JMX Connector

The JMX connector provides the ability to query JMX information from all nodes in a openLooKeng cluster. This is very useful for monitoring or debugging. Java Management Extensions (JMX) provides information about the Java Virtual Machine and all of the software running inside it. openLooKeng itself is heavily instrumented via JMX.

This connector can also be configured so that chosen JMX information will be periodically dumped and stored in memory for later access.

Configuration

To configure the JMX connector, create a catalog properties file etc/catalog/jmx.properties with the following contents:

  1. connector.name=jmx

To enable periodical dumps, define the following properties:

  1. connector.name=jmx
  2. jmx.dump-tables=java.lang:type=Runtime,presto.execution.scheduler:name=NodeScheduler
  3. jmx.dump-period=10s
  4. jmx.max-entries=86400

dump-tables is a comma separated list of Managed Beans (MBean). It specifies which MBeans will be sampled and stored in memory every dump-period. History will have limited size of max-entries of entries. Both dump-period and max-entries have default values of 10s and 86400 accordingly.

Commas in MBean names should be escaped in the following manner:

  1. connector.name=jmx
  2. jmx.dump-tables=presto.memory:type=memorypool\\,name=general,\
  3.    presto.memory:type=memorypool\\,name=system,\
  4.    presto.memory:type=memorypool\\,name=reserved

Querying JMX

The JMX connector provides two schemas.

The first one is current that contains every MBean from every node in the openLooKeng cluster. You can see all of the available MBeans by running SHOW TABLES:

  1. SHOW TABLES FROM jmx.current;

MBean names map to non-standard table names and must be quoted with double quotes when referencing them in a query. For example, the following query shows the JVM version of every node:

  1. SELECT node, vmname, vmversion
  2. FROM jmx.current."java.lang:type=runtime";
  1. node                 |              vmname               | vmversion
  2. --------------------------------------+-----------------------------------+-----------
  3. ddc4df17-0b8e-4843-bb14-1b8af1a7451a | Java HotSpot(TM) 64-Bit Server VM | 24.60-b09
  4. (1 row)

The following query shows the open and maximum file descriptor counts for each node:

  1. SELECT openfiledescriptorcount, maxfiledescriptorcount
  2. FROM jmx.current."java.lang:type=operatingsystem";
  1. openfiledescriptorcount | maxfiledescriptorcount
  2. -------------------------+------------------------
  3.                     329 |                  10240
  4. (1 row)

The wildcard character * may be used with table names in the current schema. This allows matching several MBean objects within a single query. The following query returns information from the different openLooKeng memory pools on each node:

  1. SELECT freebytes, node, object_name
  2. FROM jmx.current."presto.memory:*type=memorypool*";
  1. freebytes  |  node   |                       object_name
  2. ------------+---------+----------------------------------------------------------
  3.  214748364 | example | presto.memory:type=MemoryPool,name=reserved
  4. 1073741825 | example | presto.memory:type=MemoryPool,name=general
  5.  858993459 | example | presto.memory:type=MemoryPool,name=system
  6. (3 rows)

The history schema contains the list of tables configured in the connector properties file. The tables have the same columns as those in the current schema, but with an additional timestamp column that stores the time at which the snapshot was taken:

  1. SELECT "timestamp", "uptime" FROM jmx.history."java.lang:type=runtime";
  1. timestamp        | uptime
  2. -------------------------+--------
  3. 2016-01-28 10:18:50.000 |  11420
  4. 2016-01-28 10:19:00.000 |  21422
  5. 2016-01-28 10:19:10.000 |  31412
  6. (3 rows)