Support for MariaDB database

[HugoWenTD]

MariaDB supports Vector now. https://mariadb.com/kb/en/vector-overview/
Add new module for benchmark against MariaDB 11.8 database server.

[maumueller]

Thanks @HugoWenTD. Please make sure to add a test case to the CI as well.

[erikbern]

Are you planning to finish this @HugoWenTD ?

[HugoWenTD]

Are you planning to finish this @HugoWenTD ?

Yes Erik, I'll work on it this week.

[erikbern]

Ok, nice!

[cvicentiu]

Perhaps it would make sense to time just the "executemany" calls, not the python overhead of creating the batch?

If one is to compare these numbers to other DBs, it should time the operation the same. Is it with python overhead, or without in other DB's cases?

[erikbern]

We don't really benchmark insertions & building the index, so I don't think it's material!

[HugoWenTD]

Yeah, it's just a print, but I'll see if can change it.

[erikbern]

This is failing in CI for some reason, any idea why? https://github.com/erikbern/ann-benchmarks/actions/runs/14016901547/job/39890033550?pr=548

[HugoWenTD]

This is failing in CI for some reason, any idea why? https://github.com/erikbern/ann-benchmarks/actions/runs/14016901547/job/39890033550?pr=548

Weird. Looks it's failing with --batch option. I'll look into it.

ERROR - double free or corruption (!prev)

[HugoWenTD]

This is failing in CI for some reason, any idea why? https://github.com/erikbern/ann-benchmarks/actions/runs/14016901547/job/39890033550?pr=548

I suspect it's related to the ThreadPool and mariadb cursor.

    def batch_query(self, X: numpy.array, n: int) -> None:
        pool = ThreadPool()
        self.res = pool.map(lambda q: self.query(q, n), X)

    def query(self, v, n):
        self._cur.execute(self._sql_search, (self.vector_to_hex(v), n))

        return [id for id, in self._cur.fetchall()]

[HugoWenTD]

Updated the batch_query. It should work now.