Mapping corruption after data replication or reallocation


For the second time, restarting sequentially the two shards of a same cluster results in an incorrect mapping for one a my index. The special "_source" field is defined in the resulting mapping as a "normal" field, indexed as a standard string... The consequences are these ones : new indexation are bad (no data, only _id, and some technical fields), all queries using "_source" filtering fails.

I have a cluster with two shards. One index is defined with one shard and one replica. I sequentially restart shards, waiting a green status between first and second restart.

Elasticsearch version is 2.0.0.

Can I have help on this issue?