Help with aggregation performance improvement for nested aggregations

[frankkoornstra (Frank)]

I'm trying to find out what the underlying problem for some of my searches' and aggregations' performance is but I can't seem to find any documentation on what the terms in the profiling breakdown mean and what I can do to improve performance. Can you point me somewhere where I can find out what the following things mean (and maybe how to improve on them):

For search:

• score
• build_scorer_count
• match_count
• create_weight
• next_doc
• match
• create_weight_count
• next_doc_count
• score_count
• build_scorer
• advance
• advance_count

For aggregations:

• reduce
• build_aggregation
• build_aggregation_count
• initialize
• initialize_count
• reduce_count
• collect
• collect_count

[Update]
Aaaaaand of course, now I find https://www.elastic.co/guide/en/elasticsearch/reference/current/search-profile.html

Ok, now I can at least be more specific because it doesn't really help me that much for my specific problem

Any idea why the collect phase of this aggregator takes so long? Also, is this CPU/Memory/IO bound?

{
  "type": "NestedAggregator",
  "description": "customVariables",
  "time_in_nanos": 10816958154,
  "breakdown": {
    "reduce": 0,
    "build_aggregation": 36583,
    "build_aggregation_count": 1,
    "initialize": 15959,
    "initialize_count": 1,
    "reduce_count": 0,
    "collect": 10816896702,
    "collect_count": 8908
  },
  "children": [
    {
      "type": "GlobalOrdinalsStringTermsAggregator",
      "description": "context",
      "time_in_nanos": 10410704736,
      "breakdown": {
        "reduce": 0,
        "build_aggregation": 33198,
        "build_aggregation_count": 1,
        "initialize": 2215,
        "initialize_count": 1,
        "reduce_count": 0,
        "collect": 10410563544,
        "collect_count": 105777
      }
    }
  ]
}

With this query:

{
  "profile": true,
  "size": 0,
  "aggregations": {
    "customVariables": {
      "nested": {
        "path": "customer_input.context"
      },
      "aggregations": {
        "context": {
          "terms": {
            "field": "customer_input.context.key",
            "size": 100
          }
        }
      }
    }
  }
}

And this mapping:

{
  "properties": {
    "customer_input": {
      "dynamic": "strict",
      "properties": {
        "context": {
          "type": "nested",
          "dynamic": "strict",
          "properties": {
            "key": {
              "type": "keyword",
              "eager_global_ordinals": true,
              "ignore_above": 200
            }
          }
        }
      }
    }
  }
}

It's usually just a single shard that is slow.