MariaDB Galera Cluster presentation

HA&Solu4ons&for&MySQL&

Joﬀrey&Michaie&
Consultant&

SkySQL&AB&©&2013&Conﬁden4al&

Agenda&
• Introduc4on&
• MySQL&Replica4on&

• SemiHSynchronous&Replica4on&
• MySQL&5.6&
• MariaDB&10.0&

• MHA&
• Galera&
• Shared&Disk&
• DRBD&
• MySQL&Cluster&

Introduc4on&to&HA&

“High&availability&is&a&system&design&protocol&and&
associated&implementa6on&that&ensures&a&certain&
degree&of&opera6onal&con6nuity&during&a&given&
measurement&period”&


Up4me,&Down4me,&9s&
Availability&=&up4me&/&(up4me&+&down4me)&
&

&90%& &

&

&

&

&1&nine&&

&

&

&36.5&days&/&year &&

&

&99%& &

&

&

&

&2&nines& &

&

&3.65&days&/&year&

&

&99.9%&&

&

&

&

&3&nines& &

&

&8.76&hours&/&year&

&

&99.99%& &

&

&

&4&nines& &

&

&52&minutes&/&year&

&

&99.999%& &

&

&

&5&nines& &

&

&5&minutes&/&year&

&

&99.9999%&&

&

&

&6&nines& &

&

&31&seconds&/&year&

Availability&=&MTBF&/&(MTBF&+&MTTR)&

Terminology&
• Synchronous&vs.&Asynchronous&
• SharedHDisk&vs.&SharedHNothing&vs.&SharedHMemory&
• Single&Point&Of&Failure&H&SPOF&
• Failover&vs.&Switchover&
• Split&Brain&
• Node&Fencing,&STONITH,&Quorum&


Designing&for&HA&
• Which&level&of&availability&do&I&need?&
• How&many&nines?&

• Do&I&require&no&loss&of&data?&

• Could&I&loose&some&transac4ons?&
• Will&my&users&no4ce&or&care?&

• Do&I&need&automa4c&failover&or&is&manual&switchover&ok?&
• How&do&I&test&this?&

• Can&I&provide&a&reasonable&service&when&X&is&down?&
• Replace&X&with&each&component&of&the&service&


Before&we&talk&about&solu4ons&
• A&high&availability&setup&does&not&
replace&backups&

• Check&parameters&

• ﬂush_at_trx_commit&
• sync_binlog&
• expire_logs_day&
• binlog_format&
• sync_master_info&(on&slaves)&
• ...&

• Requirements&for&parameters&may&
change&if&using&baieryHbacked&disk&
cache&


MySQL&Replica4on&H&Asynchronous&
• Asynchronous:&3&Phases&

1.  Commit&and&write&to&binlog&on&Master&
2.  Ship&changes&to&relay&log&on&slave&
3.  Apply&changes&on&slave&&

• Master&H>&slave&rela4onship&
• MonoHthreaded&on&slaves&un4l&MySQL&5.5&
• MySQL&5.6&allows&mul4Hthreaded&

• No&conﬂict&resolu4on&

• MasterHmaster&replica4on&or&circular&
replica4on&need&applica4on&logic&

&

Slave(s)&
Master&

MySQL&Replica4on&–&SemiHSynchronous&
• Added&as&a&plugin&in&MySQL®&5.5&
• Ensures&that&changes&have&been&shipped&to&
at&least&one&slave&(or&4meouts)&&

• A&COMMIT&on&the&master&waits&for&ONE&
Slave&to&acknowledge&the&transac4on&

• Important:&The&Master&does&not&wait&for&the&

Slave&to&execute&the&transac4on,&only&to&write&
it&to&the&relay&log&
• So&the&Slave&SQL&Thread&may&s4ll&lag&behind&
the&Master&and&queries&to&the&Slave&may&s4ll&
return&old&data&

• Poten4ally&adds&latency&to&queries&

Slave(s)&
Master&

MySQL&Replica4on&for&HA&
• Master&<H>&standby&master&H&manual&failover&
• Minimal&down4me&for&changes&and&upgrades&
• SemiHsynchronous&should&be&used&
• Used&in&combina4on&with&other&HA&solu4ons&
for&geographical&replica4on&


MySQL®&5.6&–&New&Replica4on&Features&
• Global&Transac4on&ID&(GTID)&

• Makes&it&easy&to&automate&failover&and&slave&promo4on&

• Replica4on&failover&and&admin&u4li4es&
• Mul4Hthreaded&slaves&
• Replica4on&event&checksums&
• TimeHdelayed&replica4on&


MariaDB&10.0&–&New&Replica4on&Features&
• Mul4Hsource&replica4on&
• Global&Transac4on&ID&(GTID)&

• Diﬀerent&implementa4on&from&MySQL&5.6&
• GTID&per&domain&instead&of&server&


MHA&

Master&High&Availability&Manager&

• Automates&master&failover&and&slave&
promo4on&

• Monitors&the&master&or&can&integrate&with&
Pacemaker/Heartbeat&
• Failover&is&an&online&opera4on&
• Also&allows&manual&switchover&

• Short&down4me:&open&a&few&seconds&
• MySQLHReplica4on&consistency&
• No&performance&penalty&
• Drop&in&solu4on&on&exis4ng&deployment&
&

MHA&

Failover&Process&

1.  Aiempt&to&contact&MySQL&master&server&
by&SSH&

2.  If&master&server&is&alive&access&the&binary&
log&and&recover&events&

3.  Find&the&slave&with&the&most&advanced&
relay&log&

4.  Sync&all&slaves&to&the&latest&available&
binlog&event&

5.  STONITH&master&if&necessary&
6.  Promote&slave&to&master&

Galera&/&MariaDB&Galera&Cluster&
• Provides&“virtually”&synchronous&

Clients

replica4on&

• Works&with&InnoDB&
• No&slave&lag&
• Transac4ons&are&validated&by&slaves&
upon&commit&

• Cer4ﬁca4on&and&quorum&
• Transac4ons&may&be&rolled&back&
at&this&stage&

• Mul4Hmaster&or&masterHslave&possible&

MySQL

MySQL

MySQL

wsrep

wsrep

wsrep

Galera Replication

Galera&–&Transac4on&Process&
• Transac4on&is&processed&
locally&up&to&commit&4me&

Client

trx

MySQL

MySQL

MySQL

wsrep

wsrep

wsrep

Galera Replication




Client

• Transac4on&is&replicated&
to&whole&cluster&

trx

MySQL

MySQL

MySQL

wsrep

wsrep

wsrep

ws

Galera Replication


ws



Client

to&whole&cluster&

• Client&gets&OK&status&

trx

MySQL

MySQL

MySQL

wsrep

wsrep

wsrep

Galera Replication




Client

to&whole&cluster&

• Client&gets&OK&status&
• Transac4on&is&applied&in&
slaves&

MySQL

MySQL
trx

wsrep

wsrep
Galera Replication


MySQL
trx

wsrep

SharedHdisk&Solu4on&
• Ac4ve&–&Passive&replica4on&

• Failover&requires&MySQL&crash&recovery&(and&
open&ﬁle&system&crash&recovery)&

• Combined&with&Pacemaker/Heartbeat&for&
automa4c&failover&

• Virtual&IP&most&open&used&to&fail&over&

• In&theory&the&SAN&is&a&SPOF&


DRBD®&&
• Synchronous&replica4on&(three&modes)&
• Ac4ve&–&Passive&replica4on&

• Failover&requires&MySQL&crash&recovery&(and&
open&ﬁle&system&crash&recovery)&

• Combined&with&Pacemaker/Heartbeat&for&
automa4c&failover&

• Virtual&IP&most&open&used&to&fail&over&

• STONITH&or&other&fencing&mechanism&
needed&to&avoid&splitHbrain&scenarios&

• Available&on&Linux&

DRBD®&H&Architecture&


MySQL&Cluster&H&Features&
• Synchronous&replica4on&between&
nodes&

• Through&TwoHPhase&Commit&Protocol&

• ACID&transac4ons&
• Row&level&locking&
• Shared&nothing&architecture&
• No&single&point&of&failure&

• Automa4c&failover&


• InHmemory&storage&

• Some&data&can&be&stored&on&disk&
• Checkpoin4ng&to&disk&for&durability&

• Two&types&of&indexes&

• Ordered&THtrees&
• Unique&hash&indexes&

• Online&opera4ons&

• Add&node&groups&
• Sopware&upgrade&
• Some&table&altera4ons&

MySQL&Cluster&H&Architecture&


MySQL&Cluster&H&Par44oning&


MySQL&Cluster&

Network&par44oning&protocol&
Designed&to&avoid&splitHbrain&

1.  Is&at&least&one&node&from&each&node&group&present?&&
• If&not&then&the&cluster&cannot&con4nue&H&shutdown&

2.  Are&all&nodes&present&from&any&node&group?&
• If&so&then&this&is&the&only&viable&cluster&H&con4nue&

3.  Ask&the&arbitrator&

• The&arbitrator&decides&which&”cluster”&con4nues&
• If&the&arbitrator&is&not&available&the&cluster&will&shutdown&


Where&does&MySQL&Cluster&ﬁt?&
• High&demands&on&availability&(5&nines)&
• You&need&write&scalability&
• You&have&3&or&more&”machines”&available&
• Where&the&queries&and&data&model&are&simple&
• When&the&data&ﬁts&in&memory&
• When&you&have&skilled&people&


Geographical&Replica4on&
• For&geographical&(mul4Hsite)&redundacy&
synchronous&solu4ons&are&open&not&
desirable&

• Standard&Replica4on&can&be&combined&
with&any&synchronous&solu4on&locally&

• MySQL&Cluster&
• DRBD&/&shared&disk&
• Galera&


Geographical&Replica4on&
MySQL&Cluster&
• MySQL&Cluster&has&addi4onal&features&
related&to&Geographical&replica4on&

• Conﬂict&detec4on&and&automa4c&
resolu4on&

• Several&possible&resolu4on&methods&exist&

• Mul4ple&replica4on&channels&possible&
• Mul4Hsource&replica4on&possible&
• Binlog&injec4on&ensures&the&consitency&
of&binlogs&


Quick&Comparison&Chart&

*


Quick&Comparison&Chart&


Ques4ons?&


Thank&you!&

Joﬀrey&Michaie&
Joﬀrey.michaie@skysql.com&
SkySQL&Ab&
www.skysql.com&
www.facebook.com/skysql&
www.linkedin.com/company/skysql&
MySQL is a registered trademark of Oracle and/or its affiliates. MariaDB is a registered trademark of Monty Program
Ab.
SkySQL and the SkySQL logo are trademarks of SkySQL Inc. or SkySQL Ab. SkySQL is not affiliated with MySQL. All
other company and product names may be trademarks or service marks of their respective owners.


MariaDB Galera Cluster presentation

Recommended

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to MariaDB Galera Cluster presentation (20)

More from Francisco Gonçalves (20)

Recently uploaded (20)

MariaDB Galera Cluster presentation