Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

[BUG] Simple `top` queries on high-cardinality fields are very slow

Aperta
  • Qualcosa non va?
#5,673 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
52/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Tranquilla
Stack tecnologico
java, sql
Ambito
backend-api-design, performance

Direzione di ricerca

The reported entry point is a PPL top query; start with the ppl --explain logical and physical plans and their aggregation pushdown, especially the generated composite_buckets request. Compare it with the shown terms/top_values request; done means trivial top queries avoid the coordinator scan and return the expected top results.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

bug calcite performance PPL pushdown

What is the bug?
Simple top queries can take more than 30s if done on fields with high cardinality (>1mil distinct values). This happens even if there's a small top limit such as the default top 10.

How can one reproduce the bug?
Steps to reproduce the behavior:

  1. Index data with lots of distinct request_id values (>1mil, keyword-typed)
  2. source=idx | top request_id
  3. Measure

Explain plan:

> echo 'source=logs | top request_id' | ppl --explain
= Calcite Plan =
== Logical ==
LogicalSystemLimit(fetch=[10000], type=[QUERY_SIZE_LIMIT])
  LogicalProject(request_id=[$0], count=[$1])
    LogicalFilter(condition=[=($2, 10)])
      LogicalProject(request_id=[$0], count=[$1], _row_number_rare_top_=[ROW_NUMBER() OVER (ORDER BY $1 DESC, $0)])
        LogicalAggregate(group=[{0}], count=[COUNT()])
          LogicalProject(request_id=[$60])
            CalciteLogicalIndexScan(table=[[OpenSearch, logs]])


== Physical ==
EnumerableLimit(fetch=[10000])
  EnumerableCalc(expr#0..2=[{inputs}], expr#3=[10], expr#4=[=($t2, $t3)], proj#0..1=[{exprs}], $condition=[$t4])
    EnumerableWindow(window#0=[window(order by [1 DESC, 0] rows between UNBOUNDED PRECEDING and CURRENT ROW aggs [ROW_NUMBER()])])
      CalciteEnumerableIndexScan(table=[[OpenSearch, logs]], PushDownContext=[[AGGREGATION-rel#50:LogicalAggregate.NONE.[](input=RelSubset#49,group={0},count=COUNT())], OpenSearchRequestBuilder(sourceBuilder={
        "aggregations": {
          "composite_buckets": {
            "composite": {
              "size": 10000,
              "sources": [
                {
                  "request_id": {
                    "terms": {
                      "field": "request_id",
                      "missing_bucket": true,
                      "missing_order": "first",
                      "order": "asc"
                    }
                  }
                }
              ]
            }
          }
        },
        "from": 0,
        "size": 0,
        "timeout": "1m"
      }, requestedTotalSize=2147483647, pageSize=null, startFrom=0)])

The inner DSL in this case takes ~300ms (index cache disabled), while the PPL takes ~43800ms on my machine. However, it's also ordering keys lexically instead of by count using OpenSearch's own top_values aggregation.

What is the expected behavior?

At least this case of a trivial top aggregation (without trailing consuming commands) should be fast. OpenSearch itself can do this quickly (133 ms) with straightforward DSL, including returning an accurate size for the "other" bucket (sum_other_doc_count):

{
  "size": 0,
  "aggs": {
    "top_values": {
      "terms": {
        "field": "request_id",
        "size": 10
      }
    }
  }
}

{
    "took": 99,
    "timed_out": false,
    "terminated_early": true,
    "_shards": {
        "total": 1,
        "successful": 1,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 10000,
            "relation": "gte"
        },
        "max_score": null,
        "hits": []
    },
    "aggregations": {
        "top_values": {
            "doc_count_error_upper_bound": 4,
            "sum_other_doc_count": 999990,
            "buckets": [
                {
                    "key": "00000067-142d-4d90-8a2c-28409600adbd",
                    "doc_count": 1
                },
                {
                    "key": "00001910-7d05-4e6a-9716-2c74137475e6",
                    "doc_count": 1
                },
                {
                    "key": "00002676-4f8c-428a-a29a-b862ca672cb7",
                    "doc_count": 1
                },
                {
                    "key": "00003943-8463-46f2-a92b-641b4afbafb1",
                    "doc_count": 1
                },
                {
                    "key": "00003f95-b049-41de-b9be-03aae071ad99",
                    "doc_count": 1
                },
                {
                    "key": "00005502-587d-4b00-865c-86f2af5502c2",
                    "doc_count": 1
                },
                {
                    "key": "00005754-ad22-4894-a9b5-dba3007ff19a",
                    "doc_count": 1
                },
                {
                    "key": "00005c50-97d3-4c11-ad74-90a3219ca815",
                    "doc_count": 1
                },
                {
                    "key": "000066c7-0301-45aa-98b1-d4997a590214",
                    "doc_count": 1
                },
                {
                    "key": "00007501-890e-4f0b-a688-2653ade97f8b",
                    "doc_count": 1
                }
            ]
        }
    }
}

What is your host/environment?

  • mainline, also tested 3.5
  • linux (AL2023), jdk 25

Do you have any screenshots?
N/A

Do you have any additional context?
Seems to be that we don't have any top_values aggregation pushdown at all, so this query turns into a coordinator scan.

Lingua principale
Java
Stelle
175
Fork
232
Merge medio
2g 20h
PR unite (30g)
44

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di opensearch-project/sql

Tutte le issue di opensearch-project/sql

Issue simili

Altre issue su Java

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.