We are please to announce the release of Manticore Search 2.5.1.
Prebuild packages available for Debian,Ubuntu, Red Hat/CentOS, Windows and OSX can be downloaded from here.
A new endpoint is added to the HTTP API which allows writing queries in a JSON format. While it doesn't have yet all the functionality of SphinxQL, it allows performing searching and data manipulation operations (insert/update/replace/deletes).There are some limitations right now as the data manipulation operations don't handle yet MVA and JSON attributes.
Please note that JSON queries are in an early stage and may suffer changes in future. We are welcoming anyone to give it a try and send us feedback. We are not looking only for discovering potential bugs, but also for any kind of suggestion regarding this new way of connecting to Manticore Search.
New features and improvements
SHOW THREADS command will now list operations in progress like index optimizations, rotations, flushes.
New command RELOAD INDEXES added, which has same effect as sending a HUP signal to searchd daemon.
New command FLUSH LOGS added, which has the same effect as sending a USR1 signal to searchd daemon.
GROUP N BY works now correctly when used against MVA attributes.
Multi value attributes build from a sql source can now use a ranged query using the ranged step and ranged values from main sql_query.
sql_attr_multi = uint mva from ranged-main-query; select id, mva from mva where id between $start and $end
Added packages for Debian 9 and OSX 10.13. Improved support for FreeBSD,OSX and Alpine platforms.
This release brings reference counts to indexes. Rotations during high loads could get delayed by existing running queries and delay in their turn incoming queries. With the new change, existing queries are served by the old version of the index, while incoming queries get served by the new version of the index, so both rotations and queries are not stalled anymore. This means the old version will remain in memory until queries running against it finish to execute.
Another change made is related to the blackhole agent. While the master node doesn't wait for the response of the query sent to the blackhole node, it does try to connect to it and wait for a reply. This means in case of a failure of a blackhole node, the master tries to reconnect (depending on the retry count setting) ot it so there was the possibility for the master to get stalled by the blackhole node. Now the blackhole node are handled by a separate thread from the one that runs the master query to not affect the it anymore.
For a full report of changes please consult the Release note .
Early note for next release
Our next release will be in the second half of December. This release will drop support for 32bit document IDs. The 32bit IDs are from the early days of Sphinx when hardware was more limited and we feel there is no reason anymore to support them.