Need to respecify `_parent` when updating mapping of child type

[Naitreey]

Elasticsearch version: 5.2.1 docker image, 5.1.2 docker image, 5.0.1 RPM package

Plugins installed: []

JVM version: Oracle Corporation/OpenJDK 64-Bit Server VM/1.8.0_121/25.121-b13

OS version: Linux/4.3.0-1.el6.elrepo.x86_64/amd64

Description of the problem including expected versus actual behavior:
I am able to reproduce the issue presented in #15997 with Elasticsearch 5.2.1,
5.1.2 docker images, and another 5.0.1 RPM package.

Specifically:

PUT t 
{
  "mappings": {
    "parent": {},
    "child": {
      "_parent": {
        "type": "parent"
      }
    }
  }
}

PUT t/_mapping/child
{
  "properties": {}
}

results in illegal_argument_exception.

Again, providing _parent type definition in update request body worked
around the issue.

Provide logs (if relevant):

[2017-02-27T19:51:08,706][DEBUG][o.e.a.a.i.m.p.TransportPutMappingAction] [default] failed to put mappings on indices [[[t/AK3xoZCBSCK1Wj6N81GGlA]]], type [child]
java.lang.IllegalArgumentException: The _parent field's type option can't be changed: [parent]->[null]
	at org.elasticsearch.index.mapper.ParentFieldMapper.doMerge(ParentFieldMapper.java:301) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.index.mapper.FieldMapper.merge(FieldMapper.java:333) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.index.mapper.MetadataFieldMapper.merge(MetadataFieldMapper.java:73) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.index.mapper.Mapping.merge(Mapping.java:96) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.index.mapper.DocumentMapper.merge(DocumentMapper.java:333) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.metadata.MetaDataMappingService$PutMappingExecutor.applyRequest(MetaDataMappingService.java:267) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.metadata.MetaDataMappingService$PutMappingExecutor.execute(MetaDataMappingService.java:230) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.service.ClusterService.executeTasks(ClusterService.java:674) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.service.ClusterService.calculateTaskOutputs(ClusterService.java:653) ~[elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.service.ClusterService.runTasks(ClusterService.java:612) [elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.cluster.service.ClusterService$UpdateTask.run(ClusterService.java:1112) [elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.common.util.concurrent.ThreadContext$ContextPreservingRunnable.run(ThreadContext.java:527) [elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.common.util.concurrent.PrioritizedEsThreadPoolExecutor$TieBreakingPrioritizedRunnable.runAndClean(PrioritizedEsThreadPoolExecutor.java:238) [elasticsearch-5.2.1.jar:5.2.1]
	at org.elasticsearch.common.util.concurrent.PrioritizedEsThreadPoolExecutor$TieBreakingPrioritizedRunnable.run(PrioritizedEsThreadPoolExecutor.java:201) [elasticsearch-5.2.1.jar:5.2.1]
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) [?:1.8.0_121]
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) [?:1.8.0_121]
	at java.lang.Thread.run(Thread.java:745) [?:1.8.0_121]

[StefanGor]

Hi there,
I'd be interested in trying to tackle this one.

[StefanGor]

Actually, the other issue that I am working on is taking longer than expected. If anyone else wants to take this in the meantime, they are welcome to.

[PnPie]

Hello,

I see in the ParentFieldMapper.java, when we do a Mapping update of a type, we merge all the MetadataFieldMapper by calling its merge method with the new mappers, so in the implementation of ParentFieldMapper
https://github.com/elastic/elasticsearch/blob/master/core/src/main/java/org/elasticsearch/index/mapper/ParentFieldMapper.java#L300
when we merge the current with the new one, we checked if they are not equal, we throw an Exception, but if we didn't explicitly specify the _parent field in the new mapping, the parentType field of new mapper should be null, in this case we can just keep the original mapper unchanged as we intended instead of throw an Exception of trying to change it from [sometype] to [null].

I also made a PR with it, thanks for having a check.