Description
The Elasticsearch team is developing a new, faster API for autocomplete.
We need to integrate this new API into the value suggestions route (/api/kibana/suggestions/values/${index}).
API format
Courtesy of @markharwood
Request
{
"field": "obj.foo",
"string": "foo",
"case_insensitive": true,
"size": 100
}
Response
{
"_shards": {
"total": 1,
"successful": 1,
"failed": 0
},
"terms": [
"foo"
]
}
Errors will be returned in standard Elasticsearch format (root cause and all)
Case sensitivity
The new API will support both case sensitive and insensitive searches, resolving #97797
@shahzad31
Filter editor
While we're used to thinking about autocomplete suggestions in the context of the search bar's input, the same API is also used in the filter editor to provide a list of values for each field.
We could improve the performance of this component by using the new API.
@jimczi also suggested implementing pagination of the autocomplete results (which is easy to implement in ES) to improve performance even more.
Incomplete suggestions indicator
We need to consider how to let a user know, that the autocomplete results they are seeing are incomplete. This is relevant for both the search bar (KQL input area) and the field editor.
Telemetry
In #91428 we introduced telemetry to measure the quality of our autocomplete suggestions:
- how many characters did the user type before choosing a suggestion
- what was the index of the selected suggestion in the list.
This telemetry should not change so we can compare it to telemetry from 7.12 (the old, slower, implementation).