Details
-
Type:
Bug
-
Status: Resolved
-
Priority:
Critical
-
Resolution: Fixed
-
Component/s: core
-
Labels:None
Description
If a stream is open remotely using FilePath.read() and the file is not read to the very end, the log will be spammed with exceptions like:
Jan 10, 2011 1:36:51 PM hudson.remoting.ProxyOutputStream$Chunk$1 run WARNING: Failed to write to stream java.io.IOException: Pipe is already closed at hudson.remoting.FastPipedOutputStream.write(FastPipedOutputStream.java:147) at hudson.remoting.FastPipedOutputStream.write(FastPipedOutputStream.java:131) at hudson.remoting.ProxyOutputStream$Chunk$1.run(ProxyOutputStream.java:185)
...and finally disconnect the slave. Several plug-ins are affected by this bug, for instance the Emma-plug-in. Marked as critical since it yet another remote issue and keeps us from upgrade. It is also easy to reproduce.
To reproduce, just create a publisher plug-in with a perform() that looks like this:
@Override
public boolean perform(AbstractBuild build, Launcher launcher, BuildListener listener) throws IOException {
InputStream in = build.getWorkspace().child("large_file_in_workspace").read();
// Just read one byte
System.out.println(in.read());
// Close it and watch the hudson log grow and slave (most often) disconnect.
in.close();
return true;
}
Test the plug-in with Hudson 1.390+. Make sure the new publisher plug-in is used in a project running on a slave (and also make sure that the "large_file_in_workspace"-file exists in the remote workspace).
Activity
Code changed in jenkins
User: Brian Atkinson
Path:
remoting/src/main/java/hudson/remoting/Channel.java
remoting/src/main/java/hudson/remoting/Command.java
remoting/src/main/java/hudson/remoting/Pipe.java
remoting/src/main/java/hudson/remoting/PipeWindow.java
remoting/src/main/java/hudson/remoting/ProxyOutputStream.java
remoting/src/test/java/hudson/remoting/PipeTest.java
http://jenkins-ci.org/commit/core/4c10efe1a9285bcc6eed193ede82c0bdd39d658a
Log:
Merge branch 'JENKINS-8592'
Integrated in 
jenkins_main_trunk #519
JENKINS-8592 let the writer end know if the reader end of the pipe
Kohsuke Kawaguchi :
Files :
- remoting/src/main/java/hudson/remoting/Pipe.java
- remoting/src/main/java/hudson/remoting/ProxyOutputStream.java
- remoting/src/test/java/hudson/remoting/PipeTest.java
- remoting/src/main/java/hudson/remoting/PipeWindow.java
- remoting/src/main/java/hudson/remoting/Channel.java
- remoting/src/main/java/hudson/remoting/Command.java
I was told in a comment that this issue might fix JENKINS-6355 indirectly. I can confirm it has not fixed JENKINS-6355.
I'm using 1.400 and emma plugin 1.21 and issue still occurs intermittently.
sorry the JIRA in the previous comment should be: JENKINS-7586
I have been able to reproduce with a unit test. I hope to be able to get a basic fix merged into core in the next week.
Currently the pipe implementation pushes data from one side to the other without having a way to actually stop the flow from the receiving side. In the first pass I just want to have the receiving side read till the stream is exhausted. This should make the stream shut down cleanly. In a second pass I want to try to see if I can shut down the stream from the receiving side. Doing this properly may take a few tries and may be error prone, so I want to get a safe fix out the door to get things working again giving me time to experiment with a more proper fix.
Think this might work for you?