Split an event structure that includes an array into multiple distinct events with each array element.

ParameterTypeRequiredDefault ValueDescription
field[a]stringoptional[b]_events Field to split by.
stripbooleanoptional[b]false Strip the field prefix when splitting (default is false).

[a] The argument name field can be omitted.

[b] Optional parameters use their default value unless explicitly set.

Hide omitted argument names for this function

Show omitted argument names for this function

When LogScale ingests data into arrays, each array entry is turned into separate attributes named [0], [1], ... This function takes such an event and splits it into multiple events based on the prefix of such [N] attributes, allowing for aggregate functions across array values.

When the function is called, each split event generated is given a unique index ID in the _index field. This can be used to identify the individual event.

Note

The split() function is not very efficient, so it should only be used after some aggressive filtering.

split() Examples

In GitHub events, a PushEvent contains an array of commits, and each commit gets expanded into subattributes of payload.commit_0, payload.commit_1, .... LogScale cannot sum/count, etc. across such attributes. split() expands each PushEvent into one PushEvent for each commit so they can be counted.

logscale
type=PushEvent
| split(payload.commits)
| groupby(payload.commits.author.email)
| sort()

There might be a case where your parser is receiving JSON events in a JSON array, as in:

JSON
[
  {"exampleField": "value"},
  {"exampleField": "value2"}
]

In this case, your @rawstring text contains this full array, but each record in the array is actually an event in itself, and you would like to split them out.

First you need to call parseJson(), but when @rawstring contains an array, the parseJson() function doesn't assign names to the fields automatically, it only assigns indexes. In other words, calling parseJson() adds fields named something like [0].exampleField, [1].exampleField, etc. to the current event.

Since split() needs a field name to operate on before it reads indexes, it seems like we can't pass it anything here. But we can tell split() to look for the empty field name by calling split(field="").

This means that parsing the above with:

logscale
parseJson()
| split(field="")

will produce two events, each with a field named exampleField, and with an additional field, _index containing the index (count) of the original data so that each individual split() event can be identified:

Alternatively, we can tell parseJson() to add a prefix to all the fields, which can then use as the field name to split on:

logscale
parseJson(prefix="example")
| split(field="example")

Unfortunately this adds the example prefix to all fields on the new event we've split out, so you may prefer splitting on the empty field name to avoid that.