[rabbitmq-discuss] getting started, broker runs; can't get status

[Dmitriy Samovskiy]

Dave, Matthias, Christopher -

Dave Farkas wrote:
> I don't have snoopy installed, anything in /etc/ld.so.preload or can
> find a place where LD_PRELOAD is being set. SELinux is also disabled:
> 
> I've also tried to re-enable Nagle by changing the lines in the
> rabbitmq-server script with the same results.
> 
Even though this is unlikely to help, I was wondering if you could find time to do the 
following exercise. It may help determine if an issue has anything to do with rabbit in 
the first place (to make sure we are on the right track).

Please run this:

# erl -sname foo -cookie coo
Erlang (BEAM) emulator version 5.6.5 [source] [async-threads:0] [kernel-poll:false]

Eshell V5.6.5  (abort with ^G)
(foo at myvm)1> net_adm:names().
{ok,[{"foo",1292}]}
(foo at myvm)2>

Exit from there with Ctrl+C Ctrl+C.


I suspect that your response from net_adm:names() will be {error,timeout} and it will 
appear not immediately but after some time (within 30 seconds). Could you please confirm.

After you exit from erl, please do grep -r /usr/lib/erlang/erts-5.6.5/bin /var/log/*
(please replace 5.6.5 with emulator version that is displayed when you start erl). 
Anything of interest in the output? I am particularly looking for something related to 
auth, or access being denied. Interesting lines are likely to come from auth.log or its 
equivalent on your system, and there should be many similar lines (unless your syslog 
suppresses dup lines - mine didn't). If nothing shows up, maybe try similar greps - 
erlang, erts.

And finally, if you have or can get strace on the target system, could you please run this:

% strace -e trace=write -o erl_strace.log erl -sname -cookie coo


and do the same net_adm:names() in erlang shell. When it times out, exit erlang shell and 
take a look at erl_strace.log. I expect that at the end of that file you will see many 
lines like this:

--- SIGCHLD (Child exited) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---
--- SIGCHLD (Child exited) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---
--- SIGCHLD (Child exited) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---
--- SIGCHLD (Child exited) @ 0 (0) ---
--- SIGPIPE (Broken pipe) @ 0 (0) ---

Feel free to do strace with -e trace=all or -e verbose=all.


That SIGPIPE corresponds to erlang's attempt to write PORT2_REQ command to empd in the 
second connection. tcpdump shows no attempts to make a second connection either - not even 
a single SYN. It tells me that either something suppresses the connection but still has 
low-level C function return success, or something in erlang does not check return code (a 
very wild guess on my part).

All in all, I can reliably recreate this problem with snoopy on Debian Etch (if you try 
this at home, I recommend against having dpkg install snoopy in /etc/ld.so.preload - it 
will ask at install, but answer NO and instead manually set LD_PRELOAD=/lib/snoopy.so when 
you need to).

At least in scenario with snoopy, the problem has nothing to do with rabbitmq.

If there are strace or low-level OS experts on the list who would like to compare strace 
output of a good run vs run under snoopy in order to get more info, I can provide strace 
outputs.

I have also looked inside lib/erl_interface/src/epmd in erlang source for clues but did 
not find any.

What Linux distro are you using? uname -a? Any particular details how you installed the 
OS? If it's safe to share, maybe output of "rpm -qa" or "dpkg -l"?


And like Matthias said before, access to box or image of the box will be most helpful.


- Dmitriy