Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 402784

Summary: Orion Windows tests failing
Product: [ECD] Orion Reporter: John Arthorne <john.arthorne>
Component: RelengAssignee: Project Inbox <orion.releng-inbox>
Status: RESOLVED WONTFIX QA Contact:
Severity: normal    
Priority: P3 CC: webmaster
Version: 2.0   
Target Milestone: ---   
Hardware: PC   
OS: Windows 7   
Whiteboard:
Bug Depends on: 401964    
Bug Blocks:    
Attachments:
Description Flags
Patch to set proxy info in base builder
none
current best guess of proxy settngs none

Description John Arthorne CLA 2013-03-08 15:52:17 EST
The Orion tests have been failing on the hudson windows slave for the past week or so. Possibly related the webmaster has been seeing this error message when logging into the windows slave:


"Windows cannot find c:\hb\workspace\orion-jstest-windows\test\eclipse\orion.exe.  Make sure you typed the name correctly, and then try again"

See bug 401964 for more discussion.
Comment 1 John Arthorne CLA 2013-03-11 11:02:21 EDT
So what is happening here is that we are failing to even install Orion on the windows slave machine. The provisioning fails with:

Mar 10, 2013 10:54:54 PM org.apache.commons.httpclient.HttpMethodDirector executeWithRetry
INFO: I/O exception (java.net.ConnectException) caught when processing request: Connection timed out: connect

Since we can't install orion, it then fails to launch orion.exe because it's not there.
Comment 2 John Arthorne CLA 2013-03-11 11:08:28 EDT
Here is the p2 operation used to install Orion on the hudson test machine:

c:\hb\workspace\orion-jstest-windows>"c:\java\jdk1.6.0_20\jre\bin\java" -Xmx500m -jar c:\hb\workspace\orion-jstest-windows/org.eclipse.releng.basebuilder/plugins/org.eclipse.equinox.launcher.jar -application org.eclipse.equinox.p2.director -repository http://build.eclipse.org/eclipse/e4/orion/target/integration/I20130310-2230 -i org.eclipse.orion -d c:\hb\workspace\orion-jstest-windows/test/eclipse

We have not changed the "basebuilder" version used to build Orion for a long time (since June 2012). Since it fails before Orion is even installed it doesn't look like an Orion problem either. This must be related to some kind of networking problem on the hudson slave.
Comment 3 Eclipse Webmaster CLA 2013-03-12 11:28:13 EDT
I updated the windows slave to use a 'local' address(on the same subnet) for download.eclipse.org but when I manually trigger the job it still fails.  A quick peek in the log file indicates some kind of bundle resolution issue.

-M.
Comment 4 John Arthorne CLA 2013-03-13 09:24:07 EDT
It is still failing with a timeout:

Mar 12, 2013 4:28:12 PM org.apache.commons.httpclient.HttpMethodDirector executeWithRetry
INFO: I/O exception (java.net.ConnectException) caught when processing request: Connection timed out: connect
Mar 12, 2013 4:28:12 PM org.apache.commons.httpclient.HttpMethodDirector executeWithRetry
INFO: Retrying request

I am going to clear out the workspace and try fresh just to make sure there is nothing stale around.
Comment 5 David Williams CLA 2013-03-15 00:57:19 EDT
Created attachment 228455 [details]
Patch to set proxy info in base builder

I've been doing some testing on Mac and Windows and WOW, is it complicated! 

But, I think, in short, even though "downloads" was changed to "a local address", the basebuilder/p2 knows nothing about not using proxies for that address. In fact, it doesn't know to use proxies at all (otherwise, it would have been working previously, just going through downloads to get stuff ... and not sure if your tests go anywhere external, or not). A 'regular' version of Eclipse (at least a 4.2.2. level one) can detect the "system proxies" (at least on Windows, at least using Java 7, but Mac does not, or they are not set) but ... base builder does not ever do that, I don't think, on any platform, on any version of Java. In principle, you could change the .../settings/..net.prefs in base builder itself (and tag a new version, etc.) but ... putting in in your bat script should work too ... and maybe easier to change in future, just in case there are future changes to proxies :) 

Anyway, I tested the attached patch on Windows and Mac and it works as expected. If, on the other hand, I "force" the settings that base builder has (namely, nothing except 

eclipse.preferences.version=1
org.eclipse.core.net.hasMigrated=true

then I get the "infinite retry" that you are seeing. (I was using eclipse binary platform 4.2.2 for my testing, not the base builder per se). 
Hope this helps, 
Good luck!
Comment 6 David Williams CLA 2013-03-20 12:16:11 EDT
Created attachment 228722 [details]
current best guess of proxy settngs

I still see the Orion Windows test "glowing red" for long periods of time. (you might also enable the "maximum time limit" for that job, if you haven't tried that yet. 

But, in any case, I think this patch has better "default" settings for using proxies on Hudson's Window machine. It basically says to "use the system proxies" (which are set via Windows Internet Connection settings). I think the webmasters keep it "up to date" and that's where they do the tweaking to work around issue in bug 401964. 

The values set for "base builder" via this patch should work before and after that infrastructure bug is fixed. 

[I tried the "manual settings" in previous patch in our Platform tests and p2 did ok, but it caused some CVS access problems then, for some reason, so eventually ended up with this simpler version].
Comment 7 John Arthorne CLA 2013-03-21 12:42:23 EDT
Thanks David, I'll try anything!

http://git.eclipse.org/c/orion/org.eclipse.orion.server.git/commit/?id=9f638b099e83feec1a00d57fd31e72df02bfda7b

Also set the job to timeout after 10 minutes. on a good day it should only take a minute.
Comment 8 John Arthorne CLA 2013-03-22 11:02:06 EDT
It's now failing in a different way. Evidently some vestigial CVS references need cleaning up:

c:\hb\workspace\orion-jstest-windows>IF NOT EXIST c:\hb\workspace\orion-jstest-windows\org.eclipse.releng.basebuilder cvs -Q -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse ex -r v20110302 -d org.eclipse.releng.basebuilder org.eclipse.releng.basebuilder 
cvs [export aborted]: unrecognized auth response from dev.eclipse.org: cvs [pserver aborted]: /cvsroot/eclipse: no such repository
Comment 9 John Arthorne CLA 2013-06-17 09:43:29 EDT
The Orion Hudson tests are still failing due to some kind of communication problem. We will need to revisit after 3.0. For now I am removing the tests.

c:\hb\workspace\orion-jstest-windows>"c:\java\jdk1.6.0_20\jre\bin\java" -Xmx500m -jar c:\hb\workspace\orion-jstest-windows/org.eclipse.releng.basebuilder/plugins/org.eclipse.equinox.launcher.jar -application org.eclipse.equinox.p2.director -repository http://build.eclipse.org/eclipse/e4/orion/target/integration/I20130616-2230 -i org.eclipse.orion -d c:\hb\workspace\orion-jstest-windows/test/eclipse 
Jun 17, 2013 4:17:43 AM org.apache.commons.httpclient.HttpMethodDirector executeWithRetry
INFO: I/O exception (java.net.ConnectException) caught when processing request: Connection timed out: connect
Jun 17, 2013 4:17:43 AM org.apache.commons.httpclient.HttpMethodDirector executeWithRetry
Comment 10 John Arthorne CLA 2013-06-18 11:43:34 EDT
This is the commit that comments out the hudson tests for 3.0:

http://git.eclipse.org/c/orion/org.eclipse.orion.server.git/commit/?id=3ab47f43d20c6adb1ab1b24fd9782c23febc693c
Comment 11 John Arthorne CLA 2015-05-05 16:21:29 EDT
Closing as part of a mass clean up of inactive bugs. Please reopen if this problem still occurs or is relevant to you. For more details see:


https://dev.eclipse.org/mhonarc/lists/orion-dev/msg03444.html