Amazon Web Services

• CloudFormation
• Terraform
• Manual deployment

This page documents the manual deployment of YugabyteDB on six AWS EC2 instances with c5d.4xlarge as the instance type and CentOS 7 as the instance operating system. The deployment is configured for multiple availability zones (multi-AZ), with three AZs, and has a replication factor (RF) of 3. The configuration can be easily changed to handle single-AZ as well as multi-region deployments.

1. Prerequisites

Create AWS EC2 instances

Create AWS EC2 instances with the following characteristics.

• Virtual Machines: Spin up 6 (minimum 3 if you are using RF=3). Given that this is a 3-AZ deployment, a multiple of 3 is preferred.

• Operating System: CentOS 7 VMs of above type. You can use Ubuntu as well, but then some of the specific steps in terms of setting up ulimits etc. could be slightly different.

• Ports: Make sure to bring these VMs up in a Security Group where communication between instances is enabled on these ports (and not locked down by security settings). For a listing of these ports, see Default ports reference.

We now have 2 VMs each in Availability Zones us-west-2a, us-west-2b, us-west-2c respectively.

Set environment variables

Now that the six nodes have been prepared, the yb-master process will be run on three of these nodes (because RF=3) and yb-tserver will be run on all six nodes. To learn more about YugabyteDB’s server architecture, see here.

These install steps are written in a way that assumes that you will run the install steps from another node from which you can access the above six VMs over ssh.

These are some handy environment variables you can set on the node from which you are planning to do the install of the software on the six nodes.

Prepare data drives

If your AMI already has the needed hooks for mounting the devices as directories in some well defined location OR if are just trying to use a vanilla directory as the data drive for a quick experiment and do not need mounting the additional devices on your AWS volume, you can just use an arbitrary directory (like /home/$USER/ as your data directory), and YugabyteDB will create a yb-data subdirectory there (/home/$USER/yb-data) and use that. The steps below are simply a guide to help use the additional volumes (install a filesystem on those volumes and mount them in some well defined location so that they can be used as data directories by YugabyteDB).

Locate drives

On each of those nodes, first locate the SSD devices to be used as the data directories for YugabyteDB to store data on (such as RAFT/txn logs, SSTable files, logs, etc.).

NAME        MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
nvme0n1     259:1    0    40G  0 disk
└─nvme0n1p1 259:2    0    40G  0 part /
nvme1n1     259:0    0 372.5G  0 disk

OR

Notice that the 370G partition is on nvme1n1, but its MOUNTPOINT column is empty - meaning that it has not been mounted. We should prepare this drive for use by putting a reasonable filesystem on it and mounting it in some well defined location.

Create file system

Create xfs file system on those devices. The filesystem on the drives do not have to be XFS. It could be ext4 also, for instance. But we have primarily tested with xfs.

You can run this command on each node OR use the sample loop below.

OR

Verify the file system.

The above should print “xfs” for each of the nodes or drives.

Configure Drives

Add /etc/fstab entry to mount the drives on each of the nodes. This example assumes there’s one drive that you will mount at the /mnt/d0 location.

Verify that the file has the expected new entries.

Mount Drives

Mount the drive(s) at /mnt/d0 path. Note that the /mnt/d0 is just a sample path. You can use a different location if you prefer. These paths become the --fs_data_dirs argument to yb-master and yb-tserver processes in later steps.

Verify that the drives were mounted and are of expected size.

Verify system configuration

Below is an example of setting up these prerequisites in CentOS 7 or RHEL. For Ubuntu, the specific steps could be slightly different. For details, see System configuration.

Install ntp and other optional packages

OPTIONAL: You can install a few more helpful packages (for tools like perf, iostat, netstat, links)

Set ulimits

To ensure proper ulimit settings needed for YugabyteDB, add these lines to /etc/security/limits.conf (or appropriate location based on your OS).

*       -       core    unlimited
*       -       nofile  1048576
*       -       nproc   12000

Sample commands to set the above on all nodes.

Make sure the above is not overridden by files in limits.d directory. For example, if 20-nproc.conf file on the nodes has a different value, then update the file as below.

*          soft    nproc     12000
root       soft    nproc     unlimited

A sample command to set the above on all nodes.

Verify the settings

The values should be along the lines of:

open files                      (-n) 1048576
max user processes              (-u) 12000
core file size          (blocks, -c) unlimited

2. Install YugabyteDB

Note: The installation need NOT be undertaken by the root or the ADMIN_USER (centos). In the examples below, however, these commands are run as the ADMIN_USER.

Create yb-software & yb-conf directory in a directory of your choice. In this example, you use the ADMIN_USER home directory.

Download the YugabyteDB package, untar and run the post-install script to patch relative paths on all nodes.

Create ~/master & ~/tserver directories as symlinks to installed software directory.

Execute the following on master nodes only.

Execute the following on all nodes.

The advantage of using symbolic links (symlinks) is that, when you later need to do a rolling software upgrade, you can upgrade YB-Master and YB-TServer servers one at a time by stopping the YB-Master server, switching the link to the new release, and starting the YB-Master server. Then, do the same for YB-TServer servers.

3. Prepare YB-Master configuration files

This step prepares the configuration files for the three masters. The configuration files need to, among other things, have the right information to indicate which cloud, region, or availability zone each master is in.

Create YB-Master1 configuration file

Create YB-Master2 configuration file

Create YB-Master3 configuration file

Verify

Verify that all the configuration variables look right and environment variables were substituted correctly.

4. Prepare YB-TServer configuration files

Create config file for AZ1 yb-tserver nodes

Create configuration file for AZ2 YB-TServer servers

Create configuration file for AZ3 YB-TServer servers

Verify

Verify that all the configuration flags look correct and environment variables were substituted correctly.

5. Start YB-Master servers

Note: On the first time that all three YB-Master servers are started, it creates the cluster. If a YB-Master server is restarted (after cluster has been created), such as during a rolling upgrade of software, it simply rejoins the cluster.

Verify

Verify that the YB-Master servers are running.

Check the YB-Master UI by going to any of the 3 YB-Master servers.

http://<any-master-ip>:7000/

You can do so using a character mode browser (such as links for example). Try the following.

Troubleshooting

Make sure all the ports detailed in the earlier section are opened up. Else, check the log at /mnt/d0/yb-master.out for stdout or stderr output from the YB-Master server. Also, check INFO/WARNING/ERROR/FATAL glogs output by the process in the /mnt/d0/yb-data/master/logs/*

6. Start YB-TServer servers

After starting all the YB-Master servers in the previous step, start YB-TServer servers on all the nodes.

Verify that the YB-TServer servers are running.

7. Configure AZ- and region-aware placement

Note: This example is a multi-AZ (single region deployment).

The default replica placement policy when the cluster is first created is to treat all nodes as equal irrespective of the placement_* configuration flags. However, for the current deployment, you want to explicitly place 1 replica in each AZ. The following command sets replication factor of 3 across us-west-2a, us-west-2b and us-west-2c leading to the placement of 1 replica in each AZ.

Verify by running the following.

And confirm that the output looks similar to what is shown below with min_num_replicas set to 1 for each AZ.

replication_info {
  live_replicas {
    num_replicas: 3
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2a"
      }
      min_num_replicas: 1
    }
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2b"
      }
      min_num_replicas: 1
    }
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2c"
      }
      min_num_replicas: 1
    }
  }
}

Suppose your deployment is multi-region rather than multi-zone, one additional option to consider is to set a preferred location for all the tablet leaders using the yb-admin set_preferred_zones command. For multi-row or multi-table transactional operations, colocating the leaders to be in a single zone/region can help reduce the number of cross-region network hops involved in executing the transaction and as a result improve performance.

The following command sets the preferred zone to aws.us-west.us-west-2c:

Looking again at the cluster config you should see affinitized_leaders added:

replication_info {
  live_replicas {
    num_replicas: 3
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2a"
      }
      min_num_replicas: 1
    }
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2b"
      }
      min_num_replicas: 1
    }
    placement_blocks {
      cloud_info {
        placement_cloud: "aws"
        placement_region: "us-west"
        placement_zone: "us-west-2c"
      }
      min_num_replicas: 1
    }
  }
  affinitized_leaders {
    placement_cloud: "aws"
    placement_region: "us-west"
    placement_zone: "us-west-2c"
  }
}

8. Test PostgreSQL-compatible YSQL API

Connect to the cluster using the YSQL shell (ysqlsh) that is installed in the bin directory. If you want to use ysqlsh from a different node, follow the steps on the ysqlsh page.

From any node, execute the following command.

Output should be the following:

9. Test Cassandra-compatible YCQL API

Using ycqlsh

Connect to the cluster using the YCQL shell (ycqlsh) that comes installed in the bin directory. If you want to use ycqlsh from a different node, follow the steps found on the ycqlsh page.

From any node, execute the following command.

Output should be the following.

 userid | action_id | payload
--------+-----------+---------
      1 |         4 |       d
      1 |         3 |       c

Running sample workload

If you want to try the pre-bundled yb-sample-apps.jar for some sample apps, you can either use a separate load tester machine (recommended) or use one of the nodes itself.

First install Java on the machine.

Set the environment variable for the YCQL endpoint.

Here's how to run a workload with 100M key values.

% cd ~/tserver/java
% java -jar yb-sample-apps.jar --workload CassandraKeyValue --nodes $CIP_ADDR -num_threads_read 4 -num_threads_write 32 --num_unique_keys 100000000 --num_writes 100000000 --nouuid

Here's how to run a workload with 100M records with a unique secondary index.

% cd ~/tserver/java
% java -jar yb-sample-apps.jar --workload CassandraUniqueSecondaryIndex --nodes $CIP_ADDR -num_threads_read 1 -num_threads_write 16 --num_unique_keys 100000000 --num_writes 100000000 --nouuid

When workload is running, verify activity across various tablet-servers in the Master’s UI:

http://<master-ip>:7000/tablet-servers

When workload is running, verify active YCQL or YEDIS RPC calls from this links on the “utilz” page.

http://<any-tserver-ip>:9000/utilz

10. Test Redis-compatible YEDIS API

Prerequisite

Create the YugabyteDB system_redis.redis (which is the default Redis database 0) table using yb-admin or using redis-cli.

• Using yb-admin

• Using redis-cli (which comes pre-bundled in the bin directory)

Test API

11. Stop cluster and delete data

The following commands can be used to stop the cluster as well as delete the data directories.