implement two changes to better handle, identify, and recover from

[bootmanager.git] / source / steps / ChainBootNode.py

diff --git a/source/steps/ChainBootNode.py b/source/steps/ChainBootNode.py
index a83474a..d9e7226 100644 (file)
--- a/source/steps/ChainBootNode.py
+++ b/source/steps/ChainBootNode.py
@@ -1,24 +1,32 @@
 import string
 import re
 
+import InstallWriteConfig
+import UpdateBootStateWithPLC
 from Exceptions import *
 import utils
 import compatibility
 from systeminfo import systeminfo
 import BootAPI
+import notify_messages
 
 
 def Run( vars, log ):
     """
     Load the kernel off of a node and boot to it.
     This step assumes the disks are mounted on SYSIMG_PATH.
+    If successful, this function will not return. If it returns, no chain
+    booting has occurred.
     
     Expect the following variables:
     BOOT_CD_VERSION       A tuple of the current bootcd version
     SYSIMG_PATH           the path where the system image will be mounted
                           (always starts with TEMP_PATH)
     ROOT_MOUNTED          the node root file system is mounted
-
+    NODE_SESSION             the unique session val set when we requested
+                             the current boot state
+    PLCONF_DIR               The directory to store PL configuration files in
+    
     Sets the following variables:
     ROOT_MOUNTED          the node root file system is mounted
     """
@@ -34,7 +42,14 @@ def Run( vars, log ):
         SYSIMG_PATH= vars["SYSIMG_PATH"]
         if SYSIMG_PATH == "":
             raise ValueError, "SYSIMG_PATH"
-        
+
+        PLCONF_DIR= vars["PLCONF_DIR"]
+        if PLCONF_DIR == "":
+            raise ValueError, "PLCONF_DIR"
+
+        # its ok if this is blank
+        NODE_SESSION= vars["NODE_SESSION"]
+
     except KeyError, var:
         raise BootManagerException, "Missing variable in vars: %s\n" % var
     except ValueError, var:
@@ -83,11 +98,27 @@ def Run( vars, log ):
         ssh_host_key_file= None
     except IOError, e:
         pass
-    
+
+    # write out the session value /etc/planetlab/session
+    try:
+        session_file_path= "%s/%s/session" % (SYSIMG_PATH,PLCONF_DIR)
+        session_file= file( session_file_path, "w" )
+        session_file.write( str(NODE_SESSION) )
+        session_file.close()
+        session_file= None
+        log.write( "Updated /etc/planetlab/session\n" )
+    except IOError, e:
+        log.write( "Unable to write out /etc/planetlab/session, continuing anyway\n" )
+
     update_vals= {}
     update_vals['ssh_host_key']= ssh_host_key
     BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
 
+    # rewrite modprobe.conf in case there were any module changes
+    # from a new kernel installed.
+    log.write( "Rewriting /etc/modprobe.conf\n" )
+    (network_count,storage_count)= \
+             InstallWriteConfig.write_modprobeconf_file( vars, log )
 
     log.write( "Copying kernel and initrd for booting.\n" )
     utils.sysexec( "cp %s/boot/kernel-boot /tmp/kernel" % SYSIMG_PATH, log )
@@ -102,6 +133,20 @@ def Run( vars, log ):
     ROOT_MOUNTED= 0
     vars['ROOT_MOUNTED']= 0
 
+    # before we do the real kexec, check to see if we had any
+    # network drivers written to modprobe.conf. if not, return -1,
+    # which will cause this node to be switched to a debug state.
+    if network_count == 0:
+        log.write( "\nIt appears we don't have any network drivers. Aborting.\n" )
+        
+        vars['BOOT_STATE']= 'dbg'
+        vars['STATE_CHANGE_NOTIFY']= 1
+        vars['STATE_CHANGE_NOTIFY_MESSAGE']= \
+                          notify_messages.MSG_NO_DETECTED_NETWORK
+        UpdateBootStateWithPLC.Run( vars, log )
+        
+        return
+
     log.write( "Unloading modules and chain booting to new kernel.\n" )
 
     # further use of log after Upload will only output to screen