Optimize writes to InfluxDB

Use these tips to optimize performance and system overhead when writing data to InfluxDB.

• Batch writes
• Sort tags by key
• Use the coarsest time precision possible
• Use gzip compression
  • Enable gzip compression in Telegraf
  • Enable gzip compression in InfluxDB client libraries
  • Use gzip compression with the InfluxDB API
• Synchronize hosts with NTP
• Write multiple data points in one request
• Pre-process data before writing
  • Prerequisites
  • Filter data from a batch
  • Coerce data types to avoid rejected point errors
  • Merge lines to optimize memory and bandwidth
  • Avoid sending duplicate data
  • Run custom processing code

Batch writes

Write data in batches to minimize network overhead when writing data to InfluxDB.

Sort tags by key

Before writing data points to InfluxDB, sort tags by key in lexicographic order. Verify sort results match results from the Go bytes.Compare function.

# Line protocol example with unsorted tags
measurement,tagC=therefore,tagE=am,tagA=i,tagD=i,tagB=think fieldKey=fieldValue 1562020262

# Optimized line protocol example with tags sorted by key
measurement,tagA=i,tagB=think,tagC=therefore,tagD=i,tagE=am fieldKey=fieldValue 1562020262

Use the coarsest time precision possible

By default, InfluxDB writes data in nanosecond precision. However if your data isn’t collected in nanoseconds, there is no need to write at that precision. For better performance, use the coarsest precision possible for timestamps.

Specify timestamp precision when writing to InfluxDB.

Use gzip compression

Use gzip compression to speed up writes to InfluxDB. Benchmarks have shown up to a 5x speed improvement when data is compressed.

[[outputs.influxdb_v2]]
  urls = ["https://cloud2.influxdata.com"]
  # ...
  content_encoding = "gzip"

echo "mem,host=host1 used_percent=23.43234543 1641024000
mem,host=host2 used_percent=26.81522361 1641027600
mem,host=host1 used_percent=22.52984738 1641031200
mem,host=host2 used_percent=27.18294630 1641034800" | gzip > system.gzip \

curl --request POST "https://cloud2.influxdata.com/api/v2/write?org=
ORG_NAME
&bucket=
BUCKET_NAME
&precision=s" \
  --header "Authorization: Token 
API_TOKEN
" \
  --header "Content-Type: text/plain; charset=utf-8" \
  --header "Content-Encoding: gzip" \
  --data-binary @system.gzip

Synchronize hosts with NTP

Use the Network Time Protocol (NTP) to synchronize time between hosts. If a timestamp isn’t included in line protocol, InfluxDB uses its host’s local time (in UTC) to assign timestamps to each point. If a host’s clocks isn’t synchronized with NTP, timestamps may be inaccurate.

Write multiple data points in one request

To write multiple lines in one request, each line of line protocol must be delimited by a new line (\n).

Pre-process data before writing

Pre-processing data in your write workload can help you avoid write failures due to schema conflicts or resource use. For example, if you have many devices that write to the same measurement, and some devices use different data types for the same field, then you might want to generate an alert or convert field data to fit your schema before you send the data to InfluxDB.

With Telegraf, you can process data from other services and files and then write it to InfluxDB. In addition to processing data with Telegraf’s included plugins, you can use the Execd processor plugin to integrate your own code and external applications.

The following examples show how to configure the Telegraf agent and plugins to optimize writes. The examples use the File input plugin to read data from a file and use the InfluxDB v2 output plugin to write data to a bucket, but you can use any input and output plugin.

Prerequisites

Install Telegraf if you haven’t already.

Filter data from a batch

Use Telegraf and metric filtering to filter data before writing it to InfluxDB. Configure metric filters to retain or remove data elements (before processor and aggregator plugins run).

1. Enter the following command to create a Telegraf configuration that parses system usage data, removes the specified fields and tags, and then writes the data to InfluxDB:

   cat <<EOF >> ./telegraf.conf
     [[inputs.cpu]]
       # Remove the specified fields from points.
       fieldpass = ["usage_system", "usage_idle"]
       # Remove the specified tags from points.
       tagexclude = ["host"]
     [[outputs.influxdb_v2]]
       urls = ["https://cloud2.influxdata.com"]
       token = "
   API_TOKEN
   "
       organization = ""
       bucket = "
   BUCKET_NAME
   "
   EOF
   

   Replace the following:

   • ORG_NAME: the name of your organization
   • BUCKET_NAME: the name of the bucket to write data to
   • API_TOKEN: a token with write access to the specified bucket. Store this in a secret store or environment variable to avoid exposing the raw token string.

2. To test the input and processor, enter the following command:

   telegraf --test --config telegraf.conf
   

   The output is similar to the following. For each row of input data, the filters pass the metric name, tags, specified fields, and timestamp.

   > cpu,cpu=cpu0 usage_idle=100,usage_system=0 1702067201000000000
   ...
   > cpu,cpu=cpu-total usage_idle=99.80198019802448,usage_system=0.1980198019802045 1702067201000000000
   

Coerce data types to avoid rejected point errors

Use Telegraf and the Converter processor plugin to convert field data types to fit your schema.

For example, if you write the sample data in Get started home sensor data to a bucket, and then try to write the following batch to the same measurement:

home,room=Kitchen temp=23.1,hum=36.6,co=22.1 1641063600
home,room=Living\ Room temp=22i,hum=36.4,co=17i 1641067200
home,room=Kitchen temp=22.7,hum=36.5,co=26i 1641067200

InfluxDB expects co to contain an integer value and rejects points with co floating-point decimal (22.1) values. To avoid the error, configure Telegraf to convert fields to the data types in your schema columns.

The following example converts the temp, hum, and co fields to fit the sample data schema:

1. In your terminal, enter the following command to create the sample data file:

   cat <<EOF > ./home.lp
   home,room=Kitchen temp=23.1,hum=36.6,co=22.1 1641063600
   home,room=Living\ Room temp=22i,hum=36.4,co=17i 1641067200
   home,room=Kitchen temp=22.7,hum=36.5,co=26i 1641067200
   EOF
   

2. Enter the following command to create a Telegraf configuration that parses the sample data, converts the field values to the specified data types, and then writes the data to InfluxDB:

   cat <<EOF > ./telegraf.conf
   [[inputs.file]]
     ## For each interval, parse data from files in the list.
     files = ["home.lp"]
     influx_timestamp_precision = "1s"
     precision = "1s"
     tagexclude = ["host"]
   [[processors.converter]]
     [processors.converter.fields]
       ## A data type and a list of fields to convert to the data type.
       float = ["temp", "hum"]
       integer = ["co"]
   [[outputs.influxdb_v2]]
     ## InfluxDB v2 API credentials and the bucket to write to.
     urls = ["https://cloud2.influxdata.com"]
     token = "
   API_TOKEN
   "
     organization = ""
     bucket = "
   BUCKET_NAME
   "
   EOF
   

   Replace the following:

   • BUCKET_NAME: the name of the bucket to write data to
   • API_TOKEN: a token with write access to the specified bucket. Store this in a secret store or environment variable to avoid exposing the raw token string.

3. To test the input and processor, enter the following command:

   telegraf --test --config telegraf.conf
   

   Telegraf outputs the following to stdout, and then exits:

   > home,room=Kitchen co=22i,hum=36.6,temp=23.1 1641063600000000000
   > home,room=Living\ Room co=17i,hum=36.4,temp=22 1641067200000000000
   > home,room=Kitchen co=26i,hum=36.5,temp=22.7 1641067200000000000
   

Merge lines to optimize memory and bandwidth

Use Telegraf and the Merge aggregator plugin to merge points that share the same measurement, tag set, and timestamp.

The following example creates sample data for two series (the combination of measurement, tag set, and timestamp), and then merges points in each series:

1. In your terminal, enter the following command to create the sample data file and calculate the number of seconds between the earliest timestamp and now. The command assigns the calculated value to a grace_duration variable that you’ll use in the next step.

   cat <<EOF > ./home.lp
   home,room=Kitchen temp=23.1 1641063600
   home,room=Kitchen hum=36.6 1641063600
   home,room=Kitchen co=22i 1641063600
   home,room=Living\ Room temp=22.7 1641063600
   home,room=Living\ Room hum=36.4 1641063600
   home,room=Living\ Room co=17i 1641063600
   EOF
   grace_duration="$(($(date +%s)-1641063000))s"
   

2. Enter the following command to configure Telegraf to parse the file, merge the points, and write the data to InfluxDB–specifically, the configuration sets the following properties:

   • influx_timestamp_precision: for parsers, specifies the timestamp precision in the input data
   • Optional: aggregators.merge.grace extends the duration for merging points. To ensure the sample data is included, the configuration uses the calculated variable from the preceding step.

   cat <<EOF > ./telegraf.conf
   # Parse metrics from a file
   [[inputs.file]]
     ## A list of files to parse during each interval.
     files = ["home.lp"]
     ## The precision of timestamps in your data.
     influx_timestamp_precision = "1s"
     tagexclude = ["host"]
   # Merge separate metrics that share a series key
   [[aggregators.merge]]
     grace = "$grace_duration"
     ## If true, drops the original metric.
     drop_original = true
   # Writes metrics as line protocol to the InfluxDB v2 API
   [[outputs.influxdb_v2]]
     ## InfluxDB credentials and the bucket to write data to.
     urls = ["https://cloud2.influxdata.com"]
     token = "
   API_TOKEN
   "
     organization = ""
     bucket = "
   BUCKET_NAME
   "
   EOF
   

   Replace the following:

   • BUCKET_NAME: the name of the bucket to write data to
   • API_TOKEN: a token with write access to the specified bucket. Store this in a secret store or environment variable to avoid exposing the raw token string.

3. To test the input and aggregator, enter the following command:

   telegraf --test --config telegraf.conf
   

   Telegraf outputs the following to stdout, and then exits:

   > home,room=Kitchen co=22i,hum=36.6,temp=23.1 1641063600000000000
   > home,room=Living\ Room co=17i,hum=36.4,temp=22.7 1641063600000000000
   

Avoid sending duplicate data

Use Telegraf and the Dedup processor plugin to filter data whose field values are exact repetitions of previous values. Deduplicating your data can reduce your write payload size and resource usage.

The following example shows how to use Telegraf to remove points that repeat field values, and then write the data to InfluxDB:

1. In your terminal, enter the following command to create the sample data file and calculate the number of seconds between the earliest timestamp and now. The command assigns the calculated value to a dedup_duration variable that you’ll use in the next step.

   cat <<EOF > ./home.lp
   home,room=Kitchen temp=23.1,hum=36.6,co=22i 1641063600
   home,room=Living\ Room temp=22.5,hum=36.4,co=17i 1641063600
   home,room=Kitchen temp=22.7,hum=36.5,co=26i 1641063605
   home,room=Living\ Room temp=22.5,hum=36.4,co=17i 1641063605
   home,room=Kitchen temp=23.1,hum=36.6,co=22i 1641063610
   home,room=Living\ Room temp=23.0,hum=36.4,co=17i 1641063610
   EOF
   dedup_duration="$(($(date +%s)-1641063000))s"
   

2. Enter the following command to configure Telegraf to parse the file, drop duplicate points, and write the data to InfluxDB–specifically, the sample configuration sets the following:

   • influx_timestamp_precision: for parsers, specifies the timestamp precision in the input data
   • processors.dedup: configures the Dedup processor plugin
   • Optional: processors.dedup.dedup_interval. Points in the range dedup_interval to now are considered for removal. To ensure the sample data is included, the configuration uses the calculated variable from the preceding step.

   cat <<EOF > ./telegraf.conf
   # Parse metrics from a file
   [[inputs.file]]
     ## A list of files to parse during each interval.
     files = ["home.lp"]
     ## The precision of timestamps in your data.
     influx_timestamp_precision = "1s"
     tagexclude = ["host"]
   # Filter metrics that repeat previous field values
   [[processors.dedup]]
     ## Drops duplicates within the specified duration
     dedup_interval = "$dedup_duration"
   # Writes metrics as line protocol to the InfluxDB v2 API
   [[outputs.influxdb_v2]]
     ## InfluxDB credentials and the bucket to write data to.
     urls = ["https://cloud2.influxdata.com"]
     token = "
   API_TOKEN
   "
     organization = ""
     bucket = "
   BUCKET_NAME
   "
   EOF
   

   Replace the following:

   • BUCKET_NAME: the name of the bucket to write data to
   • API_TOKEN: a token with write access to the specified bucket. Store this in a secret store or environment variable to avoid exposing the raw token string.

3. To test the input and processor, enter the following command:

   telegraf --test --config telegraf.conf
   

   Telegraf outputs the following to stdout, and then exits:

   > home,room=Kitchen co=22i,hum=36.6,temp=23.1 1641063600000000000
   > home,room=Living\ Room co=17i,hum=36.4,temp=22.5 1641063600000000000
   > home,room=Kitchen co=26i,hum=36.5,temp=22.7 1641063605000000000
   > home,room=Kitchen co=22i,hum=36.6,temp=23.1 1641063610000000000
   > home,room=Living\ Room co=17i,hum=36.4,temp=23 1641063610000000000
   

Run custom preprocessing code

Use Telegraf and the Execd processor plugin to execute code external to Telegraf and then write the processed data. The Execd plugin expects line protocol data in stdin, passes the data to the configured executable, and then outputs line protocol to stdout.

The following example shows how to use Telegraf to execute Go code for processing metrics and then write the output to InfluxDB. The Go multiplier.go sample code does the following:

1. Imports influx parser and serializer plugins from Telegraf.

2. Parses each line of data into a Telegraf metric.

3. If the metric contains a count field, multiplies the field value by 2; otherwise, prints a message to stderr and exits.

4. In your editor, enter the following sample code and save the file as multiplier.go:

   package main
   
   import (
       "fmt"
       "os"
   
       "github.com/influxdata/telegraf/plugins/parsers/influx"
       influxSerializer "github.com/influxdata/telegraf/plugins/serializers/influx"
   )
   
   func main() {
       parser := influx.NewStreamParser(os.Stdin)
       serializer := influxSerializer.Serializer{}
       if err := serializer.Init(); err != nil {
           fmt.Fprintf(os.Stderr, "serializer init failed: %v\n", err)
           os.Exit(1)
       }
   
       for {
           metric, err := parser.Next()
           if err != nil {
               if err == influx.EOF {
                   return // stream ended
               }
               if parseErr, isParseError := err.(*influx.ParseError); isParseError {
                   fmt.Fprintf(os.Stderr, "parse ERR %v\n", parseErr)
                   os.Exit(1)
               }
               fmt.Fprintf(os.Stderr, "ERR %v\n", err)
               os.Exit(1)
           }
   
           c, found := metric.GetField("count")
           if !found {
               fmt.Fprintf(os.Stderr, "metric has no count field\n")
               os.Exit(1)
           }
           switch t := c.(type) {
           case float64:
               t *= 2
               metric.AddField("count", t)
           case int64:
               t *= 2
               metric.AddField("count", t)
           default:
               fmt.Fprintf(os.Stderr, "count is not an unknown type, it's a %T\n", c)
               os.Exit(1)
           }
           b, err := serializer.Serialize(metric)
           if err != nil {
               fmt.Fprintf(os.Stderr, "ERR %v\n", err)
               os.Exit(1)
           }
           fmt.Fprint(os.Stdout, string(b))
       }
   }
   

5. Initialize the module and install dependencies:

   go mod init processlp
   go mod tidy
   

6. In your terminal, enter the following command to create the sample data file:

   cat <<EOF > ./home.lp
   home,room=Kitchen temp=23.1,count=1 1641063600
   home,room=Living\ Room temp=22.7,count=1 1641063600
   home,room=Kitchen temp=23.1 1641063601
   home,room=Living\ Room temp=22.7 1641063601
   EOF
   

7. Enter the following command to configure Telegraf to parse the file, execute the Go binary, and write the data–specifically, the sample configuration sets the following:

   • influx_timestamp_precision: for parsers, specifies the timestamp precision in the input data
   • processors.execd: configures the Execd plugin
   • processors.execd.command: sets the executable and arguments for Execd to run

   cat <<EOF > ./telegraf.conf
   # Parse metrics from a file
   [[inputs.file]]
     ## A list of files to parse during each interval.
     files = ["home.lp"]
     ## The precision of timestamps in your data.
     influx_timestamp_precision = "1s"
     tagexclude = ["host"]
     # Filter metrics that repeat previous field values
   [[processors.execd]]
     ## A list that contains the executable command and arguments to run as a daemon.
     command = ["go", "run", "multiplier.go"]
     # Writes metrics as line protocol to the InfluxDB v2 API
   [[outputs.influxdb_v2]]
     ## InfluxDB credentials and the bucket to write data to.
     urls = ["https://cloud2.influxdata.com"]
     token = "
   API_TOKEN
   "
     organization = ""
     bucket = "
   BUCKET_NAME
   "
   EOF
   

   Replace the following:

   • BUCKET_NAME: the name of the bucket to write data to
   • API_TOKEN: a token with write access to the specified bucket. Store this in a secret store or environment variable to avoid exposing the raw token string.

8. To test the input and processor, enter the following command:

   telegraf --test --config telegraf.conf
   

   Telegraf outputs the following to stdout, and then exits:

   > home,room=Kitchen count=2,temp=23.1 1641063600000000000
   > home,room=Living\ Room count=2,temp=22.7 1641063600000000000