IV19713: java.nio.channels.Selector.select(long) in tight loop with zero selected keys in Linux

APAR status

• Closed as program error.

Error description

• Error Message: N/A
  .
  Stack Trace: N/A
  .
  The application observe a tight loop in the select() method of
  Selector class which continues till the channels are explicitly
  cancelled by the application.
  

Local fix

• Monitor for select() method of Selector class returns with zero
  selected keys and cancel the channels through application if it
  exceeds the expected limit.
  

Problem summary

• The root cause of the issue is that there is a very small timing
  window between a channel set to close (not open) in the Java
  layer and the actual close of the channel in the native layer.
  Hence, if the interestOps is set to 0 for that channel by the
  application during this timing window, the selector presumes
  that the channel is already closed and do not try to delete the
  channel from the select operation. Hence, it will result in a
  situation that on Java side, the interest ops is not selected
  whereas on the native side the previous interested operation is
  still selected. If the interested operation is ready on the
  native side for that channel, then the polling for that channel
  immediately returns. As the selected operation is no more
  interested in the Java layer, the application invoke the
  select() again without clearing the data from the channel. In
  the meantime, the other thread which tries to close the channel
  is waiting for the selector lock to cancel the key. As the
  select() operation immediately returns from the native side
  polling results in a tight loop, the selector lock is not
  released. Hence, the channel is not able to close which result
  in an infinite loop in select() method.
  

Problem conclusion

• This defect will be fixed in:
  6.0.0 SR11
  6.0.1 SR3
  7.0.0 SR2
  .
  The JVM is updated to ensure that the channels are deleted from
  the select operation when interestOps is set to zero
  irrespective of whether the channel is in close processing or
  not.
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  JAVA CLASS LIBS

• Fixed component ID

  620700130

Applicable component levels

• R600 PSY

     UP

• R700 PSY

     UP