[BUG] Inconsistent field collapsing search results in cases of concurrent segment search enabled and disabled

[gaobinlong]

Describe the bug

Inconsistent field collapsing search results returned by cases of concurrent segment search enabled and disabled, and seems the result of the former case is not correct.

Related component

Search

To Reproduce

1. Create index

curl -X PUT localhost:9201/abcd_test -H 'Content-Type: application/json' -d '{"settings":{"refresh_interval":"60s"}}' 

2. Index 3 documents into 3 segments

curl -XPOST "http://localhost:9201/abcd_test/_doc?refresh" -H 'Content-Type: application/json' -d'
{
  "a":1,
"c":1,
"docIndex":1
}'

curl -XPOST "http://localhost:9201/abcd_test/_doc?refresh" -H 'Content-Type: application/json' -d'
{
  "a":2,
"c":1,
"docIndex":2
}'

curl -XPOST "http://localhost:9201/abcd_test/_doc?refresh" -H 'Content-Type: application/json' -d'
{
  "a":2,
"c":1,
"docIndex":3
}'

3. Execute search

curl -XGET "http://localhost:9201/abcd_test/_search?size=2&request_cache=false" -H 'Content-Type: application/json' -d'
{
  "collapse": {
    "field": "a"
  },
"sort":[{
  "c":"desc"
}
]
}'

the result is

{"took":4027,"timed_out":false,"_shards":{"total":1,"successful":1,"skipped":0,"failed":0},"hits":{"total":{"value":3,"relation":"eq"},"max_score":null,"hits":[{"_index":"abcd_test","_id":"BymuwZgBMprIY4AAc-T_","_score":null,"_source":
{
  "a":1,
"c":1,
"docIndex":1
},"fields":{"a":[1]},"sort":[1]},{"_index":"abcd_test","_id":"CCmuwZgBMprIY4AAdOSV","_score":null,"_source":
{
  "a":2,
"c":1,
"docIndex":2
},"fields":{"a":[2]},"sort":[1]}]}}

4. Enable concurrent segment search

curl -XPUT "http://localhost:9201/abcd_test/_settings" -H 'Content-Type: application/json' -d'
{
    "index.search.concurrent_segment_search.mode": "all"
}'

5. Execute the same query, check the result

{"took":147,"timed_out":false,"_shards":{"total":1,"successful":1,"skipped":0,"failed":0},"hits":{"total":{"value":3,"relation":"eq"},"max_score":null,"hits":[{"_index":"abcd_test","_id":"7qXpxpgBgVh_OsnfGh2r","_score":null,"_source":
{
  "a":1,
"c":1,
"docIndex":1
},"fields":{"a":[1]},"sort":[1]},{"_index":"abcd_test","_id":"8KXpxpgBgVh_OsnfGx21","_score":null,"_source":
{
  "a":2,
"c":1,
"docIndex":3
},"fields":{"a":[2]},"sort":[1]}]}}

, then you can see that the second doc is different from the above one.

I think the doc with docIndex 2 should be returned because this doc has small lucene doc id.

Expected behavior

When concurrent segment search enabled and executing field collapsing search, the result should be same to the result of the same query when concurrent segment search is disabled.

Additional Details

Host/Environment (please complete the following information):

• OS: [e.g. iOS]
• Version [Main branch and 3.2]

[sandeshkr419]

[Search Triage] @jed326 @expani Do you guys have some context here?

I'm wondering if consistent results are even expected in the 1st place here?