Breaking changes in 6.7

This section discusses the changes that you need to be aware of when migrating your application to Elasticsearch 6.7.

See also Release Highlights and Release Notes.

internal version may not uniquely identify a document’s version if an indexed document wasn’t fully replicated when a primary fails. As such it is unsafe to use for optimistic concurrency control, is deprecated and the option will no longer be available in Elasticsearch 7.0.0. Please use the if_seq_no and if_primary_term parameters instead. See Optimistic concurrency control for more details.

The ingest-geoip and ingest-user-agent plugins have been converted to modules and are now included with all Elasticsearch distributions. Starting with Elasticsearch 6.7.0, attempting to install or remove these plugins will result in a no-op. In Elasticsearch 7.0.0, attempting to install or remove these plugins will result in an error. Additionally, there are two minor breaking changes here: - elasticsearch-plugin list will no longer output ingest-geoip nor ingest-user-agent - plugin.mandatory is no longer compatible with ingest-geoip nor ingest-user-agent

Elasticsearch 6.7.0 checks that there are no leftover geoip database files in the plugin configuration directory because the new module does not use them. Therefore, remove the ingest-geoip plugin prior to upgrading to 6.7.0 with the --purge option to also delete the old database files:

./bin/elasticsearch-plugin remove --purge ingest-geoip

Otherwise you will see the following error message upon startup (assuming /etc/elasticsearch/ingest-geoip as the plugin configuration directory):

expected database [GeoLite2-ASN.mmdb] to not exist in [/etc/elasticsearch/ingest-geoip]

The setting node.store.allow_mmapfs has been deprecated. Use node.store.allow_mmap instead.

Since 6.0.0, the /_cluster/state REST endpoint and cluster state API on the transport client would compute and return the compressed size of the cluster state as if it were serialized for inter-node communication. This metric would be returned in the compressed_size or compressed_size_in_bytes fields in the REST endpoint, or ClusterStateResponse#getTotalCompressedSize via the transport client. Computing this field is expensive, and has dubious reporting value as it does not reflect the actual size of the cluster state on disk. The expense of computing this field is such that in some situations it can disrupt all activity on a node including indexing activity. Therefore we have elected to by default stop computing this field in a minor release. We are providing a backwards compatibility flag via the system property -Des.cluster_state.size=true which will return to the previous 6.0.0—​6.6.2 behavior. Computing this metric and displaying this field will be completely removed in 7.0.0, and using this backwards compatibility flag will fail.

The token service no longer accepts tokens that were generated on an Elasticsearch version earlier than 6.2.0

No applications in the Elastic Stack made use of Elasticsearch tokens prior to 6.2.0, so this change does not affect access to Kibana or any official Elasticsearch clients. If you have an application that uses Elasticsearch tokens, and a token from 6.1 or earlier is presented for authentication, Elasticsearch will reject the token with a 401 status code.