[
https://issues.apache.org/jira/browse/STORM-442?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
DashengJu updated STORM-442:
----------------------------
Description:
In ShellBolt, the _readerThread read command from python/shell process, and
handle like this:
try {
ShellMsg shellMsg = _process.readShellMsg();
...
} catch (InterruptedException e) {
} catch (Throwable t) {
die(t);
}
And in the die function, getProcessTerminationInfoString will read
getErrorsString() from processErrorStream.
private void die(Throwable exception) {
String processInfo = _process.getProcessInfoString() +
_process.getProcessTerminationInfoString();
_exception = new RuntimeException(processInfo, exception);
}
so when ShellBolt got exception(for example, readShellMsg() throw NPE ) , but
it is not an error from sub process, then getProcessTerminationInfoString will
be hang because processErrorStream have no data to read.
On the other hand, as [~xiaokang] says ShellBolt should fail fast on exception
( https://github.com/apache/incubator-storm/pull/46 ) , I think it is not a
good idea to read error info from stream.
Because [~xiaokang] 's PR is based old version, so I will move his code to this
PR, and modify some other place in ShellSpout.
was:
In ShellBolt, the _readerThread read command from python/shell process, and
handle like this:
try {
ShellMsg shellMsg = _process.readShellMsg();
...
} catch (InterruptedException e) {
} catch (Throwable t) {
die(t);
}
And in the die function, getProcessTerminationInfoString will read
getErrorsString() from processErrorStream.
private void die(Throwable exception) {
String processInfo = _process.getProcessInfoString() +
_process.getProcessTerminationInfoString();
_exception = new RuntimeException(processInfo, exception);
}
so when ShellBolt got exception(for example, readShellMsg() throw NPE ) , but
it is not an error from sub process, then getProcessTerminationInfoString will
be hang because processErrorStream have no data to read.
On the other hand, as [~xiaokang] says ShellBolt should fail fast on exception
( https://github.com/apache/incubator-storm/pull/46 ) , I think it is not a
good idea to read error info from stream.
Because [~xiaokang] 's PR is based old version, so I will move his code to this
PR, and modify some other place in ShellSpout.
> multilang ShellBolt/ShellSpout die() can be hang when Exception happened
> ------------------------------------------------------------------------
>
> Key: STORM-442
> URL: https://issues.apache.org/jira/browse/STORM-442
> Project: Apache Storm (Incubating)
> Issue Type: Bug
> Affects Versions: 0.9.3-incubating
> Reporter: DashengJu
>
> In ShellBolt, the _readerThread read command from python/shell process, and
> handle like this:
> try {
> ShellMsg shellMsg = _process.readShellMsg();
> ...
> } catch (InterruptedException e) {
> } catch (Throwable t) {
> die(t);
> }
> And in the die function, getProcessTerminationInfoString will read
> getErrorsString() from processErrorStream.
> private void die(Throwable exception) {
> String processInfo = _process.getProcessInfoString() +
> _process.getProcessTerminationInfoString();
> _exception = new RuntimeException(processInfo, exception);
> }
> so when ShellBolt got exception(for example, readShellMsg() throw NPE ) ,
> but it is not an error from sub process, then
> getProcessTerminationInfoString will be hang because processErrorStream have
> no data to read.
> On the other hand, as [~xiaokang] says ShellBolt should fail fast on
> exception ( https://github.com/apache/incubator-storm/pull/46 ) , I think it
> is not a good idea to read error info from stream.
> Because [~xiaokang] 's PR is based old version, so I will move his code to
> this PR, and modify some other place in ShellSpout.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
