Galera Cluster is an open-source database cluster that provides high availability, active-active multi-master replication, and automatic failover. It allows for synchronous replication across nodes, automatic membership control, and guarantees write consistency. Performance benchmarks show Galera Cluster provides much higher throughput for reads and writes compared to traditional master-slave replication, with lower response times even under high concurrency.

1. INTRODUCTION TO GALERA CLUSTER
-GARIMA MASAND

2. SHORTCOMINGS OF MYSQL REPLICATION
Asynchronous: may result in data loss
Single Threaded: slave lag, performance
penalty
Master-Slave: read-write segregation on
application level
Master-Master: inconsistent data in case of
failures/no automated failover
High Availability: manual intervention
required for traffic shifting

3. HOW CAN WE OVERCOME THEM?

4. GALERA CLUSTER CAN HELP US

5. WHAT IS GALERA?
Galera
DB1 DB3DB2
Galera is a brainchild of Codership, open source
database cluster known for its auto healing, HA ,
distributed, scalable properties.
Compatible with MySQL, MariaDB and Percona
XtraDB
True parallel replication, on row level
Active-active multi-master topology

6. FEATURES OF GALERA CLUSTER

7. Synchronous replication
Read and write to any cluster node
Automatic membership control
Automatic node joining
True parallel replication, on row level

8. ADVANTAGES OVER MYSQL REPLICATION
No slave lag
Read and write scalability
Less manual effort to recover in case of
failures
Guaranteed write consistency

9. HOW IT WORKS?
GTID: Combination of state UUID and ordinal
sequence number
Certification based replication
Write-Set rep API
SST-IST: StateSnapshotTransfer /
IncrementalStateTransfer

10. SAMPLE GALERA CLUSTER SETUP
Galera Cluster
DB1 DB3DB2
HAPROXY/Load Balancer

11. MULTI MASTER REPLICATION VS GALERA
CLUSTER
Parameters Master-Master Galera Cluster
DDL
Bring instance out of loop and
execute DDLs Total order isolation, rolling schema upgrade
Auto_Increment
Based on no. of instances
configured in master-master
Auto increment values are not sequential for
avoiding conflicts values may have gaps
Automatic
Failover No automatic failover Auto failover
Backups
Backups can be taken from
dedicated backup slave STATE snapshot transfer as backup
Table Level LocksAvailable Table level locks are not supported
Functions
All functions supported but may
cause data discrepancy in case of
non-deterministic functions
Non-Deterministic functions like now() are
not supported

12. MIGRATING FROM M-M TO GALERA
InnoDB Engine
No explicit table locks are allowed
Mandatory primary key/delete is not allowed
No application login based on
auto_increment values
Conflicting updates should be sent to one
node

13. PERFORMANCE BENCHMARK GALERA CLUSTER VS
MULTI MASTER REPLICATION
Configuration used for Benchmarking
Parameter Galera M-M Replication
No of Nodes 3 2
Memory(Innodb) 122M 122M
Slave Threads 8 8
SST Method Rsync NA
No of Tables 10 10
No of Rows Per Table 2000000 2000000
Time Taken 60 secs for each iteration 60 secs for each iteration
*One transaction consists of 14 selects, 2 updates, 1 delete and 1 insert

14. Reads
Concurrent Threads Read Queries (Galera) Read Queries (M-M)
1 112602 21840
2 138068 27846
4 150234 36792
8 152054 49518
16 154518 68642
32 155988 77350
64 155624 89152
128 154672 97916
256 160930 100128
512 158438 105840

15. Writes
Concurrent Threads Write Queries(Galera) Write Queries(M-M)
1 32172 6240
2 39448 7956
4 42924 10512
8 43444 14148
16 44148 19612
32 44568 22100
64 44464 25472
128 44191 27976
256 45979 28608
512 45268 30240

16. Transactions Per Second
Concurrent Threads TransactionsG/sec TransactionsM/sec
1 133 26
2 164 33
4 178 43
8 179 58
16 183 58
32 184 81
64 182 105
128 180 114
256 187 116
512 181 120

17. Read/Write Requests Per Second
Concurrent Threads Read/Write RequestsG/sec Read/Write RequestsM/sec
1 2410 467
2 2955 595
4 3207 787
8 3233 1056
16 3305 1465
32 3317 1652
64 3294 1901
128 3253 2066
256 3381 2102
512 3285 2171

18. Response Time in milliseconds
Threads Response TimeG(Milisecond) Response TimeM(Milisecond)
1 7.46 38.46
2 12.18 60.38
4 22.41 91.38
8 44.47 135.89
16 87.08 196.27
32 173.31 348.11
64 347.54 604.31
128 707.11 1108.43
256 1345.56 2172.53
512 2793.23 4168.38

19. Thanks