Skip to main content
Sumo Logic

Parse JSON Formatted Logs

The JSON operator is a search operator that allows you to extract values from JSON input.

Because JSON supports both nested keys and arrays that contain ordered sequences of values, the Sumo Logic JSON operator allows you to extract:

  • Single, top-level fields.
  • Multiple fields.
  • Nested keys.
  • Keys in arrays.

The JSON operator also supports the nodrop option, which allows messages containing invalid JSON values to be displayed.

The following examples use this sample log message:

2014-03-11 15:00:42,611 -0700 INFO [hostId=prod-search-6] 
[] {"module":"stream","logMessage":"exiting search","sessionId":"90D97000","customerId": "00B12CD0"
    "version": "1"

The JSON operator uses a subset of the JSONPath format, described here:

The following table compares the JSONPath syntax elements with XPath.

JSONPath Description
$ The root object or element.
@ The current object or element.
. or [] Child operator.
n/a Parent operator.
.. Recursive descent. JSONPath borrows this syntax from E4X.
* Wildcard. All objects or elements regardless of their names.
n/a Attribute access. JSON structures don't have attributes.
[] Subscript operator. XPath uses it to iterate over element collections and for predicates. In JavaScript and JSON, it is the native array operator.
[,] Union operator in XPath results in a combination of node sets. JSONPath allows alternate names or array indices as a set.
[start:end:step] Array slice operator borrowed from ES4.
?() Applies a filter (script) expression.
() Script expression, using the underlying script engine.
n/a Grouping in XPath.

Extracting a single top-level field

The JSON operator allows you to extract a single, top-level field. For example, to extract sessionID:

_sourceCategory=stream RawOutputProcessor "\"message\""
| parse "]*" as jsonobject 
| json field=jsonobject "sessionId"
| fields -jsonobject

produces results like:

Extracting multiple fields

You can also extract multiple fields in a single operation. For example, to extract sessionID and customerID:

_sourceCategory=stream RawOutputProcessor "\"message\""
| parse "]*" as jsonobject
| json field=jsonobject "sessionId", "customerId"
| fields -jsonobject

produces these results:

In addition, you can assign names to fields that differ from their assigned names. To use sd instead ofsessionId and cid instead of customerId, like this:

_sourceCategory=stream RawOutputProcessor "\"message\"" 
| parse "]*" as jsonobject 
| json field=jsonobject "sessionId", "customerId" as sID, cID 
| fields -jsonobject

which gives you these results:

Extracting a nested key

The example log message has nested keys, which you can extract by specifying the path using dot notation:

For example, to extract the nested key type from meta, use the following query:

* | json field=jsonobject "meta.type"

Finding values in a JSON array

In some cases, fields values are actually arrays, like baselineIntervals in the example log message:

You can instruct the JSON operator to extract @baselineIntervals, like this:

* | json field=jsonobject "baselineIntervals"

It returns a list of the values in the array: ["2014-03-10T23:...", ""2014-03-11T05:..."].

like this:

To refer to one specific entry in the array, provide the array's index: 

* | json field=jsonobject "baselineIntervals[1]"

Using the nodrop option

By default, the JSON operator optimizes results by dropping messages that don't use the specified key or keys, or messages that use invalid JSON keys. Use the nodrop option to prevent this optimization, and set the extracted field values to null (empty):

* | json field=jsonobject "baselineIntervals[0]" nodrop

Using wildcard (*)

You can use wildcard (*) to access the array elements in a JSON.
For example, you can access Actor Type from the O365 JSON message using wildcard.


| json "Actor[*].Type" as Actortype

Result of the query would look like this:


Next, if required you can use these array elements to perform additional operations. For example, you can find the max of Type for a CreationTime and Id using this query:

| json field=_raw "CreationTime", "Id"
| json "Actor[*].Type" as Actortype
| extract field=ActorType"(?<Type>\d+)" multi
| max(type) by CreationTime, Id

The result would look like this:


JSON auto option

Use the json auto option in a query to automatically detect JSON objects in logs and extract the key/value pairs without the need to specify fields in a parse statement. After the query runs, you can use the Field Browser to choose the fields you’d like to display. You can also operate on the extracted fields later in the query.

If you don’t specify any additional parameters, the JSON objects are automatically detected and all of the key/value pairs are extracted. Note that messages that do not contain JSON are not dropped.

The JSON portion does not need to span the entire log message. There can be some text before and after the JSON portion. The json auto operator automatically detects where the JSON object is located and parses it.

For example, in this log message:

2015-05-04 21:51:43,289 -0700 INFO ["hostId"=stream] ["module"=stream] {"foo":{"bar":"baz"}} ...

The operator automatically detects the JSON object:


Its important to note that json auto works by searching for json blobs beginning at the end of the message. Most logs begin with a preamble, such as a timestamp, then the json blob. In cases where content appears at the end of the message after the json blob, the extraction could fail.

Having the json blob at the end of the message is recommended. Having it in the middle of the message could cause extraction failure.


 … | json auto


"{" | json auto

Additional options

* | json auto field=fieldname

Operates on a specified field. By default, json auto will attempt to extract JSON fields from the raw log message field. To have it operate on a different field, use the field option.


* | json auto field=myfield

* | json auto keys

References specific keys in json. The keys can be renamed (aliased) using as. Example:

* | json auto keys "key1", "key2" as alias1, alias2

Use the the refonly option to extract only the referenced keys. If you don't use this option, json auto will also extract all other JSON fields it fields in the message.


* | json auto keys "key1", "key2" refonly

* | json auto maxdepth

JSON is a hierarchical data structure that can have many levels of objects and arrays. For example, the following has a depth of four levels:

   "foo": {
      "bar": [
            "k1": "v1"
            "k2": "v2"
      "baz": "qux"

At depth 1 is the object containing foo. At depth 2 is the object containing bar and baz. At depth 3 is the array containing two objects. At depth 4 are two objects containing keys k1 and k2.

Use the maxdepth to specify the level at which to flatten the JSON.

The following examples show how the previous sample changes when maxdepth values are applied:

json auto maxdepth 1:

field: foo value: {"bar": [{"k1": "v1"}, {"k2", "v2"}], "baz": "qux"}

json auto maxdepth 2:

field: value: [{"k1": "v1"}, {"k2", "v2"}]

field: foo.baz value: qux


* | json auto maxdepth 2

* | json auto extractarrays

Extracts elements from flat arrays with each element of the array as a separate key value pair. Non-flat arrays (arrays containing other JSON objects or arrays) are extracted by default.

Consider this example:

{"foo": [1,2,3]}

Without the extractarrays option, json auto yields

field: foo value: [1,2,3]

With the extractarrays option, json auto yields these field-value pairs:

field: foo[0] value: 1field: foo[1] value: 2field: foo[2] value: 3

Other important notes:

1. Fields extracted using json auto need not be referenced explicitly in order to be used later in the query. For example, the user does not need to do this:

* | json auto keys "username" | count by username

The user can simply write:

* | json auto | count by username

We will infer that username must be a field extracted using json auto and treat it as such.

2. The keys extracted from json may contain dots and square brackets. Hierarchical entities within json are denoted using dot as the separator between different levels. Array indices are specified in between brackets. An example extracted field name can be:


Using this field name later in the query fails since dots and brackets are not normally allowed in the field name. To use these fields, you may escape the field using the percent sign (%).

For example, this will not work:

* | json auto | count by users[2].address.street

But this will:

* | json auto | count by %users[2].address.street