Hey Nick, The way traffic_manager, traffic_cop and traffic_server work together is that traffic_manager listens on all the ports and then passes a list of file descriptors via the -A command line argument to traffic_server. This is so if traffic_server dies the connections will not be rejected, instead they are still accepted and then when a new instance of traffic_server get's fired up the connections will then get served because the new instance will just use epoll or whatever on the old list of file descriptors. So with that I'm sure you can now understand why traffic_server would probably be dieing if you try to do this manually, you're giving it a bad list of file descriptors.
The -M command line argument is start traffic_server in management mode (i'm not 100% of what that means). Brian On Wed, Jan 18, 2012 at 9:48 AM, Nick Kew <[email protected]> wrote: > When I start trafficserver from the script, the actual program starts > with some commandline arguments. A freshly-installed trafficserver > shows something like: > > /usr/bin/traffic_server -M -A,7:X > > Trying to enable clustering using the exact instructions in TFM at > > http://trafficserver.apache.org/docs/trunk/admin/cluster-howto/index.en.html > leads to traffic_line telling me it's failed. On investigation, that's > because all trafficserver processes have exited, including traffic_cop > and traffic_manager. And they've exited silently, leaving no trace in the > logs. > This behaviour appears common to my own builds of 3.0 and 3.1, and to > the packaged version installed with yum on Amazon/ec2 servers, > which are someone else's build. > > I am able to restart trafficserver after this, but now the commandline is: > /usr/bin/traffic_server -M -A,10:X > > At this point I went hunting for any documentation of those > commandline arguments, but I haven't found it. > Any suggestions for where to look? > > -- > Nick Kew
