On 25.01.2016 13:07, Tomas Vondra wrote:
Right, it's definitely not thread-safe so there'd need to be some lock protecting the regex_t copy. I was thinking about either using a group of locks, each protecting a small subset of the affixes (thus making it possible to work in parallel to some extent), or simply using a single lock and each process would make a private copy at the beginning. In the end, I've decided to do it differently, and simply parse the affix list from scratch in each process. The affix list is tiny and takes less than a millisecond to parse in most cases, and I don't have to care about the regex stuff at all. The main benefit is from sharing parsed wordlist anyway.
This is nice decision since the affix list is small. For our task I will change shared_ispell to use this solution.
It's an old-school shared segment created by the extension at init time. You're right the size is fixed so it's possible to run out of space by loading too many dictionaries, but that was not a big deal for the type of setups it was designed for - in those cases the list of dictionaries is stable, so it's possible to size the segment accordingly in advance. But I guess we could do better now that we have dynamic shared memory, possibly allocating one segment per dictionary as needed, or something like that. regards
Yes it would be better as we will not need to define the maximum size of the shared segment in postgresql.conf.
-- Artur Zakirov Postgres Professional: http://www.postgrespro.com Russian Postgres Company -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers