Problem with parent id containing commas when talking to an index alias

[hkorte]

Hi,

I ran into problems trying to index a child document whose parent id contains commas. This only occurs using an index alias, with the index name it works. This is the resulting exception in 0.90.7:

ElasticSearchIllegalArgumentException: index/alias [parentid-testalias] provided with routing value [abc,2] that resolved to several routing values, rejecting operation
    at org.elasticsearch.cluster.metadata.MetaData.resolveIndexRouting(MetaData.java:338)
    at org.elasticsearch.action.index.IndexRequest.process(IndexRequest.java:557)
    ...

Here is some code to reproduce the problem:

Client client = new TransportClient().addTransportAddress(new InetSocketTransportAddress("localhost", 9300));

String indexName = "parentid-test";
String indexAlias = "parentid-testalias";
if (client.admin().indices().exists(new IndicesExistsRequest(indexName)).actionGet().isExists()) {
    client.admin().indices().delete(new DeleteIndexRequest(indexName)).actionGet();
}
CreateIndexRequest createIndexRequest = new CreateIndexRequest(indexName);
createIndexRequest.mapping("chld", "{\"chld\":{\"_parent\":{\"type\":\"prnt\"}}}");
client.admin().indices().create(createIndexRequest).actionGet();
client.admin().indices().prepareAliases().addAlias(indexName, indexAlias).execute().actionGet();

client.prepareIndex(indexName, "prnt", "abc,1").setSource("{\"text\":\"parent abc,1\"}").execute().actionGet();
client.prepareIndex(indexName, "chld", "def,1").setParent("abc,1").setSource("{\"text\":\"child def,1\"}").execute().actionGet();

client.prepareIndex(indexAlias, "prnt", "abc,2").setSource("{\"text\":\"parent abc,2\"}").execute().actionGet();
// this causes the exception - indexAlias and comma in parent id
client.prepareIndex(indexAlias, "chld", "def,2").setParent("abc,2").setSource("{\"text\":\"child def,2\"}").execute().actionGet();

client.close();

http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/indices-aliases.html tells me that this is intended behavior: "...search routing may contain several values separated by comma. Index routing can contain only a single value."

So, if this is how it should be, then I would suggest to add some documentation on invalid characters in IDs, as I didn't find anything regarding this. Any suggestions?

[martijnvg]

@hkorte I think this check that throws an error is only supposed to happen if you have a routing specified in the alias it self, not when someone is specifying it via routing or parent id. I think we can fix this, by only running this check if routing is resolved via the alias.

[martijnvg]

But having a comma in the routing is tricky if it used at search time, maybe this should just be not allow a comma in the routing.

[clintongormley]

Closing in favour of #6736