Optimize Ambari Performance for Large Clusters

[#1]

Whether you're creating an Ambari cluster from scratch, taking over an existing cluster, or growing your cluster over time, it is imperative to tune Ambari and MySQL to work at a large scale of 1000-3000 Ambari Agents.

Ambari Server Configs

First, increase the memory used by Ambari. For large clusters, 8 GB of memory should be sufficient. If you have more than 10 concurrent users, increase it to 16 GB.

Edit /var/lib/ambari-server/ambari-env.sh and change the -Xmn setting.

export AMBARI_JVM_ARGS=$AMBARI_JVM_ARGS' -Xms2048m -Xmx8192m

Edit /etc/ambari-server/conf/ambari.properties with the following configs

# The size of the Jetty connection pool used for handling incoming Ambari Agent requests.
# 10 hosts => 25
# 50 hosts => 35
# 100 hosts => 75
# 500 hosts => 100	
agent.threadpool.size.max=100
# Determines whether current alerts should be cached.
# Enabling this can increase performance on large cluster, but can also result in lost alert data
# if the cache is not flushed frequently.
alerts.cache.enabled=true
# The size of the alert cache.
# Less than 50 hosts => 50000
# More than 50 hosts => 100000
alerts.cache.size=100000
# The number of threads used to handle alerts received from the Ambari Agents.
# The value should be increased as the size of the cluster increases.
# Less than 50 hosts => 2
# More than 50 hosts => 4
alerts.execution.scheduler.maxThreads=4

After performing these changes, restart Ambari Server.

Move an existing Ambari DB from a spinning disk to a SSD

It is highly suggested to use a Solid State Drive for the Ambari Database since this will be much faster.

Check the throughput of the disk in which Ambari’s database (Postgres, MySQL, MariaDB, or Oracle) is on.Ideally, it should be a Solid State Drive or support at least 200 IOPS and be either on the same host as Ambari or only a 1-2 hops away.

Type	Details	IOPS	Throughput
HDD	10,000 rpm SAS drive	175-210	100 MB/s
SSD	solid-state	500+	500+ MB/s

1. ambari-server stop

2. Take a backup of the Ambari database,

		mysqldump -u root ambari > /tmp/ambari.sql

3. Stop MySQL server, copy its data, and change the directory.

service mysqld stop
cp -R -p /var/lib/mysql /mnt/disks/ssd/mysql
cat /etc/my.cnf
sed -ie 's/\/var\/lib\/mysql/\/mnt\/disks\/ssd\/mysql/g' /etc/my.cnf

4. Create symlink for sock file and start MySQL

ln -s /mnt/disks/ssd/mysql/mysql.sock /var/lib/mysql/mysql.sock
service mysqldstart

5. Ensure Ambari DB is accessible.

mysql -u root -p
show databases;
use ambari;
show tables;
select count(*) from hosts;

MySQL Optimizations

First and foremost, if you're on an older version of MySQL, you can try to update it to MySQL 5.6 or 5.7, which has a lot of performance improvements.

Connect to the MySQL DB and inspect these variables. E.g.,

SHOW VARIABLES LIKE 'name';These suggested values assume that only Ambari Database’s is on the MySQL Server.If you have other databases in the same MySQL Server, increment by these values.

WARNING: Never stop MySQL server while Ambari Server is running.

Variable	Suggested Value
innodb_log_buffer_size	512M
innodb_buffer_pool_size	16G
innodb_file_io_threads (deprecated in MySQL 5.5)	16
innodb_log_file_size	5M
innodb_thread_concurrency	32
join_buffer_size	512M
key_buffer_size	16G
max_connections	500
max_allowed_packet	1024M
max_heap_table_size	64M
query_cache_limit	16M
query_cache_size	512M
read_rnd_buffer_size	128M
sort_buffer_size	128M
table_open_cache	1024
thread_cache_size	128
thread_stack	256K

To change these values.1. Stop MySQL: service mysqld stop

2. Edit the configs in /etc/my.cnf , under the “[mysqld]” section (note, it may be in a different location).

3. Start MySQL: service mysqld start