`ignore_unavailable=true` ignored when a single index is specified

[spalger]

We noticed an edge case today when using the ignore_unavailable parameter against indices that are closed. Running the following in Sense produces a 403 error, which was very unexcepted.

curl -XPOST "http://localhost:9200/test_index/test_type" -d'
{
  "some": "value"
}'

curl -XPOST "http://localhost:9200/test/_close"

curl -XPOST "http://localhost:9200/test/_search?ignore_unavailable=true"
# {
#   "error": "ClusterBlockException[blocked by: [FORBIDDEN/4/index closed];]",
#   "status": 403
# }

Alternatively, if I send test* I get the expected empty result.

curl -XPOST "http://localhost:9200/test*/_search?ignore_unavailable=true"
# {
#   "took": 1,
#   "timed_out": false,
#   "_shards": {
#     "total": 0,
#     "successful": 0,
#     "failed": 0
#    },
#    // etc.
# }

[imotov]

A couple more examples of inconsistency:

curl -XPOST "http://localhost:9200/test/_search?ignore_unavailable=false&pretty"
# {
#   "error" : "IndexClosedException[[test] closed]",
#   "status" : 403
# }

curl -XPOST "http://localhost:9200/+test/_search?ignore_unavailable=true&pretty"
# {
#   "took" : 0,
#   "timed_out" : false,
#   "_shards" : {
#     "total" : 0,
#     "successful" : 0,
#     "failed" : 0
#   },
#   "hits" : {
#     "total" : 0,
#     "max_score" : 0.0,
#     "hits" : [ ]
#   }
# }

[javanna]

The different behaviour is caused by the fact that you are referring in the url to a specific index, in that case we ignore the ignore_unavailable option. We do consider it though if you specify multiple indices or wildcard expressions etc. That said, let's discuss if this is right or wrong :)

[clintongormley]

Yeah, it's a good question what the right behaviour here is.

[spalger]

+1 for making the parameter apply to both index names and index patterns.

I think that the current behavior would make more sense if we renamed the parameter ignore_unavailable_pattern_matches

[martijnvg]

I think it make sense that ignore_unavailable also applies if just a single concrete index or alias is specified.

[clintongormley]

Yeah I'm thinking the same thing.

[geoffroya]

+1

[ppf2]

+1 on providing a way for aliases to ignore closed indices, esp. for the time based index use case where older indices may be closed.

[spalger]

Hey @martijnvg, any word on this?