I have narrowed this down to 1.535 and up; 1.534 is fine. This issue only occurs when "TCP port for JNLP slave agents" is set to Disabled.

My best guess is that the "two simultaneous HTTP connections" fallback for the CLI connection was broken with the introduction of Jetty8 in 1.535.

From the jenkins-cli output (similar to 2.log), one can with some effort decode the server side backtrace:

java.io.IOException: Protocol violation: Unexpected single newline character in chunk size
     at hudson.util.ChunkedInputStream.getChunkSizeFromInputStream(ChunkedInputStream.java:246)
     at hudson.util.ChunkedInputStream.nextChunk(ChunkedInputStream.java:197)
     at hudson.util.ChunkedInputStream.read(ChunkedInputStream.java:120)
     ...

For the original submitter: you can workaround by re-enabling the JNLP port.

git bisect narrows this down to [b19f4650d7cc146e28398a33ad0625d46991e27f] [FIXED JENKINS-18366] Winstone is now powered by Jetty

(as I expected).

revision 99ccd7f7228ae2f846642f23f81ceb1725f61af6 again changes the behaviour - from the error situation described above to "No X-Jenkins-CLI2-Port among [X-Jenkins, null, X-Hudson, Content-Length, Expires, Set-Cookie, X-SSH-Endpoint, Server, Cache-Control, X-Hudson-Theme, X-Instance-Identity, X-Jenkins-Session, Content-Type]", apparently (implicitly?) deprecating non-JNLP CLI usage. Is this intentional?

Code changed in jenkins
User: Kohsuke Kawaguchi
Path:
test/src/test/groovy/hudson/cli/CLIActionTest.groovy
http://jenkins-ci.org/commit/jenkins/1b6ba71f9cf8ef037d7ff6522105622eff2c1f5a
Log:
JENKINS-20128 reproduced the problem

Code changed in jenkins
User: Kohsuke Kawaguchi
Path:
changelog.html
core/src/main/java/hudson/cli/CLIAction.java
http://jenkins-ci.org/commit/jenkins/9f1796e3ea8034e4a45d63e706b1c087b8f7f2e8
Log:
[FIXED JENKINS-20128]

CLI hits "/cli", so the server needs to respond to this endpoint.

Compare: https://github.com/jenkinsci/jenkins/compare/cab3d5daa7d8...9f1796e3ea80

Integrated in jenkins_main_trunk #3036

Result = SUCCESS

What was the actual cause of the regression? JENKINS-18366?

Yes - see git bisect mention in an earlier comment.

Solaris 10

• CLI times out for any command beginning at 1.535
• Last working version for CLI was Jenkins 1.534
• Please see https://issues.jenkins-ci.org/browse/JENKINS-18366?focusedCommentId=193633&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-193633

@sstack the fix is in 1.540 (and will be in 1.532.2).

I am still seeing "Protocol violation: Unexpected single newline character in chunk size' 0x0a" on Jenkins 1.549. Can you re-open?

@Jesse, 1.540 CLI is broken on Solaris.

Feb 3, 2014 7:13:41 AM hudson.remoting.SynchronousCommandTransport$ReaderThread run
SEVERE: I/O error in channel Chunked connection to http://host.net:8080/cli
hudson.remoting.DiagnosedStreamCorruptionException

@sstack might be a different bug, best to file separately (link to this one).

@sstack - did you file a separate bug on this? I am also seeing this with v1.554.1 and the issue was not present when we were on v1.534

Same problem on 1.554.3. Is there a separate bug filed or not?