MariaDB Core Architecture

How mysqld accepts a connection, parses a query, picks a plan, and reads pages from disk. Read this skill before tuning, before diagnosing slow queries, and before assuming MySQL behavior applies to MariaDB.

Quick Reference

• MariaDB = single multi-threaded mysqld process with per-connection threads or a built-in thread-pool plug-in.
• Query pipeline : parser (lex + yacc) : optimizer (cost-based, optimizer_switch flags) : executor : storage-engine handler API.
• Storage engines are plug-ins ; choice happens at CREATE TABLE ... ENGINE= time. InnoDB is the default ; Aria, MyISAM, MEMORY, MyRocks, Spider, S3, Archive, CSV, Sequence, MERGE are pluggable.
• In-memory caches per engine ; biggest is the InnoDB buffer pool, sized 60-80% of RAM on a dedicated DB host.
• Diverges from MySQL : different optimizer_switch flags (e.g. condition_pushdown_for_subquery, rowid_filter, split_materialized), built-in thread-pool (not Enterprise-only), query cache kept (deprecated, off by default), GTID format incompatible, JSON is LONGTEXT alias.
• ALWAYS run EXPLAIN before tuning : optimizer choices are visible in the plan, not in the schema.
• NEVER set innodb_buffer_pool_size blindly to 50% RAM on a dedicated DB host. Target 60-80%.
• NEVER enable every flag in optimizer_switch. Several are OFF by default (e.g. mrr, mrr_cost_based, index_merge_sort_intersection) because the cost model rejects them in common cases.

Process Model

mysqld is a single OS process with these thread roles :