- `port`: Portnumber (as string) of the InfluxDB database server
- `user`: Username for basic authentification
- `password`: Password for basic authentification
- `port`: Port number (as string) of the InfluxDB database server
- `user`: Username for basic authentication
- `password`: Password for basic authentication
- `organization`: Organization in the InfluxDB
- `ssl`: Use SSL connection
- `flush_delay`: Group metrics coming in to a single batch