[HDFS-5680] Double-JSON-encoding of JMX attributes yields malformed JSON

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Component/s: datanode, namenode
• Assignee: Unassigned
• Reporter: Eric Sirianni
• Created: 18/Dec/13 15:37
• Updated: 18/Dec/13 15:37

Description

The practice of "double JSON encoding" JMX attributes:

1. in the MXBean itself using Jetty's JSON.toString(...)
2. in the JMXJsonServlet using Jackson

results in malformed JSON documents (technically not "malformed", but doesn't properly allow for traversal into nested JSON structures).

For example:

$ curl http://localhost:50070/jmx?get=Hadoop:service=NameNode,name=NameNodeInfo::NodeUsage
{
  "beans" : [ {
    "name" : "Hadoop:service=NameNode,name=NameNodeInfo",
    "modelerType" : "org.apache.hadoop.hdfs.server.namenode.FSNamesystem",
    "NodeUsage" : "{\"nodeUsage\":{\"min\":\"0.00%\",\"median\":\"0.01%\",\"max\":\"0.01%\",\"stdDev\":\"0.00%\"}}"
  } ]

The NodeUsage attribute should instead be:

    "NodeUsage" : {"nodeUsage":{"min":"0.00%","median":"0.01%","max":"0.01%","stdDev":"0.00%"}}

The Web UI seems to be aware of this and has applied a workaround:

dfshealth.js

  // Workarounds for the fact that JMXJsonServlet returns non-standard JSON strings

Putting JMX aside, one fix would be to let JMXJsonServlet do all the encoding and simply have the MXBean return a Map. That would, however, likely not encode properly for native JMX access. I'm not familiar enough with how JMX is used elsewhere in Hadoop to suggest an alternative.