Mod Cluster encoding Issue - Urgent

[sandy2487]

We are using mod_cluster for load balancing out application. We have multiple servers connected within the application. Suppose , Server 1 and Server 2 are running Jboss , when the requests from Server 1 goes to  Server 2 , it goes through mod cluster.  Server 2 is exposing REST Web  Services.

We are using @PathVariable of Spring Framework to get the path values from the URL being sent from Server 1 on Server 2.  When we use a value like " ー " in the URL. Mod Cluster does not decode it correctly.

This character is encoded by spring at Server 1 and then the URL is sent to Server 2. Spring Framework on Server 2 will decode this and process the request. Mod Cluster is in between the requests and it does not correctly decode the UTF 8 encoding done on Server 1. Hence , the issue.

We removed the mod cluster from between and made direct calls and the issue did not happen. So we are pretty sure it is mod cluster doing something in between.

Please suggest.

[jfclere]

"ー" Doesn't look like something valid for an URL.

your description is too vague to be investigated could you come with a small example?

[sandy2487]

Jean,

"ー" does not look like a valid string in URL but it is being encoded to UTF 8 before being sent to mod_cluster. 

So mod_cluster just has to not alter the encoding which it is doing.

For Example,

Original URL : /images/Pepーsi.jpg/?productIdType=ABC

Encoded URL : /images/Pep%E3%83%BCsi.jpg/?productIdType=ABC

URL Transformed by Mod_Cluster : /images/Pepã¼si.jpg/?productIdType=ABC

Please let me know if anything else is required.

[jfclere]

Where is the URL, in an html page?

[sandy2487]

It is not in the HTML . It is being used to redirect requests between controllers.

[jfclere]

Probably you need to have nocanon in the ProxyPass directive, you should try something like:

ProxyPass / balancer://mybalancer/ nocanon

could you debug mod_cluster (LogLevel debug) in httpd.conf?

[rhusar]

Very interesting problem. For what its worth, if its possible I would avoid similar characters in URLs and canonize the URLs to ASCII or so (especially if they are accessible from the web).

It looks to me that mod_cluster converts the characters eagerly somewhere, i.e. %E3 being ã, %83 being ƒ and %BC being ¼ (in your post the 2nd character is missing Re: Mod Cluster encoding Issue - Urgent which is either a typo or the problem is even more interesting) instead of resolving to correct Japanese character which is represented by those three.

[mbabacek]

There is no such error, at least not in mod_cluster 1.2.6.

This is an output from some tests:

• Request GET: http://10.16.88.19:8847/clusterbench/requestinfo?image=Pep%E3%83%BCsi.jpg&productIdType=ABC
• Response from test Servlet:
  Query string: image=Pep%E3%83%BCsi.jpg&productIdType=ABC
  Query string UTF-8 decoded: image=Pepーsi.jpg&productIdType=ABC
  JVM route: jboss-eap-6.2
  Session ID: Y-fvBhB0yHirbQCCTLz2mfFM.jboss-eap-6.2

• Request GET: http://10.16.88.19:8847/clusterbench/requestinfo?%E2%98%B9=%E2%98%BB
• Response from test Servlet:
  Query string: %E2%98%B9=%E2%98%BB
  Query string UTF-8 decoded: ☹=☻
  JVM route: jboss-eap-6.2
  Session ID: GpJk6J91Rh2KvOmsamcZNURM.jboss-eap-6.2

10.16.88.19:8847 is Apache HTTP Server 2.2.22 with mod_cluster 1.2.6.

BTW: request.getParameterMap() would give some gibberish, because it's not automatically decoded...

Cheers