matomo-org / matomo

Empowering People Ethically with the leading open source alternative to Google Analytics that gives you full control over your data. Matomo lets you easily collect data from websites & apps and visualise this data and extract insights. Privacy is built-in. Liberating Web Analytics. Star us on Github? +1. And we love Pull Requests!
https://matomo.org/
GNU General Public License v3.0
19.94k stars 2.66k forks source link

Add support for TiDB #19655

Open bx80 opened 2 years ago

bx80 commented 2 years ago

After the linked general database abstraction issues have been completed it should be possible to create a new PDO adapter and schema definition to trial running Matomo entirely on TiDB. an open source distributed MySQL compatible database.

This change will allow for long term suitability testing of TiDB as an alternative high-performance database backend for Matomo.

Suggested Approach

Related general database abstraction issues:

19643: Abstract the database get_lock function to PDO adapter

19646: Support option for ranking queries to use window functions

19647: Abstract database specific tests and fixture code

19648: Abstract database specific helper methods to PDO adapter

19649: Support database specific DDL in migrations

19650: Use sequences instead of auto increment keys for user facing identifiers

19651: Support optional sub-query substitution for COUNT(DISTINCT x) queries

sgiehl commented 2 years ago

@bx80 I haven't yet read all the other issues in detail, but would something like #18371 be a good base for starting TiDB support as well?

bx80 commented 2 years ago

@sgiehl This would be covered by issue #19648. My suggested approach there is almost identical to your solution for #18371, the only difference is you've implemented the abstraction using a db schema class whereas I was thinking to have statement code in a db specific PDO adapter and just have schema changes in the schema class.

PDO adapter descendants get access to override some lower level functions such as query() which might be needed if we implement a multi-tenancy solution at some point and need to enforce query safe guards. Also some of the existing MySQL PDO adapter code is already too MySQL specific to work with TiDB, SingleStore, etc, which don't have InnoDB tables at all, different default ports and different version numbering.

We could allow database choice during installation and then set the existing config.ini.php adapter and schema settings to determine the schema and PDO adapter classes to use.

It looks like #18371 is already compatible with this approach. Maybe it could be finished at the same time #19648 is implemented? It'd be a good test of the abstraction structure.

dankasak commented 1 year ago

Hi there. Just a quick note on #19650 - as of TiDB v6.4.x, TiDB has a high-performance, monotonically-incrementing AUTO_INCREMENT implementation, which can be activated on a per-table basis with the AUTO_ID_CACHE 1 directive, eg CREATE TABLE t(a int AUTO_INCREMENT key) AUTO_ID_CACHE 1 See: https://docs.pingcap.com/tidb/dev/auto-increment#mysql-compatibility-mode

Therefore, #19650 may no longer be required for TiDB support. Note however that tables with AUTO_INCREMENT primary keys are still prone to write ( and potentially read ) hotspots. Please feel free to reach out if you want to discuss further.

tsteur commented 1 year ago

This is great 🎉 Thanks @dankasak