Failed to parse negative number(i. g. -1) with format [epoch_millis]

[jingshouyan]

Elasticsearch version (bin/elasticsearch --version):
7.14.1 // docker image: elasticsearch:7.14.1
Plugins installed: []
analysis-smartcn
https://github.com/medcl/elasticsearch-analysis-pinyin/releases/download/v7.14.1/elasticsearch-analysis-pinyin-7.14.1.zip
https://github.com/medcl/elasticsearch-analysis-ik/releases/download/v7.14.1/elasticsearch-analysis-ik-7.14.1.zip

to reproduce the issue:

PUT /test1
{
  "mappings": {
    "properties": {
      "date": {
        "type": "date"
      },
      "entity": {
        "type": "keyword"
      }
    }
  }
}

POST /test1/_doc
{
  "date": -1,
  "entity": "test1"
}

// response
{
  "error" : {
    "root_cause" : [
      {
        "type" : "mapper_parsing_exception",
        "reason" : "failed to parse field [date] of type [date] in document with id '80ohfnwBDmXpFPyRdOL0'. Preview of field's value: '-1'"
      }
    ],
    "type" : "mapper_parsing_exception",
    "reason" : "failed to parse field [date] of type [date] in document with id '80ohfnwBDmXpFPyRdOL0'. Preview of field's value: '-1'",
    "caused_by" : {
      "type" : "illegal_argument_exception",
      "reason" : "failed to parse date field [-1] with format [strict_date_optional_time||epoch_millis]",
      "caused_by" : {
        "type" : "date_time_parse_exception",
        "reason" : "Failed to parse with all enclosed parsers"
      }
    }
  },
  "status" : 400
}

[elasticmachine]

Pinging @elastic/es-core-infra (Team:Core/Infra)

[cbuescher]

Hi @jingshouyan,

thanks for raising this, but the behaviour is intended. I hope you don't mind me closing this issue.

Quoting from #72123:

Negative epoch times were deprecated in 6.7 (#36793) and removed in 7.0. If you are curious why, I recommend reading #40983 which had a long discussion as to the underlying reasons.

So this is working as intended. The number you are passing in is considered part of the year of the date, since no other parts exist. Please convert your date from epoch time to another date format for years prior to 1970.

[adagios]

I've found that the negative timestamps are still correctly parsed if you specify that the format is "epoch_millis" and use only second precision (timestamp should always end in 000). At least in 7.9.

Maybe it's an acceptable work-around for some cases.

[rkophs]

negative timestamps are still correctly parsed if you [...] use only second precision

This is the same conclusion that I have reached as well through testing and diving into how the current parsing logic works. I believe the regression was introduced in the way that sub-second precision is being parsed.

I'm hoping other the members of the ES community can take a look at #80208 to determine whether this is an acceptable solution to the problem.