Skip to main content
Version: 0.14

Writing Logs Using a Pipeline

This document describes how to write logs to GreptimeDB by processing them through a specified pipeline using the HTTP interface.

Before writing logs, please read the Pipeline Configuration and Managing Pipelines documents to complete the configuration setup and upload.

HTTP API​

You can use the following command to write logs via the HTTP interface:

curl -X "POST" "http://localhost:4000/v1/events/logs?db=<db-name>&table=<table-name>&pipeline_name=<pipeline-name>&version=<pipeline-version>" \
     -H 'Content-Type: application/x-ndjson' \
     -d "$<log-items>"

Request parameters​

This interface accepts the following parameters:

  • db: The name of the database.
  • table: The name of the table.
  • pipeline_name: The name of the pipeline.
  • version: The version of the pipeline. Optional, default use the latest one.

Content-Type and body format​

GreptimeDB uses Content-Type header to decide how to decode the payload body. Currently the following two format is supported:

  • application/json: this includes normal JSON format and NDJSON format.
  • application/x-ndjson: specifically uses NDJSON format, which will try to split lines and parse for more accurate error checking.
  • text/plain: multiple log lines separated by line breaks.

application/json and application/x-ndjson format​

Here is an example of JSON format body payload

[
     {"message":"127.0.0.1 - - [25/May/2024:20:16:37 +0000] \"GET /index.html HTTP/1.1\" 200 612 \"-\" \"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36\""},
     {"message":"192.168.1.1 - - [25/May/2024:20:17:37 +0000] \"POST /api/login HTTP/1.1\" 200 1784 \"-\" \"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/88.0.4324.96 Safari/537.36\""},
     {"message":"10.0.0.1 - - [25/May/2024:20:18:37 +0000] \"GET /images/logo.png HTTP/1.1\" 304 0 \"-\" \"Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:89.0) Gecko/20100101 Firefox/89.0\""},
     {"message":"172.16.0.1 - - [25/May/2024:20:19:37 +0000] \"GET /contact HTTP/1.1\" 404 162 \"-\" \"Mozilla/5.0 (iPhone; CPU iPhone OS 14_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/14.0 Mobile/15E148 Safari/604.1\""}
]

Note the whole JSON is an array (log lines). Each JSON object represents one line to be processed by Pipeline engine.

The name of the key in JSON objects, which is message here, is used as field name in Pipeline processors. For example:

processors:
  - dissect:
      fields:
        # `message` is the key in JSON object
        - message
      patterns:
        - '%{ip_address} - - [%{timestamp}] "%{http_method} %{request_line}" %{status_code} %{response_size} "-" "%{user_agent}"'
      ignore_missing: true

# rest of the file is ignored

We can also rewrite the payload into NDJSON format like following:

{"message":"127.0.0.1 - - [25/May/2024:20:16:37 +0000] \"GET /index.html HTTP/1.1\" 200 612 \"-\" \"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36\""}
{"message":"192.168.1.1 - - [25/May/2024:20:17:37 +0000] \"POST /api/login HTTP/1.1\" 200 1784 \"-\" \"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/88.0.4324.96 Safari/537.36\""}
{"message":"10.0.0.1 - - [25/May/2024:20:18:37 +0000] \"GET /images/logo.png HTTP/1.1\" 304 0 \"-\" \"Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:89.0) Gecko/20100101 Firefox/89.0\""}
{"message":"172.16.0.1 - - [25/May/2024:20:19:37 +0000] \"GET /contact HTTP/1.1\" 404 162 \"-\" \"Mozilla/5.0 (iPhone; CPU iPhone OS 14_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/14.0 Mobile/15E148 Safari/604.1\""}

Note the outer array is eliminated, and lines are separated by line breaks instead of ,.

text/plain format​

Log in plain text format is widely used throughout the ecosystem. GreptimeDB also supports text/plain format as log data input, enabling ingesting logs first hand from log producers.

The equivalent body payload of previous example is like following:

127.0.0.1 - - [25/May/2024:20:16:37 +0000] "GET /index.html HTTP/1.1" 200 612 "-" "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
192.168.1.1 - - [25/May/2024:20:17:37 +0000] "POST /api/login HTTP/1.1" 200 1784 "-" "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/88.0.4324.96 Safari/537.36"
10.0.0.1 - - [25/May/2024:20:18:37 +0000] "GET /images/logo.png HTTP/1.1" 304 0 "-" "Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:89.0) Gecko/20100101 Firefox/89.0"
172.16.0.1 - - [25/May/2024:20:19:37 +0000] "GET /contact HTTP/1.1" 404 162 "-" "Mozilla/5.0 (iPhone; CPU iPhone OS 14_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/14.0 Mobile/15E148 Safari/604.1"

Sending log ingestion request to GreptimeDB requires only modifying the Content-Type header to be text/plain, and you are good to go!

Please note that, unlike JSON format, where the input data already have key names as field names to be used in Pipeline processors, text/plain format just gives the whole line as input to the Pipeline engine. In this case we use line as the field name to refer to the input line, for example:

processors:
  - dissect:
      fields:
        # use `line` as the field name
        - line
      patterns:
        - '%{ip_address} - - [%{timestamp}] "%{http_method} %{request_line}" %{status_code} %{response_size} "-" "%{user_agent}"'
      ignore_missing: true

# rest of the file is ignored

It is recommended to use dissect or regex processor to split the input line into fields first and then process the fields accordingly.

Built-in Pipelines​

GreptimeDB offers built-in pipelines for common log formats, allowing you to use them directly without creating new pipelines.

Note that the built-in pipelines are not editable. Additionally, the "greptime_" prefix of the pipeline name is reserved.

greptime_identity​

The greptime_identity pipeline is designed for writing JSON logs and automatically creates columns for each field in the JSON log.

  • The first-level keys in the JSON log are used as column names.
  • An error is returned if the same field has different types.
  • Fields with null values are ignored.
  • If time index is not specified, an additional column, greptime_timestamp, is added to the table as the time index to indicate when the log was written.

Type conversion rules​

  • string -> string
  • number -> int64 or float64
  • boolean -> bool
  • null -> ignore
  • array -> json
  • object -> json

For example, if we have the following json data:

[
    {"name": "Alice", "age": 20, "is_student": true, "score": 90.5,"object": {"a":1,"b":2}},
    {"age": 21, "is_student": false, "score": 85.5, "company": "A" ,"whatever": null},
    {"name": "Charlie", "age": 22, "is_student": true, "score": 95.5,"array":[1,2,3]}
]

We'll merge the schema for each row of this batch to get the final schema. The table schema will be:

mysql> desc pipeline_logs;
+--------------------+---------------------+------+------+---------+---------------+
| Column             | Type                | Key  | Null | Default | Semantic Type |
+--------------------+---------------------+------+------+---------+---------------+
| age                | Int64               |      | YES  |         | FIELD         |
| is_student         | Boolean             |      | YES  |         | FIELD         |
| name               | String              |      | YES  |         | FIELD         |
| object             | Json                |      | YES  |         | FIELD         |
| score              | Float64             |      | YES  |         | FIELD         |
| company            | String              |      | YES  |         | FIELD         |
| array              | Json                |      | YES  |         | FIELD         |
| greptime_timestamp | TimestampNanosecond | PRI  | NO   |         | TIMESTAMP     |
+--------------------+---------------------+------+------+---------+---------------+
8 rows in set (0.00 sec)

The data will be stored in the table as follows:

mysql> select * from pipeline_logs;
+------+------------+---------+---------------+-------+---------+---------+----------------------------+
| age  | is_student | name    | object        | score | company | array   | greptime_timestamp         |
+------+------------+---------+---------------+-------+---------+---------+----------------------------+
|   22 |          1 | Charlie | NULL          |  95.5 | NULL    | [1,2,3] | 2024-10-18 09:35:48.333020 |
|   21 |          0 | NULL    | NULL          |  85.5 | A       | NULL    | 2024-10-18 09:35:48.333020 |
|   20 |          1 | Alice   | {"a":1,"b":2} |  90.5 | NULL    | NULL    | 2024-10-18 09:35:48.333020 |
+------+------------+---------+---------------+-------+---------+---------+----------------------------+
3 rows in set (0.01 sec)

Specify time index​

A time index is necessary in GreptimeDB. Since the greptime_identity pipeline does not require a YAML configuration, you must set the time index in the query parameters if you want to use the timestamp from the log data instead of the automatically generated timestamp when the data arrives.

Example of Incoming Log Data:

[
    {"action": "login", "ts": 1742814853}
]

To instruct the server to use ts as the time index, set the following query parameter in the HTTP header:

curl -X "POST" "http://localhost:4000/v1/events/logs?db=public&table=pipeline_logs&pipeline_name=greptime_identity&custom_time_index=ts;epoch;s" \
     -H 'Content-Type: application/json' \
     -d $'[{"action": "login", "ts": 1742814853}]'

The custom_time_index parameter accepts two formats, depending on the input data format:

  • Epoch number format: <field_name>;epoch;<resolution>
    • The field can be an integer or a string.
    • The resolution must be one of: s, ms, us, or ns.
  • Date string format: <field_name>;datestr;<format>
    • For example, if the input data contains a timestamp like 2025-03-24 19:31:37+08:00, the corresponding format should be %Y-%m-%d %H:%M:%S%:z.

With the configuration above, the resulting table will correctly use the specified log data field as the time index.

DESC pipeline_logs;
+--------+-----------------+------+------+---------+---------------+
| Column | Type            | Key  | Null | Default | Semantic Type |
+--------+-----------------+------+------+---------+---------------+
| ts     | TimestampSecond | PRI  | NO   |         | TIMESTAMP     |
| action | String          |      | YES  |         | FIELD         |
+--------+-----------------+------+------+---------+---------------+
2 rows in set (0.02 sec)

Examples​

Please refer to the "Writing Logs" section in the Quick Start guide for examples.

Append Only​

By default, logs table created by HTTP ingestion API are in append only mode.