Completion suggester returning non-deterministic results?

When I run a completion suggester I'm getting different results each time I send the same query over the same static data.

I realize that this may be an optimization for search...because users don't always need all the completions. :smiley:

This is on a vintage 7.1.1 with two shards, and I'm always getting success on both shards.

I'm looking for terms that start with "/FontIdentifier->"

I know from the data that I should be getting at least: /FontIdentifier->/Checksum, /FontIdentifier->/Family, /FontIdentifier->/Foundry, /FontIdentifier->/Kind, /FontIdentifier->/Type, /FontIdentifier->/Version

The results I'm seeing, though, generally contain 3 items, normally always /FontIdentifier->/Checksum, but then sometimes /FontIdentifier->/Foundry or /FontIdentifier->/Type or something else.

Each document that contains /FontIdentifier should roughly have all of the above. I'm storing each value as a keyword in a multi-valued field (see below).

This is the query:

GET file-observatory-eval-three-20210806/_search
  "_source": false,
  "suggest": {
    "fontIdentifier": {
      "prefix": "/FontIdentifier->",
      "completion": {
        "field": "q_parent_and_keys.completion",
        "size": 1000000,
        "skip_duplicates": true

This is the config for this field:

      "q_parent_and_keys": {
        "type": "keyword",
        "store": "true",
        "fields": {
          "case": {
            "type": "keyword",
            "normalizer": "lc_normalizer"
          "completion": {
            "type": "completion",
            "analyzer": "whitespace"

Is this a bug or a feature? Is there anything I can do with the schema or how I'm indexing the data to get what I need?

Or, is there an alternative way outside of the suggester to get all terms that match a prefixquery?

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.