LLM WikiAccess-protected knowledge portal

WIKI

[RNBK|INFR] DB engine comparison — MySQL vs PostgreSQL vs MariaDB (live-schema measurements)

AI Summary Purpose Preserve the 2026 07 09 read only live schema measurement study DAT 3313/3314 comparing three engines around partitioning, authored by 김현욱. Key points Measured facts across 4 instances 수집/dist main/dist source/prod partit

경로ai/sources/confluence/2026-07-17-db-engine-comparison.md
카테고리Source
태그#comparison #confluence #engine #kubernetes #mysql #portfolio #security #source #sources #study

# [RNBK|INFR] DB engine comparison — MySQL vs PostgreSQL vs MariaDB (live-schema measurements)

AI Summary

Purpose:

comparing three engines around partitioning, authored by 김현욱.

Key points:

partitioning = 0 tables anywhere; largest tables are billions of rows as single tables — e.g. TB_COMP_FILE_MAP_RAW_V6 6.67B rows/914GB, TB_COMP_FILE_MAP_V6_1 3.4B rows/879GB with INDEX 701GB > DATA 178GB.

composite PK (e.g. 4-column string PK on TB_COMP_LIB_VERSION), so every secondary index carries it. PostgreSQL uses a 6-byte ctid instead.

tables; PostgreSQL keeps FK (11+), has pg_partman/pg_cron automation, BRIN for append-only maps, and DETACH PARTITION CONCURRENTLY for non-blocking purge. Common constraint all three: partition key must be in every unique key/PK — our biggest table's PK(MAP_ID) and UN(OSS_ID,TAG_ID,FILE_ID) share no column, so key redesign precedes any partitioning regardless of engine.

clustered PK), code migration ≈ 2-3 weeks (FULLTEXT 8 spots, 58 views, JSON operators) + 3TB data move; MariaDB is a near-zero "scheme swap" for the license problem.

cost; if GPL is fully banned or structural improvement is an agenda → PostgreSQL port brings timestamptz/jsonb/pg_trgm/BRIN as by-products.

Relevant when:

("billions-of-rows single tables, index-larger-than-data root cause, 2038").

Do not read full document unless:

Linked documents:

Open Questions