Overview

Datadog automatically parses JSON-formatted logs. For other formats, Datadog allows you to enrich your logs with the help of Grok Parser. The Grok syntax provides an easier way to parse logs than pure regular expressions. The Grok Parser enables you to extract attributes from semi-structured text messages.

Grok comes with reusable patterns to parse integers, IP addresses, hostnames, etc. These values must be sent into the grok parser as strings.

You can write parsing rules with the %{MATCHER:EXTRACT:FILTER} syntax:

• Matcher: A rule (possibly a reference to another token rule) that describes what to expect (number, word, notSpace, etc.).

• Extract (optional): An identifier representing the capture destination for the piece of text matched by the Matcher.

• Filter (optional): A post-processor of the match to transform it.

Example for a classic unstructured log:

john connected on 11/08/2017

With the following parsing rule:

MyParsingRule %{word:user} connected on %{date("MM/dd/yyyy"):date}

After processing, the following structured log is generated:

Note:

• If you have multiple parsing rules in a single Grok parser:
  • Only one can match any given log. The first one that matches, from top to bottom, is the one that does the parsing.
  • Each rule can reference parsing rules defined above itself in the list.
• You must have unique rule names within the same Grok parser.
• The rule name must contain only: alphanumeric characters, _, and .. It must start with an alphanumeric character.
• Properties with null or empty values are not displayed.
• The regex matcher applies an implicit ^, to match the start of a string, and $, to match the end of a string.
• Certain logs can produce large gaps of whitespace. Use \n and \s+ to account for newlines and whitespace.

Matcher and filter

Here is a list of all the matchers and filters natively implemented by Datadog:

Advanced settings

At the bottom of your Grok processor tiles, there is an Advanced Settings section:

Parsing a specific text attribute

Use the Extract from field to apply your Grok processor on a given text attribute instead of the default message attribute.

For example, consider a log containing a command.line attribute that should be parsed as a key-value. You could parse this log as follows:

Using helper rules to factorize multiple parsing rules

Use the Helper Rules field to define tokens for your parsing rules. Helper rules help you to factorize Grok patterns across your parsing rules. This is useful when you have several rules in the same Grok parser that use the same tokens.

Example for a classic unstructured log:

john id:12345 connected on 11/08/2017 on server XYZ in production

Use the following parsing rule:

MyParsingRule %{user} %{connection} %{server}

With the following helpers:

user %{word:user.name} id:%{integer:user.id}
connection connected on %{date("MM/dd/yyyy"):connect_date}
server on server %{notSpace:server.name} in %{notSpace:server.env}

Examples

Some examples demonstrating how to use parsers:

• Key value or logfmt
• Parsing dates
• Alternating patterns
• Optional attribute
• Nested JSON
• Regex
• List and Arrays
• Glog format
• XML
• CSV

Key value or logfmt

This is the key-value core filter: keyvalue([separatorStr[, characterAllowList[, quotingStr[, delimiter]]]]) where:

• separatorStr: defines the separator between key and values. Defaults to =.
• characterAllowList: defines extra non-escaped value chars in addition to the default \\w.\\-_@. Used only for non-quoted values (for example, key=@valueStr).
• quotingStr: defines quotes, replacing the default quotes detection: <>, "", ''.
• delimiter: defines the separator between the different key values pairs (for example, |is the delimiter in key1=value1|key2=value2). Defaults to (normal space), , and ;.

Use filters such as keyvalue to more-easily map strings to attributes for keyvalue or logfmt formats:

Log:

user=john connect_date=11/08/2017 id=123 action=click

Rule:

rule %{data::keyvalue}

You don’t need to specify the name of your parameters as they are already contained in the log. If you add an extract attribute my_attribute in your rule pattern you will see:

If = is not the default separator between your key and values, add a parameter in your parsing rule with a separator.

Log:

user: john connect_date: 11/08/2017 id: 123 action: click

Rule:

rule %{data::keyvalue(": ")}

If logs contain special characters in an attribute value, such as / in a url for instance, add it to the allowlist in the parsing rule:

Log:

url=https://app.datadoghq.com/event/stream user=john

Rule:

rule %{data::keyvalue("=","/:")}

Other examples:

Multiple QuotingString example: When multiple quotingstring are defined, the default behavior is replaced with a defined quoting character. The key-value always matches inputs without any quoting characters, regardless of what is specified in quotingStr. When quoting characters are used, the characterAllowList is ignored as everything between the quoting characters is extracted.

Log:

key1:=valueStr key2:=</valueStr2> key3:="valueStr3"

Rule:

rule %{data::keyvalue(":=","","<>")}

Result:

{"key1": "valueStr", "key2": "/valueStr2"}

Note:

• Empty values (key=) or null values (key=null) are not displayed in the output JSON.
• If you define a keyvalue filter on a data object, and this filter is not matched, then an empty JSON {} is returned (for example, input: key:=valueStr, parsing rule: rule_test %{data::keyvalue("=")}, output: {}).
• Defining "" as quotingStr keeps the default configuration for quoting.

Parsing dates

The date matcher transforms your timestamp in the EPOCH format (unit of measure millisecond).

¹ Use the timezone parameter if you perform your own localizations and your timestamps are not in UTC. The supported format for timezones are:

• GMT, UTC, UT or Z
• +h, +hh, +hh:mm, -hh:mm, +hhmm, -hhmm, +hh:mm:ss, -hh:mm:ss, +hhmmss or -hhmmss . The maximum supported range is from +18:00 to -18:00 inclusive.
• Timezones starting with UTC+, UTC-, GMT+, GMT-, UT+ or UT-. The maximum supported range is from +18:00 to -18:00 inclusive.
• Timezone IDs pulled from the TZ database. For more information, see TZ database names.

Note: Parsing a date doesn’t set its value as the log official date. For this use the Log Date Remapper in a subsequent Processor.

Alternating pattern

If you have logs with two possible formats which differ in only one attribute, set a single rule using alternating with (<REGEX_1>|<REGEX_2>). This rule is equivalent to a Boolean OR.

Log:

john connected on 11/08/2017
12345 connected on 11/08/2017

Rule: Note that “id” is an integer and not a string.

MyParsingRule (%{integer:user.id}|%{word:user.firstname}) connected on %{date("MM/dd/yyyy"):connect_date}

Results:

Optional attribute

Some logs contain values that only appear part of the time. In this case, make attribute extraction optional with ()?.

Log:

john 1234 connected on 11/08/2017

Rule:

MyParsingRule %{word:user.firstname} (%{integer:user.id} )?connected on %{date("MM/dd/yyyy"):connect_date}

Note: A rule will not match if you include a space after the first word in the optional section.

Nested JSON

Use the json filter to parse a JSON object nested after a raw text prefix:

Log:

Sep 06 09:13:38 vagrant program[123]: server.1 {"method":"GET", "status_code":200, "url":"https://app.datadoghq.com/logs/pipelines", "duration":123456}

Rule:

parsing_rule %{date("MMM dd HH:mm:ss"):timestamp} %{word:vm} %{word:app}\[%{number:logger.thread_id}\]: %{notSpace:server} %{data::json}

Regex

Log:

john_1a2b3c4 connected on 11/08/2017

Rule:

MyParsingRule %{regex("[a-z]*"):user.firstname}_%{regex("[a-zA-Z0-9]*"):user.id} .*

List to array

Use the array([[openCloseStr, ] separator][, subRuleOrFilter) filter to extract a list into an array in a single attribute. The subRuleOrFilter is optional and accepts these filters.

Log:

Users [John, Oliver, Marc, Tom] have been added to the database

Rule:

myParsingRule Users %{data:users:array("[]",",")} have been added to the database

Log:

Users {John-Oliver-Marc-Tom} have been added to the database

Rule:

myParsingRule Users %{data:users:array("{}","-")} have been added to the database

Rule using subRuleOrFilter:

myParsingRule Users %{data:users:array("{}","-", uppercase)} have been added to the database

Glog format

Kubernetes components sometimes log in the glog format; this example is from the Kube Scheduler item in the Pipeline Library.

Example log line:

W0424 11:47:41.605188       1 authorization.go:47] Authorization is disabled

Parsing rule:

kube_scheduler %{regex("\\w"):level}%{date("MMdd HH:mm:ss.SSSSSS"):timestamp}\s+%{number:logger.thread_id} %{notSpace:logger.name}:%{number:logger.lineno}\] %{data:msg}

And extracted JSON:

{
  "level": "W",
  "timestamp": 1587728861605,
  "logger": {
    "thread_id": 1,
    "name": "authorization.go"
  },
  "lineno": 47,
  "msg": "Authorization is disabled"
}

Parsing XML

The XML parser transforms XML formatted messages into JSON.

Log:

<book category="CHILDREN">
  <title lang="en">Harry Potter</title>
  <author>J K. Rowling</author>
  <year>2005</year>
</book>

Rule:

rule %{data::xml}

Result:

{
"book": {
  "year": "2005",
  "author": "J K. Rowling",
  "category": "CHILDREN",
  "title": {
    "lang": "en",
    "value": "Harry Potter"
  }
}
}

Notes:

• If the XML contains tags that have both an attribute and a string value between the two tags, a value attribute is generated. For example: <title lang="en">Harry Potter</title> is converted to {"title": {"lang": "en", "value": "Harry Potter" } }
• Repeated tags are automatically converted to arrays. For example: <bookstore><book>Harry Potter</book><book>Everyday Italian</book></bookstore> is converted to { "bookstore": { "book": [ "Harry Potter", "Everyday Italian" ] } }

Parsing CSV

Use the CSV filter to more-easily map strings to attributes when separated by a given character (, by default).

The CSV filter is defined as csv(headers[, separator[, quotingcharacter]]) where:

• headers: Defines the keys name separated by ,. Keys names must start with alphabetical character and can contain any alphanumerical character in addition to _.
• separator: Defines separators used to separate the different values. Only one character is accepted. Default: ,. Note: Use tab for the separator to represent the tabulation character for TSVs.
• quotingcharacter: Defines the quoting character. Only one character is accepted. Default: "

Note:

• Values containing a separator character must be quoted.
• Quoted Values containing a quoting character must be escaped with a quoting characters. For example, "" within a quoted value represents ".
• If the log doesn’t contain the same number of value as the number of keys in the header, the CSV parser will match the first ones.
• Intergers and Double are automatically casted if possible.

Log:

John,Doe,120,Jefferson St.,Riverside

Rule:

myParsingRule %{data:user:csv("first_name,name,st_nb,st_name,city")}

Result:

{
  "user": {
    "first_name": "John",
    "name": "Doe",
    "st_nb": 120,
    "st_name": "Jefferson St.",
    "city": "Riverside"
  }
}

Other examples:

Use data matcher to discard unneeded text

If you have a log where after you have parsed what is needed and know that the text after that point is safe to discard, you can use the data matcher to do so. For the following log example, you can use the data matcher to discard the % at the end.

Log:

Usage: 24.3%

Rule:

MyParsingRule Usage\:\s+%{number:usage}%{data:ignore}

Result:

{
  "usage": 24.3,
  "ignore": "%"
}

ASCII control characters

If your logs contain ASCII control characters, they are serialized upon ingestion. These can be handled by explicitly escaping the serialized value within your grok parser.

Further Reading

추가 유용한 문서, 링크 및 기사:

Learn how to process your logsDOCUMENTATION How to investigate a log parsing issue?FAQ Log Parsing - Best PracticesFAQ Control the volume of logs indexed by DatadogDOCUMENTATION