Elek, Marton created HDDS-1397:
----------------------------------
Summary: Avoid the usage of signal handlers in datanodes of the
MiniOzoneClusters
Key: HDDS-1397
URL: https://issues.apache.org/jira/browse/HDDS-1397
Project: Hadoop Distributed Data Store
Issue Type: Bug
Reporter: Elek, Marton
Assignee: Elek, Marton
[~arpaga] showed me a problem that TestQueryNode.testHealthyNodesCount is
failed in the CI check of HDDS-1339.
According to the logs the test is timed out because only 4 datanodes are
started out of the 5.
The log also contained an exception from one datanode:
{code}
2019-04-04 00:26:33,583 WARN ozone.HddsDatanodeService
(LogAdapter.java:warn(59)) - failed to register any UNIX signal loggers:
java.lang.IllegalStateException: Can't re-install the signal handlers.
at org.apache.hadoop.util.SignalLogger.register(SignalLogger.java:77)
at
org.apache.hadoop.util.StringUtils.startupShutdownMessage(StringUtils.java:718)
at
org.apache.hadoop.util.StringUtils.startupShutdownMessage(StringUtils.java:707)
at
org.apache.hadoop.ozone.HddsDatanodeService.createHddsDatanodeService(HddsDatanodeService.java:126)
at
org.apache.hadoop.ozone.HddsDatanodeService.createHddsDatanodeService(HddsDatanodeService.java:108)
at
org.apache.hadoop.ozone.MiniOzoneClusterImpl$Builder.createHddsDatanodes(MiniOzoneClusterImpl.java:552)
{code}
The code which requires the signal handler is the following (signal handler is
registered in the startupShutdownMessage)
{code}
/**
* Create an Datanode instance based on the supplied command-line arguments.
* <p>
* This method is intended for unit tests only. It suppresses the
* startup/shutdown message and skips registering Unix signal handlers.
*
* @param args command line arguments.
* @param conf HDDS configuration
* @param printBanner if true, then log a verbose startup message.
* @return Datanode instance
*/
private static HddsDatanodeService createHddsDatanodeService(
String[] args, Configuration conf, boolean printBanner) {
if (args.length == 0 && printBanner) {
StringUtils
.startupShutdownMessage(HddsDatanodeService.class, args, LOG);
return new HddsDatanodeService(conf);
} else {
new HddsDatanodeService().run(args);
return null;
}
{code}
As you can read from the comment it's expected to be called with
printBanner=false to avoid the creation of the signal handler.
Note: In the startupShutdownMessage method a new signal handler is registered
and signal handlers can be registered only once:
{code}
//SignalLogger
void register(final LogAdapter LOG) {
if (registered) {
throw new IllegalStateException("Can't re-install the signal handlers.");
}
....
{code}
We have a dedicated method to create datanode service for the unit tests. The
only thing what we need is to turn OFF the signal handler registration here.
(The following code fragment shows the original state where the signal handler
creation is requested with the true parameter value)
{code}
@VisibleForTesting
public static HddsDatanodeService createHddsDatanodeService(
String[] args, Configuration conf) {
return createHddsDatanodeService(args, conf, true);
}
{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org
