MySQL Cluster Tutorial

MySQL Conference & Expo 2011

Max Mether <[email protected]>

Joffrey Michaie <[email protected]>
Johan Andersson <[email protected]>
Who are we?

• Max Mether
– Trainer and Consultant at MySQL from 2001
– Curriculum Manager at MySQL
– Training Manager at SkySQL from 2010
• Johan Andersson
– Cluster Practice Manager at MySQL from 2003
– Consultant at Severalnines from 2010
• Joffrey Michaie
– Cluster Consultant at MySQL from 2009
– Consultant at SkySQL from 2010

11.04.2011 SkySQL Ab 2011 Confidential 2

 Part 1
Introduction

11.04.2011 SkySQL Ab 2011 Confidential 3

Cluster Use Cases

• What is cluster used for?

– Telecom applications
– Online Gaming
– Financial Applications
– eCommerce
– Session Management

11.04.2011 SkySQL Ab 2011 Confidential 4

Cluster Usage

What are/will you using the cluster for??

11.04.2011 SkySQL Ab 2011 Confidential 5

Features

• Shared nothing architecture

– No single point of failure
• Synchronous replication between nodes
• ACID transactions
• Row level locking

11.04.2011 SkySQL Ab 2011 Confidential 6

Features

• In-memory storage
– Some data can be stored on disk
– Checkpointing to disk for durability
• Two types of indexes
– Ordered T-trees
– Unique hash indexes
• Online operations
– Add node groups
– Software upgrade
– Some table alterations

11.04.2011 SkySQL Ab 2011 Confidential 7

Architecture

11.04.2011 SkySQL Ab 2011 Confidential 8

Partitioning
Table

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential

Partitioning
Table

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential

Partitioning
Table

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential

Partitioning
Table

Primary Replica
Node 3 Node 4

Secondary Replica
11.04.2011 SkySQL Ab 2011 Confidential
Partitioning – 4 Data Nodes
Table

Node 3 Node 4 Node 5 Node 6

11.04.2011 SkySQL Ab 2011 Confidential 13

Partitioning – 4 Data Nodes
Table

Node Group Node Group

Node 3 Node 4 Node 5 Node 6

11.04.2011 SkySQL Ab 2011 Confidential 14

Heartbeat Circle
Node 3 Node 4

Node 6 Node 5

11.04.2011 SkySQL Ab 2011 Confidential 15

Heartbeat Circle
Node 3 Node 4

Node 6 Node 5

11.04.2011 SkySQL Ab 2011 Confidential 16

Heartbeat Circle
Node 3 Node 4

Node 6 Node 5

11.04.2011 SkySQL Ab 2011 Confidential 17

Heartbeat Circle
Node 3 Node 4

Node 6 Node 5

11.04.2011 SkySQL Ab 2011 Confidential 18

Network Partitioning Protocol

• The network partitioning protocol is designed

to avoid a split brain scenario:

1. Is there at least one node from each node

group?
2. Are all nodes present from any node group?
3. Ask the arbitrator

11.04.2011 SkySQL Ab 2011 Confidential 19

Uneven Split
Node 3 Node 4
Node Group

Node 6 Node Group Node 5

11.04.2011 SkySQL Ab 2011 Confidential 20

Even Split
Node 3 Node 4
Node Group

Node 6 Node Group Node 5

11.04.2011 SkySQL Ab 2011 Confidential 21

Durability

• In order for a node to recover faster some

data is stored locally
– The REDO log
• Synchronized by global checkpoints (GCP)
– The DataMemory
• Synchronized by local checkpoints (LCP)
• These can also be used for system recovery

11.04.2011 SkySQL Ab 2011 Confidential 22

Transactions

SQL Node
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 23

Transactions – Two Phase Commit
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 24

Transactions – Prepare Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 25

Transactions – Prepare Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 26

Transactions – Prepare Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 27

Transactions – Commit Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 28

Transactions – Commit Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 29

Transactions – Commit Phase
Transaction
SQL Node Coordinator
Transaction
request

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 30

Transactions – Commit Phase
Transaction
SQL Node Coordinator
Transaction
sucessful

Node 3 Node 4

11.04.2011 SkySQL Ab 2011 Confidential 31

Indexes

• Unique Hash Indexes

– Each table has a Primary Key hash index
– Other unique hash indexes implemented by
hidden tables
• Partitioned like tables
• Ordered indexes
– T-trees
– Local for each node

11.04.2011 SkySQL Ab 2011 Confidential 32

 Part 2
Practical Labs

11.04.2011 SkySQL Ab 2011 Confidential 33

Preparations

1. Load the virtualbox and start the system

2. Examine the Cluster configuration file
3. Start the cluster
– Start management node
– Tail the cluster log
– Start data nodes
– Start the MySQL servers
4. Load the sakila database
5. Start the MySQL clients

11.04.2011 SkySQL Ab 2011 Confidential 34

Exercise 1 – Initial test

1. Create a test table

2. Insert a row in the test table
3. Login to the other MySQL server and verify
that the table is there too
4. Drop the table

11.04.2011 SkySQL Ab 2011 Confidential 35

Exercise 2 – Backup and Restore

1. Take a backup
– Use the START BACKUP command
2. Start a clean cluster with no data
3. Use the backup to restore the data
– Use the ndb_restore utility

11.04.2011 SkySQL Ab 2011 Confidential 36

Exercise 3 – Node Recovery

1. Examine the log during the process

2. Kill one of the data nodes with the kill
command
3. Execute a query, is the cluster working?
4. Restart the node
5. Execute a query

11.04.2011 SkySQL Ab 2011 Confidential 37

Exercise 4 – NDBINFO

1. Go to the ndbinfo schema

2. Examine the tables

11.04.2011 SkySQL Ab 2011 Confidential 38

Exercise 5 – Resource Limits

1. Issue the statements that run into limits

2. Change the configuration file
– Set MaxNoOfConcurrentOperations to
40000
3. Do a ”rolling restart”
– Restart each node one by one
4. Re-issue the failed statement

11.04.2011 SkySQL Ab 2011 Confidential 39

Exercise 6 – Partial Restore

1. Restore one table from the backup

– Restore the table directly or
– Extract the contents to plainfile and import

11.04.2011 SkySQL Ab 2011 Confidential 40

Exercise 7 – Query Optimization

1. Run the Query

– Watch query time
2. Use EXPLAIN, show indexes
– Watch the query execution plan and cardinality
3. Rewrite the query
– Watch query time

11.04.2011 SkySQL Ab 2011 Confidential 41

 Part 3
Best Practices

11.04.2011 SkySQL Ab 2011 Confidential 42

Agenda
• Cluster Setup
– Recommended Setup
– Networking & Hardware Selection
• Disk Data Tables
• Configuration
• Administration
– Online/Offline Operations
– Backup and restore
• Monitoring
11.04.2011 SkySQL Ab 2011 Confidential 43
Recommended Setup

Clients

Load Balancer(s)

Redundant switches

SQL+Mgm SQL+Mgm
+AppServer +AppServer
+WebServer... +WebServer...

Bonding

Data node Data node

11.04.2011 SkySQL Ab 2011 Confidential 44

Networking
• Dedicated >= 1GB/s networking
• Prevent network failures (NIC x 2, Bonding)
• Use Low-latency networking (Dolphin...)
–Especially when>= 8 data nodes or want higher
throughput and lower latency
• No security layer to management node
(remote shutdown allowed ....)
• Enable port 1186 access only from cluster
nodes and administrators
11.04.2011 SkySQL Ab 2011 Confidential 45
Hardware – Data Nodes

• One data node can use 8 cores (Cluster 7.0+)

• CPU: 2 x 4 core (Nehalem works really well)
– Fast CPU → fast processing of messages
• RAM: As much as you need
– 10GB data set will require 20GB of RAM
– Each node will then need 2 x 10 / #data nodes
(For example 2 data nodes → 10GB → 16GB good)
• Disk: 10xDataMemory + space for BACKUP +
TableSpace (if disk data tables)
11.04.2011 SkySQL Ab 2011 Confidential 46
Hardware – MySQL Servers

• CPU: 2 – 16 cores
• RAM: Not so important – 4GB enough
(depends on connections and buffers)
• Disks: Used mainly for logging
– Binary log needed for replication

11.04.2011 SkySQL Ab 2011 Confidential 47

 Disk Subsystem
low-end mid-end high-end

LCP LCP LCP

REDOLOG REDOLOG REDOLOG

1 x SATA 7200RPM 1 x SAS 10KRPM 4 x SAS 10KRPM

• For a read-most, write • Heavy duty (many MB/s) • Heavy duty (many MB/s)
not so much • No redundancy • Disk redundancy (RAID1+0)
• No redundancy (but other data node is hot swap
(but other data node is the mirror)
the mirror)

• REDO, LCP, BACKUP – written sequentually in small chunks (256KB)

• If possible, use Odirect = 1

11.04.2011 SkySQL Ab 2011 Confidential 48

Filesystem

• Most customers uses EXT3(Linux) and UFS

(Solaris)
– Ext2 could be an option (but recovery is longer)
• XFS – we haven't experienced so much...
• ZFS
– You must separate journal (Zil) and filesystem
• Mount with noatime
• Raw device is not supported
11.04.2011 SkySQL Ab 2011 Confidential 49
Disk Data Storage
Minimal recommended high-end

LCP UNDOLOG
REDOLOG (REDO LOG)
UNDOLOG
TABLESPACE 1
TABLESPACE

TABLESPACE 2
2 x SAS 10KRPM (preferably)
(REDO LOG / UNDO LOG)
LCP

4 x SAS 10-15KRPM (preferably)

• Use High-end for heavy read write (1000's of 10KB records per sec) of data
(e.g Content Delivery platforms)
• SSD for TABLESPACE is also interesting – not much experience of this yet
• Having TABLESPACE on separate disk is good for read perf.
• Enable WRITE_CACHE on devices

11.04.2011 SkySQL Ab 2011 Confidential 50

Configuration – Disk Data Storage
• Use Disk Data tables for
– Simple accesses (read/write on PK)
– Same for innodb – easily DISK BOUND (iostat)
• Set
– DiskPageBufferMemory=3072M
• is a good start if you rely a lot on disk data – like the Innodb_Buffer_Pool,
but set it as high as you can!
• Increased chance that a page will be cached
– SharedGlobalMemory=384M-1024M
– UNDO_BUFFER=64M to 128M (if you write a lot)
• You cannot change this BUFFER later!
• Specified at LOGFILE GROUP creation time
– DiskIOThreadPool=[ 8 .. 16 ] (Cluster 7.0+)

11.04.2011 SkySQL Ab 2011 Confidential 51

Configuration - General

– MaxNoOfExecutionThreads<=#cores
• Contention can occur → unexpected behaviour
– RedoBuffer=32-64M
• If you need to set it higher your disks are too slow
– FragmentLogFileSize=256M
– NoOfFragmentLogFiles=
6 x DataMemory (in MB) / (4x 256MB)
• Most common issue – redo log too small
– Try the configurator:
www.severalnines.com/config
11.04.2011 SkySQL Ab 2011 Confidential 52
Application - Primary Keys

• Always define a primary key

– Tables without primary keys are accepted
• A hidden primary key is created
• The hidden PK is not replicated
• There are recovery issues with hidden PKs
• Application behavior (KEY NOT FOUND.. etc)
• At least have a
id BIGINT AUTO_INCREMENT PRIMARY KEY
– Even if you don't need it for your applications
11.04.2011 SkySQL Ab 2011 Confidential 53
Application - Query Cache

• Don't cache everything in the Query Cache

– Expensive to invalidate over N mysql servers
– A write on one server will force the others to
purge their cache
• For tables that change seldom (or read-only)
– Set query_cache_type=2 (DEMAND)
SELECT SQL_CACHE <cols> .. FROM
table;
– This can be good for STATIC data
11.04.2011 SkySQL Ab 2011 Confidential 54
Application – Transaction Size
• Transactions (large updates)
–NDB designed for many and short transactions
• Recommended to UPDATE / DELETE in small chunks
• Use LIMIT 10000 until all records are UPDATED/DELETED
• MaxNoOfConcurrentOperations
–Limit for how many records than can be modified
simultaneously on one data node
–MaxNoOfConcurrentOperations=1000000 will use
1GB of RAM
• Use only if necessary
11.04.2011 SkySQL Ab 2011 Confidential 55
Application – Table Locks

• Table lock commands are local only

–FLUSH TABLE WITH READ LOCK;
–LOCK TABLES <table> READ;
• You must get the lock on all mysql servers

11.04.2011 SkySQL Ab 2011 Confidential 56

Application – Schema Operations

• Don't use too much CREATE/DROP TABLE

of NDB tables
–It is a heavy operation within Cluster
–Takes much longer than with standard MySQL

11.04.2011 SkySQL Ab 2011 Confidential 57

REDO Log Optimizations

• Some tables account for a lot of writes, but do

not need to be recovered (session tables)
– A session table is often unnecessary to REDO LOG
and to CHECKPOINT
• Create these tables as 'NO LOGGING' tables:
mysql> set @ndb_curr_val=@@ndb_table_no_logging;
mysql> set ndb_table_no_logging=1;
mysql> create table session_table(..) engine=ndb;
mysql> set ndb_table_no_logging=@ndb_curr_val;

– session_table will not be REDO logged

→ No disk activity for this table!

11.04.2011 SkySQL Ab 2011 Confidential 58

ALTER TABLE
• NOT online operations:
• Rename a table
• Change data type
• Change storage size
• Drop column
• Rename column
• Add/Drop a PRIMARY KEY
• Online operations:
• Add column (ALTER ONLINE …)
• CREATE INDEX
• Online add node (see my presentation from last year how to do it)
• Altering a 1GB table offline requires 1GB extra
11.04.2011 SkySQL Ab 2011 Confidential 59
Administration Layer
• Introduce a MySQL Server for administration purposes!
• Should never ever get application requests
• Simplifies heavy (non online) schema changes

App layer

SQL layer

#give explicit nodeid in config.ini:

[mysqld]
id=8
hostname=X
Storage layer
# in my.cnf:
syncrepl ndb_connectstring=”nodeid=8;x,y”
ndb_cluster_connnection_pool=1

Admin layer

11.04.2011 SkySQL Ab 2011 Confidential 60

Online Upgrades

• OS, SW version (7.0.x → 7.1.x)

• Configuration
– increase DM, IM, Buffers, redo log, [mysqld] slots
• Hardware (upgrade more RAM etc)
• Adding data nodes (from 7.0)
– See Johan’s presentation from the conference
2009

11.04.2011 SkySQL Ab 2011 Confidential 61

Backup

• Backup of NDB tables

– Online – can have ongoing transactions
– Consistent – only committed data
– ndb_mgm -e “START BACKUP”
• Copy backup files from data nodes to safe
location
• Non-NDB tables must be backed up separately
• MySQL system tables are stored only in
MyISAM
11.04.2011 SkySQL Ab 2011 Confidential 62
Backup
• You want to backup (for each mysql server)
– mysql database
– Triggers, SP ...
• Use 'mysqldump‘
mysqldump mysql > mysql.sql
mysqldump --no-data
--no-create-info -R >
routines.sql
• Copy my.cnf & config.ini files
11.04.2011 SkySQL Ab 2011 Confidential 63
Monitoring

• Mandatory to monitor
– CPU/Network/Memory usage
– Disk capacity (I/O) usage
– Network latency between nodes
– Node status ...
– Used Index/Data Memory
• www.severalnines.com/cmon- monitors data
nodes and MySQL servers

11.04.2011 SkySQL Ab 2011 Confidential 64