Change cluster configuration

Sometimes there might be a need to move a YugabyteDB universe deployed on a set of nodes to a completely different set of nodes in an online manner. Some scenarios that require such a cluster change are:

  • changing the instance or machine type.
  • updating the instance images on a regular basis (for example, AMI rehydration in AWS).
  • moving to a different set of zones, regions, or data centers.

This page provides the steps needed to perform such a data move in an online manner from the initial setup to the final setup as described below. This tutorial assumes that you are familiar with the YugabyteDB process architecture.

Example scenario

Initial configuration

We will assume the following about the initial setup of the universe:

  • Universe has six nodes: node1, node2, node3, node4, node5 and node6, each representing the IP address of a machine/VM.
  • Replication factor (RF) for this universe is three.
  • Nodes node1, node2 and node3 are the three master nodes.
  • All six nodes run tservers.

Desired configuration

We will transform the universe into the following final setup:

  • The universe will have six different nodes: node7, node8, node9, node10, node11 and node12.
  • Replication factor (RF) is still three.
  • Nodes node7, node8, node9 will be the three master nodes.
  • All six nodes run YB-TServers.
  • The original six nodes node1, node2, node3, node4, node5 and node6 will no longer be part of this universe.

Prerequisites

Ensure universe is in healthy state

This is to ensure that we do not inadvertently cause any further under replication on top of ongoing failures.

  • All YB-Master processes are running and able to talk to each other. This can be verified by checking the master UI (ex., http://node1:7000/) and ensure the Masters tab shows those three with one in the LEADER for RAFT Role.
  • All of the YB-TServer processes are running and heartbeating to the master leader. This can be verified by checking the master UI (ex., http://node1:7000/tablet-servers) which lists all the tablet servers in ALIVE state.

Ensure new machines are ready

Spin up a new set of VMs or servers (with the new AMI, for example) with IPs node7, node8, node9, node10, node11 and node12.

1. Configure new machines

Use these two steps to configure the six new machines:

2. Start YB-Master processes

Run the command below to bring up the new YB-Master process on the new master nodes node7, node8 and node9. When the master_addresses parameter is not set, this master process starts running without joining any existing master quorum. These nodes will be added to the master quorum in a later step.

  1. ~/master/bin/yb-master                \
  2.     --fs_data_dirs <data directories> \
  3.     >& /mnt/d0/yb-master.out &

NoteThe master_addresses parameter should not be set for these new masters.

Refer to starting master processes for further parameters and options.

3. Start YB-TServer processes

Run the following command to bring up the tserver processes on all the new nodes node7, node8, node9, node10, node11 and node12.

  1. export MASTERS=node1:7100,node2:7100,node3:7100,node7:7100,node8:7100,node9:7100
  2. ~/tserver/bin/yb-tserver              \
  3.     --tserver_master_addrs $MASTERS   \
  4.     --fs_data_dirs <data directories> \
  5.     >& /mnt/d0/yb-tserver.out &

Refer to starting tserver processes for further parameters and options.

NoteThe tserver_master_addrs parameter includes the new master IPs as well, so that they can keep heartbeating/reporting to the new master even after the old masters are removed from master quorum.

Now that the YB-TServer processes are running, we should verify that all the twelve tservers (six old and six new) are heartbeating to the master leader. Go to http://node1:7000/tablet-servers and confirm that twelve servers are in ALIVE status.

4. Perform data move

The data on this cluster can now be moved. First, we blacklist the old tablet servers to move the data away from them into the new set of tablet servers.

The commands below can be run from one of the old master nodes. You can first blacklist the six old tservers:

  1. export MASTERS=node1:7100,node2:7100,node3:7100
  2. ~/master/bin/yb-admin -master_addresses $MASTERS change_blacklist ADD node1:9100 node2:9100 node3:9100 node4:9100 node5:9100 node6:9100

Verify that the blacklist info looks similar to the output below:

  1. ~/master/bin/yb-admin -master_addresses $MASTERS get_universe_config
  2. Config:
  3. version: 5
  4. server_blacklist {
  5.   hosts {
  6.     host: "node1"
  7.     port: 9100
  8.   }
  9.   hosts {
  10.     host: "node2"
  11.     port: 9100
  12.   }
  13.   ...
  14.   ...
  15.   hosts {
  16.     host: "node6"
  17.     port: 9100
  18.   }
  19. }

Next, wait for the data move to complete. You can check the percentage completion by running the command below.

  1. ~/master/bin/yb-admin -master_addresses $MASTERS get_load_move_completion
  2. Percent complete = 66.6

In the example above, the data move is 66.6% done. Rerun this command periodically till the returned value reaches 100.

NoteThe time needed for this data move depends on the following:

  • number of tablets/tables
  • size of each of those tablets
  • disk/ssd transfer speeds
  • network bandwidth between the new nodes and the existing ones.

5. Master quorum change

Now we move the master quorum from the old set of masters node1,node2,node3 to the new set of masters node7,node8,node9. This is done by adding one new master followed by removing one old master sequentially, till all the old masters are removed. This can be run from one of the new masters.

ADDSERVER step adds a new master and REMOVE_SERVER step removes an old master from the master quorum. After every step, it is recommended to check the Masters state on master UI home page (i.e., _http://node7:7000).

NoteIf there is any error log reported on the command line from the steps below, double check the master UI, as it might be a transitional error message and can be ignored.

  1. export MASTERS=node1:7100,node2:7100,node3:7100,node7:7100,node8:7100,node9:7100
  2. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config ADD_SERVER node7 7100
  3. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config REMOVE_SERVER node1 7100
  4. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config ADD_SERVER node8 7100
  5. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config REMOVE_SERVER node2 7100
  6. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config ADD_SERVER node9 7100
  7. ~/master/bin/yb-admin -master_addresses $MASTERS change_master_config REMOVE_SERVER node3 7100

Now we ensure that the master leader is one of the new master nodes.

  1. $ export MASTERS=node7:7100,node8:7100,node9:7100
  2. $ ~/master/bin/yb-admin -master_addresses $MASTERS list_all_masters
  1. Master UUID         RPC Host/Port          State      Role
  2. ...                   node8:7100           ALIVE     FOLLOWER
  3. ...                   node9:7100           ALIVE     FOLLOWER
  4. ...                   node7:7100           ALIVE     LEADER

And confirm the same on http://node7:7000/, that the set of master IPs in the Masters list are only the three new ones.

On a new masters’ UI page, ensure that all the new tablet servers are reporting to the master leader and have the tablet load distributed. For example, http://node7:7000/tablet-servers should show the Load on six new tservers. The old tserver can be in DEAD status.

6. Update master addresses on tservers

The tserver_master_addrs parameter for all the new tserver processes needs to be set to the list of three new master IPs node7:7100,node8:7100,node9:7100 for future use.

TipUpdating master addresses is needed in case the yb-tserver process is restarted.

7. Cleanup

The old nodes are not part of the universe any more and can be shutdown.Once the old tserver processes are terminated, you can cleanup the blacklist from the master configuration using the command below.

  1. ~/master/bin/yb-admin -master_addresses $MASTERS change_blacklist REMOVE node1:9100 node2:9100 node3:9100 node4:9100 node5:9100 node6:9100

TipCleaning up the blacklist server will help reuse the older IPs in case they get recycled.

Ensure there are no server_blacklist entries returned by the command:

  1. ~/master/bin/yb-admin -master_addresses $MASTERS get_universe_config