Using JBoss 5.1.0.GA, mod_cluster 1.2.0.Final and Spring based web services ( REST template ), shows the following in the error_log. These web services return mixed data ( text - html and binary - images )

[Fri Nov 16 13:24:58 2012] [error] ajp_handle_cping_cpong: awaited CPONG, received 3

[Fri Nov 16 13:24:58 2012] [error] (20014)Internal error: proxy: AJP: cping/cpong failed to 10.1.1.85:8009 (10.1.1.85)

[Fri Nov 16 13:24:58 2012] [error] ajp_handle_cping_cpong: awaited CPONG, received 3

[Fri Nov 16 13:24:58 2012] [error] (20014)Internal error: proxy: AJP: cping/cpong failed to 10.1.1.85:8009 (10.1.1.85)

[Fri Nov 16 13:24:58 2012] [error] ajp_handle_cping_cpong: awaited CPONG, received 3

[Fri Nov 16 13:24:58 2012] [error] (20014)Internal error: proxy: AJP: cping/cpong failed to 10.1.1.85:8009 (10.1.1.85)

[Fri Nov 16 13:24:58 2012] [error] ajp_handle_cping_cpong: awaited CPONG, received 3

[Fri Nov 16 13:24:58 2012] [error] (20014)Internal error: proxy: AJP: cping/cpong failed to 10.1.1.85:8009 (10.1.1.85)

[Fri Nov 16 13:24:58 2012] [error] ajp_handle_cping_cpong: awaited CPONG, received 3

[Fri Nov 16 13:24:58 2012] [error] (20014)Internal error: proxy: AJP: cping/cpong failed to 10.1.1.85:8009 (10.1.1.85)

The load balancer and cluster node are configured to use AJP. 

The mod_cluster configuration includes flushPackets (set to true), flushWait (set to 1) to remove the additional 3 bytes, but the error seems to persist.

The end-user symptom is that some of the images are shown, while some are not. Doing a refresh on the browser, results in a seemingly random display of images ( some images get displayed, others don't)

There are a number of related threads on this topic, however, none seem to be conclusive

https://community.jboss.org/message/606285#606285

https://community.jboss.org/message/610085#610085

https://community.jboss.org/thread/162578?start=15&tstart=8

Any suggestions on why this is happening ?

I did try to use HTTP instead of AJP, by commenting out the AJP connector details in the server.xml, restarting both the load balancer and the cluster node, the error mentioned above does not come up, but the end-user symptom still persists. The flushWait, flushPackets are not changed, but they would anyways not apply to HTTP.