David Judd created STORM-2439:
---------------------------------
Summary: HealthCheck feature does not work
Key: STORM-2439
URL: https://issues.apache.org/jira/browse/STORM-2439
Project: Apache Storm
Issue Type: Bug
Components: storm-core
Affects Versions: 1.0.3
Reporter: David Judd
There are a few issues with this feature:
1. The default timeout value produces `java.lang.ClassCastException:
java.lang.Integer cannot be cast to java.lang.Long at
org.apache.storm.command.HealthCheck.processScript(HealthCheck.java:79)`
because the value, 5000, is automatically deserialized by Jackson as an
Integer, but we attempt to cast it to a long. (I successfully worked around
this by setting a timeout greater than the maximum int.)
2. The documentation says that a script should print "ERROR" if the node is
unhealthy, but in fact the script must *also* exit with a non-zero exit code.
This appears to be the opposite of what is intended, given a comment that says
"We treat non-zero exit codes as indicators that the scripts failed to execute
properly, not that the system is unhealthy". I believe the test in this line is
inverted:
https://github.com/apache/storm/blob/70102643e74d577728adf5f8719920d1bf60e98a/storm-core/src/jvm/org/apache/storm/command/HealthCheck.java#L97
3. Even with workarounds for the above two bugs, a failing health check does
not cause workers to shut down in my testing with Storm 1.0.3. I have not
determined the cause, but because the previous two issues suggest to me that
this code is rarely if ever tested, I do not plan to investigate further at the
moment.
If this feature is, as it appears, untested and non-functional, I would suggest
that it be removed from the code and documentation.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
