edwardt / ejabberd-trace

Easy tracing of users connected to ejabberd

ejabberd-trace

This is an extremely simple utility for easy tracing of connections made to ejabberd XMPP server.

In the future it might be expanded but just as well it might be abandoned due to proving to be useless.

The project is in rapid development - you have been warned. (Read: I don't care for API backwards compatibility for now).

Getting started

First, please start the application. This makes it auto-detect server string type in use.

> application:start(ejabberd_trace).

Tracing a connected user by JID

Trace a user who is connected to an ejabberd node:

> ejabberd_trace:user("asd@localhost/a-resource").
> % or
> ejabberd_trace:user("asd@localhost").

If you don't specify a resource and there are multiple available:

> ejabberd_trace:user("asd@localhost").
{error,{multiple_sessions,[{{"asd","localhost","psi"},
                            <0.327.0>},
                           {{"asd","localhost","x3"},<0.307.0>}]}}
> ejabberd_trace:user("asd@localhost/x3").

Tracing a user that is going to connect soon by JID

It's possible to trace a user who is not connected yet but will connect in a short period of time using only the JID.

It's useful in case you need to trace all communication (including connection initiation) happening between a c2s process and other parts of the system (including TCP connection) - this variant requires specifying the resource.

> ejabberd_trace:new_user("asd@localhost/x4").

Keep in mind this might be heavy on the system - until it's known which of the connected processes is the one you want to trace all newly connecting c2s processes are traced and the traces cached. Once it's known (by inspecting the XMPP stream) which process is the one to be traced its trace cache is flushed to the trace handler; the rest of the trace cache is discarded.

Display c2s process state by JID

> ejabberd_trace:state("asd@localhost").

Display incoming/outgoing XMPP streams for a given user

This gives a complete pretty-printed log of the whole client to server connection for a single full JID:

> ejabberd_trace:new_user("alice@localhost/escalus-default-resource",
                          ejabberd_trace_filter:stream(),
                          fun ejabberd_trace_format:stream/2).

Example output:

in :
<stream:stream to='localhost' version='1.0' xml:lang='en' xmlns='jabber:client'
               xmlns:stream='http://etherx.jabber.org/streams'>

out :
<?xml version='1.0'?>
<stream:stream xmlns='jabber:client' xmlns:stream='http://etherx.jabber.org/streams'
               id='1217298371' from='localhost' version='1.0' xml:lang='en'>

out :
<stream:features>
  <mechanisms xmlns='urn:ietf:params:xml:ns:xmpp-sasl'>
    <mechanism>DIGEST-MD5</mechanism>
    <mechanism>PLAIN</mechanism>
    <mechanism>SCRAM-SHA-1</mechanism>
  </mechanisms>
  <c ver='mfN6SdQ3DGO7/QUHHftElVDFZ7k='
     node='http://www.process-one.net/en/ejabberd/'
     hash='sha-1'
     xmlns='http://jabber.org/protocol/caps'/>
  <register xmlns='http://jabber.org/features/iq-register'/>
  <sm xmlns='urn:xmpp:sm:3'/>
</stream:features>

in :
<auth mechanism='PLAIN' xmlns='urn:ietf:params:xml:ns:xmpp-sasl'>AGFsaWNlAG1ha290YQ==</auth>

out :
<success xmlns='urn:ietf:params:xml:ns:xmpp-sasl'/>

...

Filters

It's possible to filter the traces caught for the traced process. The available filters are:

• raw_traces - all raw trace messages
• dbg - all trace messages as formatted by dbg
• tx - messages sent on socket as formatted by dbg
• rx - messages received on socket as formatted by dbg
• routed_out - messages routed to the c2s process from other parts of the server
• routed_in - messages routed by the c2s process to other part of the server

ToDo

• Implement routed_out filter.

• Implement routed_in filter.

• Make filter and formatter choices independent. Filter should only filter traces and pass them to a formatter.

• Make filters composable. It should be possible to display routed_in and tx traces simultaneously.

  There are two combinators available: `any` and `all`.
  

• Sort out logging. There are debug messages logged unconditionally right now, e.g.:

    >>>>> caught send_element
    >>>>> found no Jid
  
  It would be nice to keep them, but not display all the time.
  It should also be possible to print the traces to a file.
  
  These are only compiled in when the `debug` compile flag is set.
  

• Trace multiple subsequent connections of a given user. A test story may involve the same full jid connecting more than once to fulfill one scenario.

• Enable functioning in a distributed environment.

  This seems to work. Sometimes.
  

• In case of BOSH users the trigger might happen on two distinct events: the call in c2s process or sending the message from BOSH socket. Find a way to discriminate one (store opts in trace server ets? and verify in do_trace_user?).

• Auto-detect server string type - there's no reason to make user set this manually. This is done when application:start/1 is called, but maybe it could be even more automatic?