It is worth noting that the self identification is not perfect. It gathers a few different identities for itself; including localhost, the output of hostname, and a non loopback output of ipconfig. However it is possible to have a machine configured where that is not sufficient so start-all works but start-here does not.
Sent from my phone, please pardon the typos and brevity. On May 6, 2013 8:04 AM, "Aaron" <[email protected]> wrote: > Ahhh cool...so start-all.sh & start-here.sh are "smart" and by that I > mean, will only start the certain processes based upon which machine they > are on. And if you do a start-here.sh on a box, it will look through every > file, and only start the processes for which it "finds itself." > > So, if a machine is in master & gc, but not in tracer....and when I call > start-here.sh on that box, it will only start the master & gc > processes..and NOT tracer? Assuming the answer is yes, but just confirming. > > For slave, does that assume 2 processes: tserver & logger? so, if a > machine is in the slave file, it will start both of those? > > > > On Fri, May 3, 2013 at 4:06 PM, John Vines <[email protected]> wrote: > >> Those files are used for 2 purposes. >> 1. Start-all.sh, start-here.sh, and tup.sh read all of the files (tup >> only uses slaves) to start up the processes. Start-all will go through each >> file for each process, ssh to that server (no ssh if the current host) and >> start the process. Start-all uses tup for the tserver and logger processes. >> Start-here reads through each file, and if the machines match, it will >> start that process locally. >> >> 2. The files are used in attempts to bind to a port. This is for when >> your on a machine with multiple ports, to help it bind on the port you >> want. This MAY only actually occur if you're using the above mentioned >> scripts. >> >> So that said, as long as the configuration of those files is correct for >> the node you're on, you can run them there. And on that same note, the >> reason you want to keep those various server list files synced is so you >> can run start-all/tup from any node and have the same effect. >> >> >> On Fri, May 3, 2013 at 3:47 PM, Aaron <[email protected]> wrote: >> >>> Was curious if anyone knew of some decent write-ups on various shell >>> scripts? Also, curious about master, slaves, gc, tracers, etc conf files. >>> Main reason I ask is about what .sh to run "where." >>> >>> For example, have "head" system that is running >>> master/gc/monitor/tracer, then we have other various nodes running >>> tserver/logger. >>> >>> 1a. Does every node, including the "head" node need all the conf files >>> the same...masters/slaves/tracers, etc? >>> >>> 1b. Or does the "master" node only need to know about where the slaves >>> are, so it uses the slaves file. And conversely, do the tserver nodes only >>> need to have a master conf file to put themselves to the master? >>> >>> 2. I saw someplace not to use tup.sh on the compute/storage nodes...do >>> I just use start-server.sh to only bring up the logger/tserver..and NOT the >>> other services? >>> >>> Basically, there are more questions like this, but, wanted to ask if >>> anyone out there has some write-ups to I can RTFM first. Basically, >>> looking for a little more into the weeds that the >>> Administration/Installation section of the manual. >>> >>> Thanks in advance, >>> Aaron >>> >> >> >
