0

I have a 2 DC(each 2 node) cluster. The two data-centers(DC1,DC2) are about 5000 km apart.

For one of the datacenters(DC1), the number of read requests are about 10 times the other node. The load policy used is Token Aware(DC Aware Round Robin).

When I try to make queries from devcenter, the load is balanced among the two nodes in DC1. Gossip is normal among the nodes. The snitch used is GossipPropertyFileSnitch.

The variable phi_convict_threshold is 12 in the yaml. What could be the problem. Any direction you may think is appreciated.

Japneet Singh
  • 21
  • 3

1 Answers1

0

Why did you set phi_convict_threshold to 0.1 ? The usual value is 8 and for cloud deployments (AWS for ex) it's recommended to set it to 12.

The lower the phi_convict_threshold is set, the more false positive gossip protocol will mark a node as being DOWN

doanduyhai
  • 8,712
  • 27
  • 26