Drill has a collection of JSON configuration options to allow you to configure how Drill interprets JSON files. These are set at the global level, however the HTTP plugin allows you to configure these options individually per connection and override the Drill defaults. The options are:
allowNanInf
: Configures the connection to interpret NaN
and Inf
valuesallTextMode
: By default, Drill attempts to infer data types from JSON data. If the data is malformed, Drill may throw schema change exceptions. If your data is inconsistent, you can enable allTextMode
which when true, Drill will read all JSON values as strings, rather than try to infer the data type.readNumbersAsDouble
: By default Drill will attempt to interpret integers, floating point number types and strings. One challenge is when data is consistent, Drill may throw schema change exceptions. In addition to allTextMode
, you can make Drill less sensitive by setting the readNumbersAsDouble
to true
which causes Drill to read all numeric fields in JSON data as double
data type rather than trying to distinguish between ints and doubles.enableEscapeAnyChar
: Allows a user to escape any character with a \skipMalformedRecords
: Allows Drill to skip malformed records and recover without throwing exceptions.skipMalformedDocument
: Allows Drill to skip entire malformed documents without throwing errors.All of these can be set by adding the jsonOptions
to your connection configuration as shown below:
"jsonOptions": { "allTextMode": true, "readNumbersAsDouble": true }
One of the challenges of querying APIs is inconsistent data. Drill allows you to provide a schema for individual endpoints. You can do this in one of three ways:
The schema provisioning currently supports complex types of Arrays and Maps at any nesting level.
"jsonOptions": { "schema": { "columns":[ { "name":"outer_map", "type":"ARRAY<STRUCT<`bigint_col` BIGINT, `boolean_col` BOOLEAN, `date_col` DATE, `double_col` DOUBLE, `interval_col` INTERVAL, `int_col` BIGINT, `timestamp_col` TIMESTAMP, `time_col` TIME, `varchar_col` VARCHAR>>","mode":"REPEATED" }, { "name":"field1", "type":"VARCHAR", "mode":"OPTIONAL" }, ] } }
One of the major challenges of interacting with JSON data is when the schema is inconsistent. Drill has a UNION
data type which is marked as experimental. At the time of writing, the HTTP plugin does not support the UNION
, however supplying a schema can solve a lot of those issues.
Drill offers the option of reading all JSON values as a string. While this can complicate downstream analytics, it can also be a more memory-efficient way of reading data with inconsistent schema. Unfortunately, at the time of writing, JSON-mode is only available with a provided schema. However, future work will allow this mode to be enabled for any JSON data.
You can enable JSON mode simply by adding the drill.json-mode
property with a value of json
to a field, as shown below:
"jsonOptions": { "readNumbersAsDouble": true, "schema": { "type": "tuple_schema", "columns": [ { "name": "custom_fields", "type": "ARRAY<STRUCT<`value` VARCHAR PROPERTIES { 'drill.json-mode' = 'json' }>>", "mode": "REPEATED" } ] } }