Multi data path config support writes a file to a data location based on the available size (by default). There is a Lucene file called segments.gen that has the same name, and only in that case, we need to make sure we alway write it to the same data location, otherwise, the index will have multiple segments.gen files, and the shard can seem to be corrupted.
The message if this case happens is that segments_xxx file was not found, in which case, a find for segments.gen can yield multiple files. Deleting the segments.gen files will cause the shard to recover properly (as its an extra protection layer to resolve the segments header by Lucene)
Make sure the segments.gen file is writtne to the same directory every time
fixes#4674
this will allow to easily access them regardless of the transport address used, and allow to remove reverse lookup of host names from ip address extracted from inet transport address where needed
Recycling is not thread-local anymore by default but instead there are several
pools of objects to recycle that threads may use depending on their id.
Each pool is protected by its own lock so up to ${number of pools} threads may
recycler objects concurrently.
Recyclers have also been refactored for better composability, for example there
is a soft recycler that creates a recycler that wraps data around a
SoftReference and a thread-local recycler that can take any factory or recyclers
and instantiates a dedicated instance per thread.
RecyclerBenchmark has been added to try to figure out the overhead of object
recycling depending on the recycler type and the number of threads trying to
recycle objects concurrently.
Close#4647
A lot of different API's currently use different names for the
same logical parameter. Since lucene moved away from the notion
of a `similarity` and now uses an `fuzziness` we should generalize
this and encapsulate the generation, parsing and creation of these
settings across all queries.
This commit adds a new `Fuzziness` class that handles the renaming
and generalization in a backwards compatible manner.
This commit also added a ParseField class to better support deprecated
Query DSL parameters
The ParseField class allows specifying parameger that have been deprecated.
Those parameters can be more easily tracked and removed in future version.
This also allows to run queries in `strict` mode per index to throw
exceptions if a query is executed with deprected keys.
Closes#4082
Because one URI parameter can contain either nodeIds or a list of metrics,
the check to detect if this parameter is either a nodeId or a metric needs
to be more accurate.
If a type or path is missing in the REST test yaml file, it is
automatically replaced with _all. This makes it hard to test changes
in the api, for example adding the possibility to leave the index
blank in addition to _all and * in the uri.
closes#4657
`allocation.disable_new_allocation`, `allocation.disable_allocation`, `allocation.disable_replica_allocation`,
in favour for the enable allocation decider which has a single option `allocation.enable` wich can be set to the following values:
`none`, `new_primaries`, `primaries` and `all` (default).
Closes#4488
Common changes:
Lastname, Firstname -> Firstname Lastname
Name I -> Name
Title-style capitalization
Removed L-to-R characters ( <e200> )
Removed duplicates
Alphabetical order
Named wildcards were not always properly replaced with proper values by PathTrie.
Delete index (curl -XDELETE localhost:9200/*) worked anyway as the named wildcard is the last path element (and even if {index} didn't get replaced with '*', the empty string would have mapped to all indices anyway). When the named wildcard wasn't the last path element (e.g. curl -XPOST localhost:29200/*/_close), the variable didn't get replaced with the current '*' value, but with the empty string, which leads to an error as empty index is not allowed by open/close index.
Closes#4564
The new internal get index settings api is more efficient when it comes to sending the index settings from the master to the client via the
Also the get index settings support now all the indices options.
Closes#4620
The FVH was throwing away some boosts on queries stopping a number of
ways to boost phrase matches to the top of the list of fragments from
working.
The plain highlighter also doesn't work for this but that is because it
doesn't support the concept of the same term having a different score at
different positions.
Also update documentation claiming that FHV is nicer for weighing terms
found by query combinations.
Closes#4351
Important: This breaks backwards compatibility with 0.90
* Removed endpoints: /_cluster/nodes, /_cluster/nodes/nodeId1,nodeId2
* Disallow usage of parameters, but make required metrics part of URI
* Changed NodesInfoRequest to return everything by default
* Fixed NPE in NodesInfoResponse
Closes#4055
Luca Cavanna
Simon Willnauer
Shay Banon
spenceralger
Spencer Alger
Adrien Grand
Andrew Raines
uboness
Clinton Gormley
Alexander Reelsen
Martijn van Groningen
Britta Weber
David Pilato
Leonardo Menezes
Nik Everett
Honza Král
ohnorobo
Costin Leau
Igor Motov