Availability of Cassandra

Question

I am running into an issue "me.prettyprint.hector.api.exceptions.HUnavailableException: : May not be enough replicas present to handle consistency level." when I have RF=1, Read Consistency Level = 1 and one of the nodes in 6 node ring/ cluster is down. All of my reads are failing with this exception. Any idea? Ideally only reads that are looking for data in the node which is down should fail and all other reads should be successful?

Answer 1

I've seen something similar when I mis-configured my replication settings, specifically I had the wrong datacenters named om the replication strategy. Double check what your DCs are (assuming you're using NetworkTopologyStrategy).

If you don't already know your DC names, in a shell on one of the nodes run:

$ nodetool -h localhost ring
Address         DC          Rack        Status State   Load            Owns    Token                                       
                                                                               141784319550391000000000000000000000000     
172.26.233.135  Cassandra   rack1       Up     Normal  25.75 MB        16.67%  0                                           
172.26.233.136  Cassandra   rack1       Up     Normal  26.03 MB        16.67%  28356863910078200000000000000000000000      
172.26.233.137  Cassandra   rack1       Up     Normal  27.19 MB        16.67%  56713727820156400000000000000000000000      
172.26.233.138  Cassandra   rack1       Up     Normal  26.78 MB        16.67%  85070591730234600000000000000000000000      
172.26.233.139  Solr        rack1       Up     Normal  24.47 MB        16.67%  113427455640313000000000000000000000000     
172.26.233.140  Solr        rack1       Up     Normal  26.66 MB        16.67%  141784319550391000000000000000000000000 

You can see we have two DCs, Cassandra and Solr (this is a DSE cluster).

In cassandra-cli:

use Keyspace1;
describe;

CLI will print the strategy options:

Keyspace: Catalog:
  Replication Strategy: org.apache.cassandra.locator.NetworkTopologyStrategy
  Durable Writes: true
    Options: [DC1:3]
...

We have a mis-match. Cassandra is looking for a datacenter named DC1 hence the UnavailableException. We need to update the replication options to match the actual DCs in our cluster. In CLI, update the strategy options for your keyspace using the data center names:

update keyspace Keyspace1 with strategy_options = {Cassandra:3,Solr:2};

Answer 2

There could be a few possibilities:

• You're running a multi-row query (get_range, get_indexed_slices, multiget, or the cql equivalents) that requires multiple nodes to be up
• Your cluster is unbalanced, with the down node owning most of the ring; a bad multi-dc configuration could also produce something similar
• Your cluster wasn't in a good state to begin with, where some nodes don't see others. Make sure nodetool ring shows the same output when run against each node in the cluster

If none of those are the cause, double check that you're specifying the consistency level correctly with Hector and cqlsh.