Tag Archives: row-based

Why does MySQL replication fail?

When considering active-active multi-master, you must consider it’s foundation technology. Although MySQL replication is straightforward to setup, it can fail in a myriad of ways. Most of those are known and well understood. We can solve them only if we use the technology in the standard way.

Click through to the end for multi-master solutions that work with MySQL.

Reason 2 – Replication is brittle to start with

– statement based – throw it over the fence architecture
– use non-deterministic functions causes unpredictable behavior
– no globally unique transaction id (like Oracle SCN – system change number)
– no built in checksums – pt-table-checksum is a must
– replication position syncing to index files prone to breaking
– temp tables disappear after restart
– row-based still a new code path – doesn’t support zero downtime
– even with row-based mysql can fall back to statement
– row-based does not include SQL in binlogs
– MySQL replica slaves die frequently, require re-clone from authoritative master

NEXT: Reason 3 – MySQL row-based replication has limitations

PREV: Reason 1 – auto increment settings create new problems

Want more? Grab our Scalable Startups monthly for more tips and special content. Here’s a sample