Crash-safe semisync replication

Abstract

This presentation dwells into recent improvements made to the semisync replication recovery. Namely, when a crashed former semisync master recovers from its binary log the latter may contain events for transactions that were prepared by the storage engine and written to the binary log, but that were never actually received by a slave server (therefore) neither committed by the storage engine. According the general recovery rules the crashed master server was used to commit those transactions and thus become inconsistent with its former slaves should those servers meanwhile have switched over to a new master.

The solution presented addresses that through refining the server recovery in such a way that when the old master is restarted in the semisync slave mode, past recovery, its binlog will not have any extra transaction.

Other links

Date and time

  • Tuesday 5 October, 18.20 – 18.50 CEST (UTC +2), 12.20pm – 12:50pm New York time, 00:20 – 00:50 Beijing/Singapore time

Presenters

Andrei Elkin
Leads MariaDB replication at MariaDB Corporation

Andrei Elkin, Espoo Finland, leads the MariaDB replication. He holds PhD in math and
physics from the Saint-Peterstung state university (1997), and since
then has been working with database replication and clustering
technologies.


Sujatha Sivakumar
Software developer at MariaDB Corporation

Sujatha Sivakumar, Bangalor India, is an experienced engineer who joint
MariaDB in 2019 having almost a decade of MySQL Server development
and maintenance background.