Node Instance Restart

Prior to release v3.18.1, to restart a Duroslight/GFTL/node instance, you were required to restart the node manager service. This operation takes time, and in the case of a dual-node server, if for example there is a problem with one Duroslight, you still need to restart the node manager service, which will impact both nodes.

The new API documented below allows you to restart only a single node-instance control plane (running inside the NM service) and only its own Duroslight service and its gftl-instance, without impacting the GFTL service (i.e., without impacting the other node’s GFTL instance). You can also perform the restart only for the node’s control plane and Duroslight instances, without its GFTL instance. If you are not restarting the GFTL instance, the restart time is shorter, and this reduces the rebuild time after the restart. The API can be executed remotely, so you can run the API on Server A to restart a node on any other server in the cluster.

In general, the API will fail if the restart causes a PG to go into read-only mode. If you must restart, in this case, you should use the --force flag to override.

For additional information on this API, see lbcli restart node-instance.

There are a few options here, detailed in the table below:

Node InstanceDuroslightGFTLDuration
Restart a nodeYesYesNo~30sec
Restart a node (and GFTL)YesYesYes~60sec
Stop a nodeYesYesNo~3sec
Start a nodeYesYesNo~30sec

lbcli Examples

Status of the node: the node-manager server is up and running:

Shell
Copy

Validate that the other services are running (GFTL and Duroslight):

Shell
Copy

Example of restarting a node instance (Prompt Y to verify):

Shell
Copy

Example of restarting a node instance and GFTL (--consider-backend=true):

Shell
Copy

Example of how to restart a node, even if it will cause a PG to become read-only:

Shell
Copy

If you try to restart or stop a node instance when another node is down (which could cause PGs to go into read-only state), the API will fail. This is a safeguard, and if it is required, you should use the --force flag:

Shell
Copy

Example of stopping a node instance (Prompt Y to verify):

Shell
Copy

Example of starting a node instance (Prompt Y to verify):

Shell
Copy
Type to search, ESC to discard
Type to search, ESC to discard
Type to search, ESC to discard