Hi Michael,
The new autocruncher is ready for the big recrunch. It is in
/home/daq/blast/autocruncher directory. I tried to implement
everything I thought necessary for the recrunch. If I left
anything else out, please let me know.
Also, in the process, the lrn needed a minor change, and lrd
was changed to take advantage of the autocruncher. I committed
the changes into CVS, and tagged them as v3_4_11 (I think),
and made the necessary links in bin directory to make them
default for the blast account.
Here are the summary of changes seen by the user (This is part
of the README file located in /home/daq/blast/autocruncher/src):
- A_serverlist is removed. Use A_config instead. See the
example A_config file in this directory.
- It is possible to export blast environments to the remote servers.
This is configured through A_config, and it is useful for
different set of recrunch jobs, if you want to route the output
files, and use different PARAM directory.
- It is possible to request more than one job per run
(example lrn, then root charge.C). This is configured
through A_config.
- It recognizes #####-##### run number range notation in A_runlist
file, and sends the runs one-by-one. The A_runlist file is also
reformatted, whenever user changes it. A_runlist's inode does
not change after run submissions (this used to confuse emacs.)
- There is no hard coded path, everything is defined according
to the current path where you start the autocruncher. You
can copy the whole autocruncher directory, and start it at
that point. The daemons running on the servers will aware of this
automatically. This is handy for the next item.
- It is possible to run more than one autocruncher sessions. They
should be completely independent. Each autocruncher sees what
is running on the servers, but takes actions only for those
submitted by itself.
- It is possible to share the servers between several autocruncher
sessions; all of them will obey the process limits. Scheduling is
handled by "first-come-first-serve" basis. On average they should
get equal shares. The only requirement is having the complete
autocruncher files in a different directory. Just copy with
"cp -a ~blast/autocruncher newplace ; cd newplace", configure
the A_config file, and start it. It is recommended to have
different labels for the jobs (see A_config COMMAND) to
distinguish the jobs visually (even this is not a requirement,
as the owned processes are marked as '*' on the screen.)
- The daemons running on servers quit automatically after
autocruncher exits, and restarted when the autocruncher is
started. The information regarding the ongoing processes
are preserved.
- And, a minor improvement: It shows the number of runs waiting
in the queue.
- Complete source code is in /home/daq/blast/autocruncher/src
And, off course I should add that, I tested this and I expect it
to be working just fine, but in case it behaves badly under heavy
load, let me know.
Cheers,
Taylan
-- ---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--- Taylan Akdogan Massachusetts Institute of Technology akdogan@mit.edu Department of Physics Phn:+1-617-258-0801 Laboratory for Nuclear Science Fax:+1-617-258-5440 Room 26-559 ---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=---
This archive was generated by hypermail 2.1.2 : Mon Feb 24 2014 - 14:07:32 EST