Changeset 5bf359d for fedd/federation/proxy_emulab_segment.py

Timestamp:

May 28, 2010 3:16:46 AM (14 years ago)

Author:

Ted Faber <faber@…>

Branches:

axis_example, compt_changes, info-ops, master, version-3.01, version-3.02

Children:

2f6820c

Parents:

06cc65b

Message:

More refactoring. Neaten up the code for creating segments in emulab and make the local and proxy class structures parallel. The code is more readable this way, I hope.

File:

• fedd/federation/proxy_emulab_segment.py (modified) (7 diffs)

fedd/federation/proxy_emulab_segment.py

@@ -14 +14 @@
 
 class start_segment(proxy_segment):
+    """
+    This starts an experiment on an emulab accessed remotely via ssh.  Most of
+    the experiment constuction has been done by the emulab_access object.  This
+    just does the wrangling of the emulab commands and collected the node to
+    physical mapping.  The routine throws service errors.
+    """
+
     def __init__(self, log=None, keyfile=None, debug=False):
         proxy_segment.__init__(self, log=log, keyfile=keyfile, debug=debug)
@@ -28 +35 @@
 
     def get_state(self, user, host, pid, eid):
+        """
+        Return the state of the experiment as reported by emulab
+        """
         # command to test experiment state
         expinfo_exec = "/usr/testbed/bin/expinfo"  
@@ -80 +90 @@
 
     def get_mapping(self, user, host, pid, eid):
+        """
+        Get the physical to virtual mapping from the expinfo command and save
+        it in the self.map member.
+        """
         # command to test experiment state
         expinfo_exec = "/usr/testbed/bin/expinfo"  
@@ -143 +157 @@
 
 
-    def __call__(self, parent, eid, pid, user, tclfile, tmpdir, timeout=0):
-        """
-        Start a sub-experiment on a federant.
-
-        Get the current state, modify or create as appropriate, ship data
-        and configs and start the experiment.  There are small ordering
-        differences based on the initial state of the sub-experiment.
-        """
-        # ops node in the federant
-        host = "%s%s" % (parent.ops, parent.domain)
+    def make_null_experiment(self, user, host, pid, eid, tmpdir):
+        """
+        Create a null copy of the experiment so that we capture any logs there
+        if the modify fails.  Emulab software discards the logs from a failed
+        startexp
+        """
+        try:
+            f = open("%s/null.tcl" % tmpdir, "w")
+            print >>f, self.null
+            f.close()
+        except EnvironmentError, e:
+            raise service_error(service_error.internal,
+                    "Cannot stage tarfile/rpm: %s" % e.strerror)
+
+        if not self.scp_file("%s/null.tcl" % tmpdir, user, host):
+            return False
+        self.log.info("[start_segment]: Creating %s" % eid)
+        timedout = False
+        try:
+            if not self.ssh_cmd(user, host,
+                    ("/usr/testbed/bin/startexp -i -f -w -p %s " + 
+                    "-e %s null.tcl") % (pid, eid), "startexp",
+                    timeout=60 * 10):
+                return False
+        except self.ssh_cmd_timeout:
+            timedout = True
+
+        if timedout:
+            state = self.get_state(user, host, pid, eid)
+            if state != "swapped":
+                return False
+        return True
+
+    def set_up_experiment_filespace(self, user, host, pid, eid, tmpdir):
+        """
+        Send all the software and configuration files into the experiment's
+        file space.  To reduce the number of ssh connections, we script many
+        changes and execute the script.
+        """
         # Configuration directories on the remote machine
         proj_dir = "/proj/%s/exp/%s/tmp" % (pid, eid)
@@ -159 +202 @@
         lsoftdir = "%s/software" % tmpdir
 
-        state = self.get_state(user, host, pid, eid)
-
-        if not self.scp_file(tclfile, user, host):
-            return False
-        
-        if state == 'none':
-            # Create a null copy of the experiment so that we capture any
-            # logs there if the modify fails.  Emulab software discards the
-            # logs from a failed startexp
-            try:
-                f = open("%s/null.tcl" % tmpdir, "w")
-                print >>f, self.null
-                f.close()
-            except EnvironmentError, e:
-                raise service_error(service_error.internal,
-                        "Cannot stage tarfile/rpm: %s" % e.strerror)
-
-            if not self.scp_file("%s/null.tcl" % tmpdir, user, host):
-                return False
-            self.log.info("[start_segment]: Creating %s" % eid)
-            timedout = False
-            try:
-                if not self.ssh_cmd(user, host,
-                        ("/usr/testbed/bin/startexp -i -f -w -p %s " + 
-                        "-e %s null.tcl") % (pid, eid), "startexp",
-                        timeout=60 * 10):
-                    return False
-            except self.ssh_cmd_timeout:
-                timedout = True
-
-            if timedout:
-                state = self.get_state(user, host, pid, eid)
-                if state != "swapped":
-                    return False
-        
         # Open up a temporary file to contain a script for setting up the
         # filespace for the new experiment.
@@ -237 +245 @@
                             user, host, "%s/%s" % (softdir, f)):
                         return False
-        # Stage the new configuration (active experiments will stay swapped
-        # in now)
+        return True
+
+    def swap_in(self, user, host, pid, eid):
+        """
+        Swap experiment in.  This includes code to cope with the experiment
+        swaping command timing out, but the experiment being swapped in
+        successfully.
+        """
+        self.log.info("[start_segment]: Swapping %s in" % eid)
+        timedout = False
+        try:
+            if not self.ssh_cmd(user, host,
+                    "/usr/testbed/bin/swapexp -w %s %s in" % (pid, eid),
+                    "swapexp", timeout=25*60):
+                return False
+        except self.ssh_cmd_timeout:
+            timedout = True
+       
+        # If the command was terminated, but completed successfully,
+        # report success.
+        if timedout:
+            self.log.debug("[start_segment]: swapin timed out " +\
+                    "checking state")
+            state = self.get_state(user, host, pid, eid)
+            self.log.debug("[start_segment]: state is %s" % state)
+            return state == 'active'
+
+        return True
+
+
+    def __call__(self, parent, eid, pid, user, tclfile, tmpdir, timeout=0):
+        """
+        Start a sub-experiment on a federant.
+
+        Get the current state, modify or create as appropriate, ship data
+        and configs and start the experiment.  There are small ordering
+        differences based on the initial state of the sub-experiment.
+        """
+        # ops node in the federant
+        host = "%s%s" % (parent.ops, parent.domain)
+        state = self.get_state(user, host, pid, eid)
+
+        if not self.scp_file(tclfile, user, host):
+            return False
+        
+        if state == 'none':
+            # Put a dummy in place to capture logs, and establish an experiment
+            # directory.
+            if not self.make_null_experiment(user, host, pid, eid, tmpdir):
+                return False
+
+        if not self.set_up_experiment_filespace(user, host, pid, eid, tmpdir):
+            return False
+
+        # With the filespace in place, we can modify and swap in.
         self.log.info("[start_segment]: Modifying %s" % eid)
         try:
@@ -253 +314 @@
         # Active experiments are still swapped, this swaps the others in.
         if state != 'active':
-            self.log.info("[start_segment]: Swapping %s" % eid)
-            timedout = False
-            try:
-                if not self.ssh_cmd(user, host,
-                        "/usr/testbed/bin/swapexp -w %s %s in" % (pid, eid),
-                        "swapexp", timeout=25*60):
-                    return False
-            except self.ssh_cmd_timeout:
-                timedout = True
-           
-            # If the command was terminated, but completed successfully,
-            # report success.
-            if timedout:
-                self.log.debug("[start_segment]: swapin timed out " +\
-                        "checking state")
-                state = self.get_state(user, host, pid, eid)
-                self.log.debug("[start_segment]: state is %s" % state)
-                return state == 'active'
+            if not self.swap_in(user, host, pid, eid):
+                return False
         # Everything has gone OK.
         self.get_mapping(user, host, pid,eid)