[groonga-talk:316] [ANN] Groonga 6.0.1

[HAYASHI K. <ha...@cl...>]

Hi,

Groonga 6.0.1 has been released!
  http://groonga.org/docs/news.html#release-6-0-1

Install: http://groonga.org/docs/install.html
Charcteristics:  http://groonga.org/docs/characteristic.html

The topics in this release:

* [hash] Supported 4GiB over total key size
* Supported to remove dependent tables
* [experimental] Supported to split chunks in static indexing

This release includes some bug fixes of load command, so upgrade Groonga!

## [hash] Supported 4GiB over total key size

In this release, 4GiB over total key size is supported for a hash table.

In the previous versions of Groonga, there is a limitation about total
key size of a table - 4GiB total key size.

Such a limitation is now relaxed. By using KEY_LARGE flag with a table,
the value of new limitation of the total key size is changed to 1TiB.

Here is the example to use KEY_LARGE flag:

  table_create Users TABLE_HASH_KEY|KEY_LARGE ShortText

As a result, it handles more large table data in a hash table.

## Supported to remove dependent tables

In this release, a feature to remove dependent tables is supported for
table_remove and logical_table_remove command.

In the previous versions of Groonga, removing tables are failed by
default when the following conditions are met:

* One or more tables use the table as key type.
* One or more columns use the table as value type.

This is a intended behavior because it is aimed not to remove
referenced table unexpectedly by default.

On the other hand, it is a bit hard to remove if there are many
dependent tables which you want to remove at once.

For such a purpose, the feature "--dependent yes" is implemented.

Here is the sample schema:

  table_create ReferencedTable TABLE_HASH_KEY ShortText
  table_create Table1 TABLE_HASH_KEY ReferencedTable
  table_create Table2 TABLE_NO_KEY
  column_create Table2 reference_column COLUMN_SCALAR ReferencedTable

When "table_remove ReferencedTable" is executed, it just fails because
ReferencedTable is the key of Table1 and referenced from
reference_column of Table2.

Execute "table_remove ReferencedTable --dependent yes" is executed,
Table1 and reference_column of Table2 are removed correctly.

NOTE: there is a case that "--dependent yes" consumes a lot of
resources if there are many dependent tables because it opens all
dependent tables and columns temporary. In such a case, reduce the max
number of threads by "thread_limit 1" command.

## [experimental] Supported to split chunks in static indexing

In this release, splitting chunks in static indexing are supported.

It is aimed to achieve fast phrase search when frequently used word and
rarely used word are specified at the same time.

To enable this feature, set the following environment variable and
execute offline index construction:

  GRN_INDEX_CHUNK_SPLIT_ENABLE=1

And enable the following environment variable:

  GRN_II_CURSOR_SET_MIN_ENABLE=1

This feature is disabled by default because it is not widely tested
yet, but it is verified to improve phrase search performance in a test
case which is described above.

## Improvements

  * [mruby] Updated bundled mruby to fix a crash bug related to GC and
    backtrace.
  * Exported the following API
     * grn_expr_take_obj()
     * grn_request_canceler_cancel_all()
     * grn_obj_remove_dependent()
     * grn_obj_is_text_family_type()
  * [hash] Supported 4GiB over total key size when KEY_LARGE flag is
    set to a table:
  * table_create Users TABLE_HASH_KEY|KEY_LARGE ShortText
  * [load] Support request ID with raw JSON input case:
    POST /d/load?table=XXX&request_id=x and

    load --table XXX --request_id x
    [
    ...
    ]
  * [shutdown] Added mode argument to shutdown immediately. Use
    "shutdown --mode immediate" in such a purpose.
  * [Mac OS X] Added a instruction to setup MeCab dictionary for
    Homebrew.
  * [load] Supported to stop load when cancel is requested.
  * [table_remove] Supported to remove dependent tables. Use
    "--dependent yes" for it.
  * [logical_table_remove] Supported to remove dependent tables.
  * [Windows] Supported memory debug mode on Windows.
  * Supported to dump allocation information by status on memory debug
    mode
  * [I18N] Added installation step for Sphinx.
  * [experimental] Supported to split chunks in static indexing. Use
    GRN_INDEX_CHUNK_SPLIT_ENABLE=yes to enable it. [GitHub#479]

## Fixes

  * [load] Fixed a crash bug when nonexistent column is
    specified. [GitHub#493]
  * [load] Fixed a bug that load command does not return error code
    correctly. [GitHub#495]
  * [load] Fixed a memory leak when parsing columns parameter in load
    command.
  * [load] Fixed a bug that only the first array in --values is
    handled as a list of column names if --columns is not
    specified. [GitHub#497]
  * [load] Fixed to check --columns more precisely [GitHub#496]
  * Fixed a insufficient critical section handling for thread-safety.
  * [column_create] Fixed a crash bug when failed to create a column.
  * [table_remove] Fixed a crash bug to remove nonexistent
    table. [GitHub#504]
  * Fixed a bug that offline index construction against WITH_POSITION
    + non-text VECTOR column ignores position.
  * [grndb] Fixed a bug that cycle reference causes stack over flow.
  * [deb] Dropped support for Ubuntu 15.04 (Vivid Vervet)

## Thanks

  * YUKI Hiroshi


-- 
HAYASHI Kentaro <ha...@cl...>