The functions provided by the shared_ispell module are shown in Table F.50.

The module needs to allocate space in the shared memory segment. So add this to the config file (or update the current values):

# libraries to load
shared_preload_libraries = 'shared_ispell'

# config of the shared memory
shared_ispell.max_size = 32MB

To find out how much memory you actually need, use a large value (e.g. 200MB) and load all the dictionaries you want to use. Then use the shared_ispell_mem_used() function to find out how much memory was actually used (and set the shared_ispell.max_size GUC variable accordingly).

Don't set it exactly to that value, leave there some free space, so that you can reload the dictionaries without changing the GUC max_size limit (which requires a restart of the DB). Something like 512kB should be just fine.

The extension defines a shared_ispell template that you may use to define custom dictionaries. E.g. you may do this:

CREATE TEXT SEARCH DICTIONARY english_shared (
    TEMPLATE = shared_ispell,
    DictFile = en_us,
    AffFile = en_us,
    StopWords = english
);

CREATE TEXT SEARCH CONFIGURATION public.english_shared
    ( COPY = pg_catalog.simple );

ALTER TEXT SEARCH CONFIGURATION english_shared
    ALTER MAPPING FOR asciiword, asciihword, hword_asciipart,
                    word, hword, hword_part
    WITH english_shared, english_stem;

We can test created configuration:

SELECT * FROM ts_debug('english_shared', 'abilities');
   alias   |   description   |   token   |         dictionaries          |   dictionary   |  lexemes  
-----------+-----------------+-----------+-------------------------------+----------------+-----------
 asciiword | Word, all ASCII | abilities | {english_shared,english_stem} | english_shared | {ability}
(1 row)

Or you can update your own text search configuration. For example, you have the public.english dictionary. You can update it to use the shared_ispell template:

ALTER TEXT SEARCH CONFIGURATION public.english
    ALTER MAPPING FOR asciiword, asciihword, hword_asciipart,
                    word, hword, hword_part
    WITH english_shared, english_stem;

Tomas Vondra <tomas.vondra@2ndquadrant.com>, Prague, Czech Republic