运行cqlsh遇连接拒绝/无法连接服务器，求助排查yaml配置问题

Absolutely—missing the rpc_address parameter in your cassandra.yaml is almost certainly the root cause of your connection errors ("unable to connect to any server" and "connection refused"). Here's a detailed breakdown of why this happens, how to verify it, and how to fix it:

1. Why rpc_address is Critical for cqlsh #

Cassandra's cqlsh utility relies on the Thrift protocol (the default for cqlsh connections in Cassandra 3.x) to communicate with the node. The rpc_address configures which IP address the Thrift service binds to.

By default, Cassandra sets rpc_address: localhost if the parameter isn't explicitly defined. If you've removed this parameter entirely from your trimmed yaml, the Thrift service may fail to start correctly, or only bind to the loopback interface (localhost). This means:

• If you're running cqlsh from a remote machine, it can't reach the node's Thrift service.
• Even if you run cqlsh locally, misconfiguration here can prevent the service from listening properly, leading to connection refusals.

2. How to Verify the Issue #

Before making changes, confirm this is the problem with these quick checks:

• Check Cassandra logs: Look in your logs/system.log file for entries related to Thrift. If you don't see a line like Starting Thrift service on /192.168.x.x:9160, the service didn't start due to missing configuration.
• Check listening ports: Run this command to see if the Thrift port (9160) or native CQL port (9042) is active:

  netstat -tulpn | grep cassandra
  

  If port 9160 isn't listed, the Thrift service isn't listening—confirming the rpc_address issue.

3. Fix the Configuration #

Add the rpc_address parameter to your cassandra.yaml file, using the same value as your listen_address (for a single-node cluster) or 0.0.0.0 to allow connections from any address (use cautiously in production):

# Replace with your node's actual IP or listen_address value
rpc_address: 192.168.1.100

For multi-node clusters or environments with network address translation (NAT), you may also need to set broadcast_rpc_address to the public IP of the node, but this isn't necessary for a single-node setup.

After adding the parameter, restart Cassandra:

service cassandra restart

Then test the connection again with:

cqlsh <your-node-ip>

4. Additional Troubleshooting Steps #

If the issue persists after adding rpc_address, check these other common culprits:

• Verify listen_address: Ensure this is set to your node's actual IP (not localhost if you need remote access).
• Firewall rules: RHEL 6.9 uses iptables—confirm ports 9160 (Thrift) and 9042 (native CQL) are open:

  iptables -A INPUT -p tcp --dport 9160 -j ACCEPT
  iptables -A INPUT -p tcp --dport 9042 -j ACCEPT
  service iptables save
  

• Service status: Confirm Cassandra is running with service cassandra status. If it's not, check the system log for startup errors (e.g., invalid yaml syntax).

内容的提问来源于stack exchange，提问作者Vishal Sharma