Embeddable, in-memory, document-oriented database with a high-level Query builder interface.
hash
/tree
-indexes. It requires extra memory, but gives ~20-50%% overall speedup (depending on the indexes selectivity). The column subindex may be disabled with the dense
index optionhash
/store
-indexes: now it uses column subindexes if possible for the cache efficiency. In some cases this approach provides up to 45% performance gain for the low-selectivity indexesstring
store(-)
indexes. Previosly store(-)
indexes have used the column for int
, int64
, double
, uuid
and bool
types only. The column subindex may be disabled with the dense
index optionWhereUUID
-method. Now it works for non-index fields tooAltLinux P10
array_remove
and array_remove_once
functions for the UPDATE-queries. Details
sparse
-indexessparse
-indexes. From now sparse
-indexes always use more effective IdSets insteadDeleteMeta
method and simplified meta storing/reading logicdistinct
result in cases when WHERE
-condition contains duplicate values (e.g. distinct(id) WHERE id IN (1,1,1,3)
)merge_limit
value (new values range is [1, 0x1FFFFFFF]
)EnumMeta
and DeleteMeta
functionsRenameNamespace
function - now upsert after namespace's renaming works properlyDELETE
-method for the /metabykey
HTTP-endpointAltLinux P10
joined
and composite
fields behavior: from now fields with those tags will not be created in the databaseRelease
build with GCC-12Google Benchmark
library during the buildclang-tidy
configs into GitHub
source treeclock_gettime
-calls. This gives up to 20% overall performance on the some operation systems, which using libstdc++ without vDSO (for example, on Centos7)Reindexer::Connect
-call is now thread-safeset_leader_node
commandconfig_rollback_timeout_sec
into sharding.conf
(option for debug purposes)COUNT()
for fulltext queries, when preselect_before_ft
is enabled in the index configLIMIT
/OFFSET
behavior for MERGE
-queries, when preselect_before_ft
is enabled in the index configexplain
results for the subqueries
Delete
and Update
queriesCOUNT_CACHED(*)
/CachedTotal()
aggregation in the queries with INNER JOINS. Now it's possible to cache total count results for such queriesbm25_config
field in the fulltext settings
EnableStorage
API was removedENABLE_V3_FOLLOWERS
cmake option, it is able to replicate to the reindexer v3 followers (v3.21.0 or newer)explain
for subqueries will be implement in the next releases)WithStrictJoinHandlers
. This option allows to validate JoinHandlers usage in runtimecproto
-binding. Error messages will no longer be redirected to stdout instead of user's loggercache
in the namespaces
entity of the #config
-namespace)urpcaddr
-flag in server's config)http-write-timeout
and http-read-timeout
options in server's config)Request-Timeout
header in HTTP-requests to setup individual timeout for each requestucproto://<socket_path>:/<dbname>
, example: ucproto:///tmp/reindexer.sock:/my_db
)ucproto://<socket_path>:/<dbname>
, example: ucproto:///tmp/reindexer.sock:/my_db
)EnableLogger
method betweenn Go and C++ mutexes