[HDFS-11308] NameNode doFence state judgment problem

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.7.1
• Component/s: auto-failover

Environment

CentOS Linux release 7.1.1503 (Core)

• Assignee: Unassigned
• Reporter: Wally Tang
• Created: 09/Jan/17 22:05
• Updated: 02/Oct/19 17:14
• Resolved: 10/Jan/17 20:42

Description

In our Cluster, I found some abnormal in ZKFC log

[2017-01-10T01:42:37.168+08:00] [INFO] hadoop.ha.SshFenceByTcpPort.doFence(SshFenceByTcpPort.java 147) [Health Monitor for NameNode at xxx-xxx-172xxx.hadoop.xxx.com/xxx.xxx.172.xxx:8021-EventThread] : Indeterminate response from trying to kill service. Verifying whether it is running using nc...
[2017-01-10T01:42:37.234+08:00] [WARN] hadoop.ha.SshFenceByTcpPort.pump(StreamPumper.java 88) [nc -z xxx-xxx-172xx.hadoop.xx.com 8021 via ssh: StreamPumper for STDERR] : nc -z xxx-xxx-172xx.hadoop.xxx.com 8021 via ssh: nc: invalid option -- 'z'
[2017-01-10T01:42:37.235+08:00] [WARN] hadoop.ha.SshFenceByTcpPort.pump(StreamPumper.java 88) [nc -z xxx-xxx-172xx.hadoop.xxx.com 8021 via ssh: StreamPumper for STDERR] : nc -z xxx-xxx-17224.hadoop.xxx.com 8021 via ssh: Ncat: Try `--help' or man(1) ncat for more information, usage options and help. QUITTING.

When I perform nc an exception occurs, the return value is 2, and cannot confirm sshfence success，this may lead to some problems

SshFenceByTcpPort.java

rc = execCommand(session, "nc -z " + serviceAddr.getHostName() +
            " " + serviceAddr.getPort());
        if (rc == 0) {
          // the service is still listening - we are unable to fence
          LOG.warn("Unable to fence - it is running but we cannot kill it");
          return false;
        } else {
          LOG.info("Verified that the service is down.");
          return true;          
        }

Issue Links

• duplicates: HDFS-3618 SSH fencing option may incorrectly succeed if nc (netcat) command not present