oops, too much messing about

[nodemanager.git] / plugins / omf_resctl.py

diff --git a/plugins/omf_resctl.py b/plugins/omf_resctl.py
index 7d82e79..4f6d5b1 100644 (file)
--- a/plugins/omf_resctl.py
+++ b/plugins/omf_resctl.py
@@ -13,7 +13,8 @@ import subprocess
 import tools
 import logger
 
-priority = 50
+# we need this to run after sliverauth
+priority = 150
 
 def start():
     pass
@@ -22,15 +23,16 @@ def start():
 # hard-wire this for now
 # once the variables are expanded, this is expected to go into
 config_ple_template="""---
-# Example:
-# _slicename_ = nicta_ruby
-# _hostname_ = planetlab1.research.nicta.com.au
-# _xmpp_server_ = xmpp.planet-lab.eu
 # we extract expires time here, even in a comment so that the
 # trigger script gets called whenever this changes
 # expires: _expires_
- 
-:uid: _slicename_@_hostname_
+
+# these are not actual OMF parameters, they are only used by the trigger script
+:hostname: _hostname_
+:slicename: _slicename_
+
+# OMF configuration
+:uid: _slicename_%_hostname_
 :uri: xmpp://_slicename_-_hostname_-<%= "#{Process.pid}" %>:_slicename_-_hostname_-<%= "#{Process.pid}" %>@_xmpp_server_
 :environment: production
 :debug: false
@@ -47,6 +49,7 @@ yaml_slice_path="/etc/omf_rc/config.yml"
 # given that we're now responsible for fetching this one, I have to
 # decide on an actual path - not jsut a name to search for in PATH
 omf_rc_trigger_script="/usr/bin/plc_trigger_omf_rc"
+omf_rc_trigger_log="/var/log/plc_trigger_omf_rc.log"
 
 # hopefully temporary: when trigger script is missing, fetch it at the url here
 omf_rc_trigger_url="http://git.mytestbed.net/?p=omf.git;a=blob_plain;f=omf_rc/bin/plc_trigger_omf_rc;hb=HEAD"
@@ -111,17 +114,20 @@ def GetSlivers(data, conf = None, plc = None):
             try:
                 fetch_trigger_script_if_missing (slicename)
                 # the trigger script actually needs to be run in the slice context of course
+                # in addition there is a requirement to pretend we run as a login shell
+                # hence sudo -i
                 slice_command = [ "sudo", "-i",  omf_rc_trigger_script ]
                 to_run = tools.command_in_slice (slicename, slice_command)
-                logger.log("command_in_slice: %s"%to_run)
-                sp=subprocess.Popen(to_run, stdout=subprocess.PIPE,stderr=subprocess.STDOUT)
-                (out,err)=sp.communicate()
-                retcod=sp.returncode
-                # we don't wait for that, try to display a retcod for info purpose only
-                # might be None if that config script lasts or hangs whatever
-                logger.log("omf_resctl: %s: called OMF rc control script (imm. retcod=%r)"%(slicename,retcod))
-                logger.log("omf_resctl: got stdout\n%s"%out)
-                logger.log("omf_resctl: got stderr\n%s"%err)
+                log_filename = "/vservers/%s/%s"%(slicename,omf_rc_trigger_log)
+                logger.log("omf_resctl: starting %s"%to_run)
+                logger.log("redirected into %s"%log_filename)
+                logger.log("*not* waiting for completion..")
+                with open(log_filename,"a") as log_file:
+                    subprocess.Popen(to_run, stdout=log_file,stderr=subprocess.STDOUT)
+                # a first version tried to 'communicate' on that subprocess instance
+                # but that tended to create deadlocks in some cases
+                # causing nodemanager to stall...
+                # we're only losing the child's retcod, no big deal
             except:
                 import traceback
                 traceback.print_exc()