Change IP addresses
Give a server or the console a new LAN IP address, then update the clients, the VIP and the console URL that use the old address.
What happens
Replication, failover, backups and metrics use the mesh overlay address,
which does not change. Only the client connection strings, the VIP and the
console URL (TSE_PUBLIC_URL) use LAN addresses.
| Change | Effect on the cluster | What you do |
|---|---|---|
| A server gets a new address in the same subnet | None | Update the clients that use the LAN address of the node. |
| A server moves to a different subnet | None, but the node cannot hold the VIP | Update the clients, then repair the VIP. |
The console gets a new address, and TSE_PUBLIC_URL is a DNS name |
None after the DNS update | Update the DNS record. |
The console gets a new address, and TSE_PUBLIC_URL is an IP address |
The servers cannot reach the console | Keep the old address on the console host, or change TSE_PUBLIC_URL and enroll each server again. |
Before you start
- For a planned change, the service Overview shows the service as healthy, and no operation runs. If the address already changed and the service is not healthy, contact support before you start another operation.
- From its new address, each server can reach the console URL and the mesh
ports
8090/tcp,41641/udpand3478/udp. See TLS modes. - Add the new address to each firewall rule that has the old address.
A server gets a new address in the same subnet
Change the address on the server or in the DHCP service. The mesh finds the new address.
Test the server
On the Servers page, open the menu of the server and select Test server. If the test fails, check the network access of the server to the console and the firewall rules. If you cannot restore access, contact support.
Check the new address
Once an hour, the console reads the LAN address of each server of a database service. It updates the server and the connection strings, and adds an entry on the Operations page. On a service without a VIP, the Overview shows a LAN address changed banner for 7 days.
To read the address now, open the server and select Refresh on its Storage card. Make sure that the server and the connection information show the new address.
Update the clients
- Through the VIP: no change.
- A connection string with all nodes, no VIP: it still connects through the other nodes. Replace the old address.
- Direct endpoints (the primary on
:5432or:6432): replace the address. - Single-node service: replace the address in each client.
To prevent this, give each server a static address or a DHCP reservation.
A server moves to another subnet
Do the steps in the previous section first. With no VIP, that is all.
With a VIP, the node has no address in the subnet of the VIP after the move. Other nodes in the old subnet can still hold the VIP. If all nodes move, connections through the VIP fail. Within an hour, the Overview shows a VIP is stranded banner that names the node.
Repair the VIP in one of these two ways:
-
Keep the VIP. Give the node an address in the subnet of the VIP again, for example on a second interface or a VLAN. Then restart keepalived on the node:
sudo systemctl restart keepalivedThe banner goes away after the next hourly check. If it stays, contact support.
-
Change the VIP. Use this when all nodes have an address in the new subnet. Select Change VIP on the banner. Enter a free address in the new subnet. Its last octet must be different from the last octet of the VIP of each other service. Move the clients to the new VIP. See Change the VIP.
Change VIP refuses to start while the service is not running, a member is down, or a member waits for Replace a server.
The console gets a new address
While the servers cannot reach the console, connections that depend on the console relay stop. A server that restarts cannot join the mesh again until it can reach the console. Do the steps below soon after the move. See Console outage.
To move the console to a new host, restore a console backup on the new host. The same address cases apply. See Restore the console.
TSE_PUBLIC_URL is a DNS name
Lower the TTL
If the move is still planned, decrease the TTL of the DNS record before the move.
Update the DNS record
Point the DNS record to the new address.
Test the servers
On the Servers page, run Test server for each server. If a test
fails, run this on that server. Replace console.example.com with the
host name in TSE_PUBLIC_URL:
getent hosts console.example.comIf it prints the old address or nothing, check the DNS record and the DNS settings of the server. If the server still cannot connect, contact support.
TSE_PUBLIC_URL is an IP address
Each server keeps the console address from its enrollment. Do not change
TSE_PUBLIC_URL unless you enroll each server again (below). Keep the old
address available on the console host:
Keep the old address
Add the old IP address to the console host as a second address, and make it persistent in the network configuration. Or, on a router or a firewall, forward the mesh ports of the old address to the new address.
Test the servers
On the Servers page, run Test server for each server. If a test fails, check the address or the forwarding and the network access of the server. If the server still cannot connect, contact support.
Change TSE_PUBLIC_URL
When you change TSE_PUBLIC_URL, you must enroll each server again.
Re-enroll command keeps the mesh identity and the overlay address of the
server, and the members stay in their clusters.
Before you start:
- The new URL is reachable from each server.
- In
providedTLS mode, your certificate has the new address. - You recorded the overlay address of each server.
Set the new URL
On the console host, run ./timescale-enterprise.sh in the install
directory (/opt/tse by default) and select Change the listener. At
Server-facing console URL (TSE_PUBLIC_URL), enter the new URL. Use a
DNS name, so that the next move is only a DNS change. Keep the current
values for the other prompts.
The installer restarts the console. In self-signed mode, the console
also makes a new browser certificate.
Re-enroll each server
For each server, one at a time: on the Servers page, open the menu of the server, select Re-enroll command, and run the command on the server. The command shows kept the mesh identity. Then run Test server, and compare the overlay address with the address that you recorded. For a service or DCS member, make sure that the topology shows it as healthy.
If the command does not show kept the mesh identity, stop. Also stop if the test fails, the address changed, or the member is not healthy. Contact support before you re-enroll the next server.
Check the services
On each service Overview, make sure that the service is healthy. If it is not, contact support.