See issues on GitHub

Release Notes

Incompatible changes:

  • Gateway: Internal refactoring, requires manual upgrade when using fs gateway. (#232)
  • This release allows a newly started node to reuse up-to-date files in the working directory, without having to pull them from the gateway/master every time. Makes restarting a node much faster. This feature requires complete data re-indexing. (Note that when re-indexing data the metadata is automatically re-created too.)
  • The service wrapper has been moved to its own repo: http://github.com/elasticsearch/elasticsearch-servicewrapper. (#243)
  • Jgroups has been removed – now uses Zen instead
  • The terms API has been replaced by facets in search. (#242)
  • The search.facets API has changed (see below)
  • The shard level index.store.memory settings have been replaced with a node level memory store. (#235)
  • REST API: Cluster state (where possible) now returns the mappings as JSON, rather than as a string containing JSON.
  • REST Search API: Change score to _score to denote sorting by hit score. (#271)
  • Removed cloud plugin
  • Mapping: Revise dynamic mapping (into default), merge default to new mappings. (#275)

Enhancements:

Query DSL

  • Term Facets: A lot of work has been done to improve facet support:
    • Specify term facets that return the N most frequent terms that a field has, either globally, or bound by the search query (#207). See docs.
    • Facets can be filtered, by the same filters used in queries (#217). See docs.
    • A list of terms can be excluded from the results (#246). See docs.
    • Statistical facets: min, max, total, count, sum_of_squares, variance and std_deviation for numeric fields (#212). See docs.
    • Scripted statistical facets: define a script to be evaluated with statistical facets (#227). See docs.
    • Histogram facets: Create buckets from the values of one field (eg timestamp) and query total and count for another field in the same doc (#219). See docs.
  • Terms Factes: Allow to provide regex controlling which terms should be included. (#277)
  • Custom scripts:
    • Custom score (scripted) query. (#220)
      • The custom_score query uses the mvel scripting language to provide custom scoring for searches. For instance, you could use this to make documents with a more recent last_modified date more relevant than older documents. These mini-scripts can also accept parameters.
    • Script fields – searches can use the mvel scripting language to return calculated fields. (#221)
  • Fuzzy query support. (#213)
  • Alternate syntax for prefix (docs), term (docs), wildcard (docs), and span_term (docs) queries. (#192)
  • and, or and not filters (#216). See docs.
  • Cluster state (see docs):
    • Added parameters to filter the state on nodes, routing_table, metadata, and indices. (#234)
    • Now also includes information about index aliases, blocks and master_node id.
  • Added /_mapping, /{index}/_mapping and /{index}/{type}/_mapping to REST interface (#222). See docs.
  • Return the maxScore per search and _score for each search hit. (#205)

Other

  • Memory management much improved
  • Make replication type for index, delete, and delete_by_query operations configurable: async or sync (#196).
  • Java API has been improved (#199)
  • FS Gateway: Added index.gateway.fs.native_copy to control where native file copying will be used. (#202)
  • Plugins: Easier to plugin a custom DSL query/filter parsers. (#208)
  • Flush API: Added the full parameter for a complete clean. (#210)
  • Gateway: Configurable recovery_after_time and recover_after_nodes. (#223)
  • Automatic (and configurable) management of indexing buffer size. (#241)
  • Better behaviour when the whole cluster is shut down. (#250)
  • Time values (eg durations) now support days as well, eg “2d"
  • Zen Discovery: Configure which nodes are or are not allowed to become master with discovery.zen.master. (#248)
  • Analysis:
    • Fields that are not_analyzed should automatically default to keyword analyzer. (#229)
    • When specifying empty array for stopwords, use an empty list for stopwords. (#230)
  • Make merging mappings smarter. (#253)
  • Translog: Implement a file system based translog and make it the default. (#260)
  • If there isn't a valid score for result hits then return null for max_score value. (#263). The same logic applies also to _score values of individual documents.
  • Support Cross-Origin resource in http/rest module. (#218)
  • Cluster Health API: Add wait_for_nodes (accepts “N", “<N", “>N", “<=N", and “>=N"). (#269)
  • Implemented specific cloud plugin for AWS with S3 gateway and EC2 discovery.
  • Analysis: Add pattern analyzer. (#276)

Bug fixes:

  • Zen Discovery:
    • Ungraceful shutdown of the master and start of replacement node might cause the cluster not to elect a new master. (#200)
    • When a master node is forcefully killed, other nodes might not monitor the other elected master. (#224)
    • A node might get into an infinite state of trying to find a master (when client / non_master) nodes exists. (#247)
  • Don't create / use the work directory if not needed (for example, on client / non data) nodes. (#249)
  • Nodes Info API: Failed to generate REST response when node attributes are set. (#201)
  • REST API does not expose node-master status. (#203)
  • Query DSL: field query does not take into account allow_leading_wildcards. (#236)
  • Added checks to ensure survival when a cached stream is removed mid process
  • Search requests hangs when no indices exists. (#209)
  • Don't fail when a flush cannot be completed
  • Cluster state is now refreshed on restart, clearing out old information.
  • Block operations performed on a cluster until it has recovered from the gateway. (#239)
  • Put Mapping fails when an analyzer is specified that was not configured. (#252)
  • Sorting breaks when sorting on a field that has no value in some documents.
  • Facet results vary depending on size. (#259)
  • Querying mapping on a non-master throws an error. (#261)
  • Put Mapping: When updating existing mappings, the request returns with acknowledged `false`. (#262)
  • Search: Sending a request that fails to parse can cause file leaks. (#270)
  • Mapping: Dynamic mapping definitions are ignored. (#274)

Internal:

  • Library upgrades:
    • Lucene: now at version 3.0.2.
    • netty: now at version 3.2.1 Final
    • async http client: now at 1.0.0
  • Use Java's user.dir as the default path.home location
  • cached termsSet used in dfs phase
  • Logging improvements
  • Update build files to reference hamcrest
  • Always start the unicast ping discovery, so unicast discovery will work even when using multicast
  • Rest interface now accepts HTTP OPTIONS and HEAD
  • Clean up files that are in the working directory but not in the gateway
  • Big jarjar refactoring, guice now jarjar'ed
  • Tcp Transport: Reduce transport.tcp.connection_per_node from 5 to 1. (#225)
  • Detect write errors early and notify transport handler
  • Connect to a node when it joins the cluster on the disco level, so if it fails, it will be propagated back and the node will not be added to the cluster state
  • Simplify netty transport to use single channel
  • Refactor builder requets into common base class
  • Groovy:
    • Refactor util.xcontent to common.xcontent in client
    • Add prepare methods to groovy APIs
  • Gateway: HDFS gateway refactored to use the new common blobstore
  • Improve join process in cluster, fetch the cluster meta-data on join and handle new meta data