Fix time series timestamp meta missing

[weizijun]

I find a time series timestamp meta missing case.

here is the reproduce steps:

1. create a time_series index, and set the timestamp field some meta.
2. index a doc with a new field that is not in mappings, it will call mappings marge.
3. then the timestamp field meta is missing.

the reason that meta is missing is when a new field comes, MappingParser.parse will build a new mapping with new fields. And merge the new mappings with exist mapping.
the new mapping have no timestamp field, so it will auto add timestamp field, the timestamp is without user's meta info.
And merge method build a new timestamp field to override the user's timestamp field.
It cause the timestamp meta missing.

I fixed the case, by move timestamp logic from MappingParser.parse to create index logic.
And move the tests to a new IT test.
I add a test to test case, TimeSeriesModeIT.testAddTimeStampMeta will fail in the pre-commit.

[elasticmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[weizijun]

@elasticmachine update branch

[nik9000]

For my own education I reproduced this:

curl -uelastic:password -HContent-Type:application/json -XPUT localhost:9200/foo?pretty -d'{
  "settings": {
    "index": {
      "mode": "time_series",
      "routing_path": "dim"
    }
  },
  "mappings": {
    "properties": {
      "@timestamp": {
        "type": "date",
        "meta": {
          "m": "words words words"
        }
      },
      "dim": {
        "type": "keyword",
        "time_series_dimension": true
      }
    }
  }
}'


curl -uelastic:password localhost:9200/foo/_mappings?pretty
# meta is present

curl -uelastic:password -HContent-Type:application/json -XPOST localhost:9200/foo/_doc?pretty -d'{
  "@timestamp": "2021-11-16T09:48:20Z",
  "dim": "a",
  "cat": "mouse"
}'

curl -uelastic:password localhost:9200/foo/_mappings?pretty
# meta is absent

[nik9000]

I'd tend to want to do these as yaml rest tests instead of ESIntegTestCase. The yaml tests are much more copy and paste-ish, we run them during rolling upgrade testing and during backwards compatibility testing.

I'm not sure that it's worth rewriting it now though.

[weizijun]

fine, I move the test from TimeSeriesModeTests to TimeSeriesModeIT, some tests are the same with yaml tests. Maybe later, we can remove these tests.

[nik9000]

I think IndexSettings.getMode can't be null - it'd default to STANDARD.

I think I'd prefer to move this merge helping code to IndexMode too, if we can. If every behavior change in different index modes is a method on IndexMode then folks reading it can scan it for the kinds of behavior that it influences. Hard checks for a particular mode like this make that much harder.

[weizijun]

fine, I'm trying to move these code into IndexMode

[weizijun]

I change the code, is it ok?

[nik9000]

Makes sense to me.

[nik9000]

@elasticmachine, test this please

[nik9000]

@elasticmachine update branch

[nik9000]

@elasticmachine test this please

[nik9000]

@weizijun the errors in the build look related to your change.

[weizijun]

@weizijun the errors in the build look related to your change.

ok, I will fixed the tests

[nik9000]

@elasticmachine, test this please

[martijnvg]

LGTM 👍

[nik9000]

Thanks for your patience on this one @weizijun !

[weizijun]

Thank you, @nik9000 , @martijnvg !