Guava's caches have overflow issues around 32GB with our default segment
count of 16 and weight of 1 unit per byte. We give them 100MB of headroom
so 31.9GB.
This limits the sizes of both the field data and filter caches, the two
large guava caches.
Closes#6268
Because json objects are unordered this also adds an explicit order syntax
that looks like
"highlight": {
"fields": [
{"title":{ /*params*/ }},
{"text":{ /*params*/ }}
]
}
This is not useful for any of the builtin highlighters but will be useful
in plugins.
Closes#4649
When multiple date formats are specified using the || syntax in the field mappings the date_histogram aggregation breaks. This is because we are getting a parser rather than a printer from the date formatter for the object we use to convert the DateTime values back into Strings. Simple fix to get the printer from the date format and test to back it up
Closes#6239
Assertion was triggered for percolating documents with nested object
in mapping if the document did not actually contain a nested object.
Reason:
MultiDocumentPercolatorIndex checks if the number of documents is
actualu >1. Instead we can just use the SingleDocumentPercolatorIndex
in this case.
closes#6263
The `FieldDataWeighter` allowed to use a concrete subclass of the caches
generic type to be used that causes ClassCastException and also trips the
CirciutBreaker to not be decremented appropriately.
This was tripped by settings randomization also part of this commit.
Closes#6260
Today we check if the DocIdSet we filter by is `fast` but the check fails
if the DocIdSet if wrapped in an `ApplyAcceptedDocsFilter` which is always
the case if the index has deleted documents. This commit unwraps
the original DocIdSet in the case of deleted documents.
Closes#6247
AllTokenStream, used to index the _all field, adds some overhead, but
it's not necessary when no fields were boosted or when positions are
not indexed the _all field.
Closes#6187Closes#6219
This adds support for sending a list of benchmark names and/or wildcard
patterns when submitting an abort request. Standard wildcards such as:
"*", "*xxx", and "xxx*" are supported. Multiple benchmark names and
wildcard patterns can be submitted together as a comma-separated list
from the REST API or as a string array from the Java API.
Closes#6185
This change adds a new cluster state that waits for the replication of a shard to finish before starting snapshotting process. Because this change adds a new snapshot state, an pre-1.2.0 nodes will not be able to join the 1.2.0 cluster that is currently running snapshot/restore operation.
Closes#5531
This is yet another safety guard to make sure we don't delete data if the local copy is the only one (even if it's not part of the cluster state any more)
Closes#6191
In some places we want to delay the start of a shard recovery because the source node is not ready to receive. At the moment the retry logic ignores the time delay parameter (`retryAfter`) causing a busy waiting like scenario. This is fixed in this commit.
Closes#6226
This commit upgrades to the latest Lucene 4.8.1 release including the
following bugfixes:
* An IndexThrottle now kicks in when merges start falling behind
limiting index threads to 1 until merges caught up. Closes#6066
* RateLimiter now kicks in at the configured rate where previously
the limiter was limiting at ~8MB/sec almost all the time. Closes#6018
Just follow "lowercase" token filter example and register "uppercase" token filter as an exposed token filter. This will not, by itself, test whether ES is correctly handling "uppercase" TF; this is more of a "code as documentation" fix.
This commit add `dummy docs` to `ElasticsearchIntegrationTest#indexRandom`.
It indexes document with an empty body into the indices specified by the docs
and deletes them after all docs have been indexed. This produces gaps in
the segments and enforces usage of accept docs on lower levels to ensure
the features work with delete documents as well.
We currently ask `MatchDocIdSet#matchDoc(int)` before consulting
the accept docs. This can also have a negative performance impact
since `matchDoc(int)` calls might be way more expensive than
acceptDocs calls.
Closes#6234
Nik Everett
Adrien Grand
Martijn van Groningen
Karel Minarik
Simon Willnauer
Alex Ksikes
Colin Goodheart-Smithe
javanna
Clinton Gormley
Britta Weber
Lee Hinman
Alexander Reelsen
mikemccand
Andrew Selden
Igor Motov
Boaz Leskes
Tiago Alves Macambira
Itamar Syn-Hershko