Release Notes
Breaking changes:
- Remove
RestActions#splitXXX(String) methods #3680 - Debian & RPM packages: Disable immediate restart on package upgrade #3685
- Flush API: Removed the refresh flag #3689
- Optimize API: Removed the refresh flag #3690
- Thread Pool: Removed blocking type option #3531
New features:
- Function scoring. Do not miss the great explanation at the github issue #3423
- An API to clear scroll requests has been added #3657
- Shard allocation takes free disk space into account #3480
- Suggestions can be highlighted in the phrase suggester #3442
- Random ordering (as part of the new function scoring) #1170
Enhancements:
- A dedicated suggest thread pool has been added #3698
- Debian package: Support for ARMHF in JDK_DIRS #3659
- Rename
IndexShard#searcher() to IndexShard#acquireSearcher() #3653 - CompletionSuggest throws an exception if input string contains a reserved character #3648
- Thread pools: Allow to control the number of processors sizes are based on #3643
BytesStreamOutput default size should be 2k instead of 32k #3638- Optionally ignore index names in request body for multi-get/search/bulk when indices are already given in url #3636
- Use different LZF compression on Solaris to prevent segfaults #3634
- Make the acceptable compression overhead used by MultiOrdinals configurable and default to FASTEST #3623
- Remove
org.elasticsearch.common.UUID and replace it with a simplified version #3605 - Preventing possible stack overflow in completion suggester when using long input strings #3596
- Support
_name for queries as already supported by filters #3581 - Improved refresh logic when replica move to started #3573
- Smarter default to
index.index_concurrency #3546 - Bound processor size based calculations to a maximum of 32 #3545
- Add option to disable printing out readable size and time values #3432
- Renamed
readable_format flag to human #3541 - Improved exception handling in actions when forking to a thread pool #3524
- Make
RestSearchAction#parseSearchXXX(RestRequest) public #3499 - Raised default timeout for deleting an index #3498
- Improved backwards compatibility handling for NGram / EdgeNGram analysis #3489
- Changed default
operation_threading from single_thread to thread_per_shard #3483 - Suggest ignores empty shards #3473
- Completion suggestion enhancements
- Support FuzzySuggester #3465
- Allow payload to be an arbitrary value #3550
- Statistics for completion suggest are exposed now #3522
- Expose
IndexWriter#setUseCompoundFile() via engine settings. #3461 - Improved filtering by
_parent field #3454 - Plugin Manager enhancements
- Support silent mode #3628
- Support
-remove group/artifact/version naming scheme #3421 - Install site plugin with custom url doesn't filter directory #3582
- Index template enhancements
- Enable
GET /_template to show all templates #2532 - Add client method to get a specific index template #3439
- Mapping: Changing
_source excludes/includes is now possible at runtime #3491
Bug fixes:
- Briefly delete manifested mapping type on a node without reason #3697
- Debian package: Init script fails due to unsupported syntax on non-Bash shells #3691
- Rare race condition when introducing new fields into a mapping #3667
JsonGenerationException thrown in SuggestResponse#toString method #3661- CompletionStats: A resource leak was fixed, since requested searchers were not closed #3652
- CompletionStats: Possible
NullPointerException was fixed #3619 - BalancedShardsAllocator prematurely modified
unassigned shards list #3610 - Closing an already closed index lead to a
NullPointerException #3601 ArrayIndexOutOfBoundsException when using empty preference parameter #3591- Forced awareness fails to balance shards #3580
- Cluster settings update could hang if received settings which were not dynamically updatable #3560
NullPointerException during concurrent merges #3555- Plugin Manager can not download _site plugins from github #3551
- Wrong analyzer used when indexing dynamic property #3544
- Date math not working correctly due to lower casing #3540
- Dynamic templates from an index template could be skipped if a new type already have dynamic templates #3538
- Setting index/bulk thread pools with
queue_size could cause replica shard failures #3526 - Removed static versions of MatchAllDocsQuery #3521
NullPointerException during discovery #3515- Certain exceptions could result in an unreleased search context #3513
- Concurrent Put Mapping API to multiple indices/types could return prematurely #3507
- Phrase queries automatically generated by query string ignored boosts #3503
- Put mapping requests were prematurely acknowledged if other nodes were quicker than master #3487
FastVectorHighlighter failed with StackOverflow on terms with large TermFrequency #3486- A
NullPointerException for POST mode facets was fixed if facet_filter accepts no documents #3479 - An empty top level filter lead to failed search requests #3477
- Changing the
_default_ mapping was validated is if it was a normal type #3476 - When replacing an existing
_default_ type, the old one gets now correctly merged into the new mapping #3474 - A possible
NullPointerException in BytesRefOrdValComparator was fixed #3470 - Prevent Phrase Suggester from failing on missing fields #3469
- Inconsistent usage of
ScriptScoreFunction in FiltersFunctionScoreQuery #3464 - The
bin/plugin shell script did not exit with a useful exit code on error #3463 - More-Like-This query returned all documents if non of the fields in the document are supported #3453
- Debian package: Older version of start-stop-daemon did not support
--status parameter #3452 - Geoshape filter could not handle multiple shapes #3242
- CompletionFieldMapper ignored
path: just_name #3669 - Many cores could lead to huge thread pools #3478