Client Connections
This group provides metrics associated to client connections to the nodes.
If you are seeing non zero values for these metrics it means the node is in an overloaded state. Cassandra versions prior to 3.11.6 when overloaded would result in blocked Native Transport Requests.
By default when the overload state is reached cassandra will apply back-pressure on the client which will result in an increase in PausedConnections. If the client started its connection with THROW_ON_OVERLOAD the overload state will result in an increase of RequestDiscarded.
Troubleshooting: Consistent non zero values indicates for these metrics indicate the cluster is overloaded. To address, load on the cluster needs to be reduced or processing capacity added. Instaclustr support can assist with horizontally or vertical scaling processing capacity as appropriate.
RequestDiscarded
Monitors requests that been discarded due to the node being overloaded from clients that have started with THROW_ON_OVERLOAD set to True. The metric is provided in two ways:
one_minute_rate – In the last minute the requests that have been discarded
value – The total requests that have been discarded
You can also get this metric through the Monitoring API with the query param metrics=n::requestDiscarded
PausedConnections
Monitors requests from clients that have had their requests paused due to the node being overloaded from clients that have started with THROW_ON_OVERLOAD as default or set to False:
value – The current connections that have been paused (back-pressure applied)
You can also get this metric through the Monitoring API with the query param metrics=n::pausedConnections