Skip to main content

Analysis of network issues with tcpdump and wireshark

I've recently had to deal with the application connectivity issue (details on stackoverflow) that appeared after the migration of the application to a new server. It resulted in "Connection timed out" Java exceptions in certain cases. The answer was on the surface but I didn't know where to look at exactly. So I had to investigate and apply network sniffing tools such as tcpdump and Wireshark. Here I'd like to share my experience with the network analysis.
  1. The issue and the cause.
  2. Wireshark.
  3. Tcpdump.
The issue and the cause
The following exception was thrown by Saxon XSLT processor when the document function was invoked:
Caused by: org.apache.commons.lang.exception.NestableRuntimeException: net.sf.saxon.trans.DynamicError: net.sf.saxon.trans.DynamicError: Connection timed out
        at Method)
It clearly shows that some resource was not accessible although I could easily access the URL that was an argument of the document function. Finally it appeared that there was a DOCTYPE declaration of a DTD resource in the target xml file. So Saxon apparently failed when it could not access that DTD to perform the validation. To find out this root cause I had to use network tools.

First of all, I tried to reproduce the issue on my developer's Windows-based laptop but could not do this that confirmed it's a server configuration issue. However, before debugging on QA environment I analyzed the application network activity locally with Wireshark that appears to be the most popular network packet analyzer with GUI. This tool provides numerous filtering options so you can locate whatever you want on the network. It helped me to capture the HTTP request sent by Saxon and to see all request headers. Afterwards I simulated the identical request with the same headers with wget and curl on the QA environment. But generally it didn't help so I had to move to the QA environment.

On the QA environment we have RHEL system installed so tcpdump appeared to be the best fit. It is another very popular network packet analyzer but only command-line. To investigate the issue further, I've used tcpdump to record the network activity for the current host for two scenarios: sending suspicious HTTP request with curl and doing it via the application itself. Curl worked fine, so the issue proved to be application-related. Indeed curl did not send another request to fetch the DTD file while Saxon did. So this is how curl command looked like:
curl -v -H "Pragma: no-cache" -H "User-Agent: Java/1.6.0_21" -H "Cache-Control: no-cache" -H "Accept: text/html, image/gif, image/jpeg, *; q=.2, */*; q=.2" -H "Connection: keep-alive" http://suspicious-request-URL-here
This is how tcpdump command looked like:
sudo /usr/sbin/tcpdump -i eth2 -s 512 -l -A host
It prints all the requests coming to and from the specified host. I'm not pasting those results here as they are too big. But having compared the tcpdump outputs for two scenarios, I've found out there were additional requests sent from the application. It did not reveal the exact URL though but it helped me to guess the root cause and locate that DTD declaration. To conclude, the network packet analysis can be very useful for debugging. I'm pretty confident with client-side tools built into browsers (e.g. Firebug). As for the server-side, I'm still not an expert here and it may be possible to discover much more details with the tools I mentioned above.


Popular posts from this blog

Connection to Amazon Neptune endpoint from EKS during development

This small article will describe how to connect to Amazon Neptune database endpoint from your PC during development. Amazon Neptune is a fully managed graph database service from Amazon. Due to security reasons direct connections to Neptune are not allowed, so it's impossible to attach a public IP address or load balancer to that service. Instead access is restricted to the same VPC where Neptune is set up, so applications should be deployed in the same VPC to be able to access the database. That's a great idea for Production however it makes it very difficult to develop, debug and test applications locally. The instructions below will help you to create a tunnel towards Neptune endpoint considering you use Amazon EKS - a managed Kubernetes service from Amazon. As a side note, if you don't use EKS, the same idea of creating a tunnel can be implemented using a Bastion server . In Kubernetes we'll create a dedicated proxying pod. Prerequisites. Setting up a tunnel.

Notes on upgrade to JSF 2.1, Servlet 3.0, Spring 4.0, RichFaces 4.3

This article is devoted to an upgrade of a common JSF Spring application. Time flies and there is already Java EE 7 platform out and widely used. It's sometimes said that Spring framework has become legacy with appearance of Java EE 6. But it's out of scope of this post. Here I'm going to provide notes about the minimal changes that I found required for the upgrade of the application from JSF 1.2 to 2.1, from JSTL 1.1.2 to 1.2, from Servlet 2.4 to 3.0, from Spring 3.1.3 to 4.0.5, from RichFaces 3.3.3 to 4.3.7. It must be mentioned that the latest final RichFaces release 4.3.7 depends on JSF 2.1, JSTL 1.2 and Servlet 3.0.1 that dictated those versions. This post should not be considered as comprehensive but rather showing how I did the upgrade. See the links for more details. Jetty & Tomcat. JSTL. JSF & Facelets. Servlet. Spring framework. RichFaces. Jetty & Tomcat First, I upgraded the application to run with the latest servlet container versio

Extracting XML comments with XQuery

I've just discovered that it's possible to process comment nodes using XQuery. Ideally it should not be the case if you take part in designing your data formats, then you should simply store valuable data in plain xml. But I have to deal with OntoML data source that uses a bit peculiar format while export to XML, i.e. some data fields are stored inside XML comments. So here is an example how to solve this problem. XML example This is an example stub of one real xml with irrelevant data omitted. There are several thousands of xmls like this stored in Sedna XML DB collection. Finally, I need to extract the list of pairs for the complete collection: identifier (i.e. SOT1209 ) and saved timestamp (i.e. 2012-12-12 23:58:13.118 GMT ). <?xml version="1.0" standalone="yes"?> <!--EXPORT_PROGRAM:=eptos-iso29002-10-Export-V10--> <!--File saved on: 2012-12-12 23:58:13.118 GMT--> <!--XML Schema used: V099--> <cat:catalogue xmlns:cat=