source: fedd/federation/proxy_protogeni_segment.py @ c2c153b

#!/usr/local/bin/python

import sys, os
import re

import tempfile
import subprocess
import logging 
import time
import random
import string
import signal

import xml.parsers.expat
from threading import Thread

from proxy_segment import proxy_segment
from service_error import service_error
from remote_service import service_caller
from util import fedd_ssl_context

import topdl

class segment_base(proxy_segment):
    class ProtoGENIError(Exception): 
        def __init__(self, op, code, output):
            Exception.__init__(self, output)
            self.op = op
            self.code = code
            self.output = output

    def __init__(self, log=None, keyfile=None, debug=False, 
            ch_url=None, sa_url=None, cm_url=None):
        proxy_segment.__init__(self, log=log, keyfile=keyfile, debug=debug)

        self.ProtoGENIError = start_segment.ProtoGENIError
        self.ch_url = ch_url
        self.sa_url = sa_url
        self.cm_url = cm_url

        self.call_SetValue = service_caller('SetValue')

        self.debug_fail = ['Resolve']
        self.debug_response = {
                'RedeemTicket': ("XML blob1", "XML blob2"),
                'SliceStatus': { 'status': 'ready' },
            }


    def pg_call(self, url, method, params, context):
        max_retries = 5
        retries = 0

        s = service_caller(method, request_body_name="", strict=False)
        self.log.debug("Calling %s %s" % (url, method))
        if not self.debug:
            while retries < max_retries:
                r = s.call_xmlrpc_service(url, params, context=context)
                code = r.get('code', -1)
                if code == 0:
                    # Success leaves the loop here
                    return r.get('value', None)
                elif code == 14 and retries +1 < max_retries:
                    # Busy resource
                    retries+= 1
                    self.log.info("Resource busy, retrying in 30 secs")
                    time.sleep(30)
                else:
                    # NB: out of retries falls through to here
                    raise self.ProtoGENIError(op=method, 
                            code=r.get('code', 'unknown'), 
                            output=r.get('output', 'No output'))
        else:
            if method in self.debug_fail:
                raise self.ProtoGENIError(op=method, code='unknown',
                        output='No output')
            elif self.debug_response.has_key(method):
                return self.debug_response[method]
            else:
                return "%s XML blob" % method


class start_segment(segment_base):
    def __init__(self, log=None, keyfile=None, debug=False, 
            ch_url=None, sa_url=None, cm_url=None):
        segment_base.__init__(self, log=log, keyfile=keyfile, debug=debug,
                ch_url=cm_url, sa_url=sa_url, cm_url=cm_url)


    # Turn the manifest into a dict were each virtual nodename (i.e. the topdl
    # name) has an entry with the allocated machine in hostname and the
    # interfaces in 'interfaces'.  I love having XML parser code lying around.
    def manifest_to_dict(self, manifest, ignore_debug=False):
        if self.debug and not ignore_debug: 
            self.log.debug("Returning null manifest dict")
            return { }

        # The class allows us to keep a little state - the dict under
        # consteruction and the current entry in that dict for the interface
        # element code.
        class manifest_parser:
            def __init__(self):
                self.d = { }
                self.current_key=None

            # If the element is a node, create a dict entry for it.  If it's an
            # interface inside a node, add an entry in the interfaces list with
            # the virtual name and component id.
            def start_element(self, name, attrs):
                if name == 'node':
                    self.current_key = attrs.get('virtual_id',"")
                    if self.current_key:
                        self.d[self.current_key] = {
                                'hostname': attrs.get('hostname', None),
                                'interfaces': { }
                                }
                elif name == 'interface' and self.current_key:
                    self.d[self.current_key]['interfaces']\
                            [attrs.get('virtual_id','')] = \
                            attrs.get('component_id', None)
            #  When a node is finished, clear current_key
            def end_element(self, name):
                if name == 'node': self.current_key = None


        mp = manifest_parser()
        p = xml.parsers.expat.ParserCreate()
        # These are bound to the class we just created
        p.StartElementHandler = mp.start_element
        p.EndElementHandler = mp.end_element

        p.Parse(manifest)
        return mp.d


    def generate_portal_configs(self, parent, topo, pubkey_base, 
            secretkey_base, tmpdir, master, leid, connInfo, services, nodes):

        def conninfo_to_dict(key, info):
            """
            Make a cpoy of the connection information about key, and flatten it
            into a single dict by parsing out any feddAttrs.
            """

            rv = None
            for i in info:
                if key == i.get('portal', "") or \
                        key in [e.get('element', "") \
                        for e in i.get('member', [])]:
                    rv = i.copy()
                    break

            else:
                return rv

            if 'fedAttr' in rv:
                for a in rv['fedAttr']:
                    attr = a.get('attribute', "")
                    val = a.get('value', "")
                    if attr and attr not in rv:
                        rv[attr] = val
                del rv['fedAttr']
            return rv

        # XXX: un hardcode this
        def client_null(f, s):
            print >>f, "Service: %s" % s['name']

        def client_smb(f, s):
            print >>f, "Service: %s" % s['name']
            smbshare = None
            smbuser = None
            smbproj = None
            for a in s.get('fedAttr', []):
                if a.get('attribute', '') == 'SMBSHARE':
                    smbshare = a.get('value', None)
                elif a.get('attribute', '') == 'SMBUSER':
                    smbuser = a.get('value', None)
                elif a.get('attribute', '') == 'SMBPROJ':
                    smbproj = a.get('value', None)

            if all((smbshare, smbuser, smbproj)):
                print >>f, "SMBshare: %s" % smbshare
                print >>f, "ProjectUser: %s" % smbuser
                print >>f, "ProjectName: %s" % smbproj

        client_service_out = {
                'SMB': client_smb,
                'tmcd': client_null,
                'seer': client_null,
                'userconfig': client_null,
            }
        # XXX: end un hardcode this


        seer_out = False
        client_out = False
        for e in [ e for e in topo.elements \
                if isinstance(e, topdl.Computer) and e.get_attribute('portal')]:
            myname = e.name[0]
            type = e.get_attribute('portal_type')

            info = conninfo_to_dict(myname, connInfo)

            if not info:
                raise service_error(service_error.req,
                        "No connectivity info for %s" % myname)

            # Translate to physical name (ProtoGENI doesn't have DNS)
            physname = nodes.get(myname, {'hostname': myname})['hostname']
            peer = info.get('peer', "")
            ldomain = parent.domain;

            mexp = info.get('masterexperiment',"")
            mproj, meid = mexp.split("/", 1)
            mdomain = info.get('masterdomain',"")
            muser = info.get('masteruser','root')
            smbshare = info.get('smbshare', 'USERS')
            ssh_port = info.get('ssh_port', '22')

            active = info.get('active', 'False')

            cfn = "%s/%s.gw.conf" % (tmpdir, myname.lower())
            tunnelconfig = parent.attrs.has_key('TunnelCfg')
            try:
                f = open(cfn, "w")
                if active == 'True':
                    print >>f, "active: True"
                    print >>f, "ssh_port: %s" % ssh_port
                    if type in ('control', 'both'):
                        for s in [s for s in services \
                                if s.get('name', "") in parent.imports]:
                            p = urlparse(s.get('server', 'http://localhost'))
                            print >>f, 'port: remote:%s:%s:%s' % \
                                    (p.port, p.hostname, p.port)

                if tunnelconfig:
                    print >>f, "tunnelip: %s" % tunnelconfig
                # XXX: send this an fedattr
                #print >>f, "seercontrol: control.%s.%s%s" % \
                        #(meid.lower(), mproj.lower(), mdomain)
                print >>f, "peer: %s" % peer.lower()
                print >>f, "ssh_pubkey: /usr/local/federation/etc/%s" % \
                        pubkey_base
                print >>f, "ssh_privkey: /usr/local/federation/etc/%s" % \
                        secretkey_base
                f.close()
            except IOError, e:
                raise service_error(service_error.internal,
                        "Can't write protal config %s: %s" % (cfn, e))
            
            # XXX: This little seer config file needs to go away.
            if not seer_out:
                try:
                    seerfn = "%s/seer.conf" % tmpdir
                    f = open(seerfn, "w")
                    if not master:
                        print >>f, "ControlNode: control.%s.%s%s" % \
                            (meid.lower(), mproj.lower(), mdomain)
                    print >>f, "ExperimentID: %s" % mexp
                    f.close()
                except IOError, e:
                    raise service_error(service_error.internal, 
                            "Can't write seer.conf: %s" %e)
                seer_out = True

            if not client_out and type in ('control', 'both'):
                try:
                    f = open("%s/client.conf" % tmpdir, "w")
                    print >>f, "ControlGateway: %s" % physname.lower()
                    for s in services:
                        if s.get('name',"") in parent.imports and \
                                s.get('visibility','') == 'import':
                            client_service_out[s['name']](f, s)
                    # Does seer need this?
                    # print >>f, "ExperimentID: %s/%s" % (mproj, meid)
                    f.close()
                except IOError, e:
                    raise service_error(service_error.internal,
                            "Cannot write client.conf: %s" %s)
                client_out = True



    def export_store_info(self, cf, nodes, ssh_port, connInfo):
        """
        For the export requests in the connection info, install the peer names
        at the experiment controller via SetValue calls.
        """

        for c in connInfo:
            for p in [ p for p in c.get('parameter', []) \
                    if p.get('type', '') == 'output']:

                if p.get('name', '') == 'peer':
                    k = p.get('key', None)
                    surl = p.get('store', None)
                    if surl and k and k.index('/') != -1:
                        if self.debug:
                            req = { 'name': k, 'value': 'debug' }
                            self.call_SetValue(surl, req, cf)
                        else:
                            value = nodes.get(k[k.index('/')+1:], 
                                    {}).get('hostname',"")
                            if value:
                                req = { 'name': k, 'value': value }
                                self.call_SetValue(surl, req, cf)
                            else:
                                self.log.error("No hostname for %s" % \
                                        k[k.index('/'):])
                    else:
                        self.log.error("Bad export request: %s" % p)
                elif p.get('name', '') == 'ssh_port':
                    k = p.get('key', None)
                    surl = p.get('store', None)
                    if surl and k:
                        req = { 'name': k, 'value': ssh_port }
                        self.call_SetValue(surl, req, cf)
                    else:
                        self.log.error("Bad export request: %s" % p)
                else:

                    self.log.error("Unknown export parameter: %s" % \
                            p.get('name'))
                    continue

    def configure_nodes(self, topo, nodes, user, host, sshd, sshd_config,
            gate_cmd, node_cmd, pubkey, secretkey, stagingdir, tmpdir):

        fed_dir = "/usr/local/federation"
        ssh = "/usr/bin/ssh -n -i .ssh/id_rsa -o 'ForwardX11 no' -o 'StrictHostKeyChecking no' "
        scp = "/usr/bin/scp -i .ssh/id_rsa -o 'ForwardX11 no' -o 'StrictHostKeyChecking no' "
        ifconfig = "/sbin/ifconfig"
        tar = "/bin/tar"

        for e in [ e for e in topo.elements if isinstance(e, topdl.Computer)]:
            vname = e.name[0]
            node = nodes.get(vname, {})
            pname = node.get('hostname', None)
            if pname:
                script = open("%s/%s.startup" %(tmpdir, pname), "w")
                # Reset the interfaces to the ones in the topo file
                for i in [ i for i in e.interface \
                        if not i.get_attribute('portal')]:
                    pinf = node['interfaces'].get(i.name, None)
                    addr = i.get_attribute('ip4_address') 
                    netmask = i.get_attribute('ip4_netmask') or '255.255.255.0'
                    if pinf and addr:
                        print >>script, \
                                "%s %s %s netmask %s"  % \
                                (ifconfig, pinf, addr, netmask)
                    else:
                        self.log.error("Missing interface or address for %s" \
                                % i.name)
                    
                for s in e.software:
                    # XXX: Just tarfiles for now
                    if not (s.location and s.install):
                        continue
                    s_base = s.location.rpartition('/')[2]
                    print >>script, "%s %s@%s:%s/%s ." % \
                            (scp, user, host, stagingdir, s_base)
                    print >>script, \
                            "%s -C %s -xzf %s" % (tar, s.install, s_base)
                for f in ('hosts', pubkey, secretkey, 'client.conf', 
                        'userconf', 'seer.conf'):
                    print >>script, "%s %s@%s:%s/%s %s/etc" % \
                            (scp, user, host, stagingdir, f, fed_dir)
                if sshd:
                    print >>script, "%s %s@%s:%s %s/bin" % \
                            (scp, user, host, sshd, fed_dir)
                if sshd_config:
                    print >>script, "%s %s@%s:%s %s/etc" % \
                            (scp, user, host, sshd_config, fed_dir)
                # Look in tmpdir to get the names.  They've all been copied
                # into the (remote) staging dir
                if os.access("%s/%s.gw.conf" % (tmpdir, vname), os.R_OK):
                    print >>script, "%s %s@%s:%s/%s.gw.conf %s/etc" % \
                            (scp, user, host, stagingdir, vname, fed_dir)


                # Start commands
                if e.get_attribute('portal') and gate_cmd:
                    # Portals never have a user-specified start command
                    print >>script, gate_cmd
                elif node_cmd:
                    # XXX: debug
                    print >>script, "sudo perl -I%s/lib %s/bin/import_key.pl /users/%s/.ssh/authorized_keys /root/.ssh/authorized_keys" % (fed_dir, fed_dir, user)
                    # XXX: debug
                    # start routing on nodes
                    print >>script, "sudo perl %s/bin/protogeni_routing.pl" % \
                            fed_dir
                    if e.get_attribute('startup'):
                        print >>script, "%s \\$USER '%s'" % \
                                (node_cmd, e.get_attribute('startup'))
                    else:
                        print >>script, node_cmd
                script.close()
                if not self.scp_file("%s/%s.startup" % (tmpdir, pname), 
                        user, pname):
                    self.log.error("Could not copy script to %s" % pname)
            else:
                self.log.error("Unmapped node: %s" % vname)

    def start_node(self, user, host, node):
        # Place an identity on the node so that the copying can succeed
        self.ssh_cmd(user, host, "scp .ssh/id_rsa %s:.ssh" % node)
        self.ssh_cmd(user, node, "sudo /bin/sh ./%s.startup &" % node)

    def start_nodes(self, user, host, nodes):
        threads = [ ]
        for n in nodes:
            t = Thread(target=self.start_node, args=(user, host, n))
            t.start()
            threads.append(t)

        done = [not t.isAlive() for t in threads]
        while not all(done):
            self.log.info("Waiting for threads %s" % done)
            time.sleep(10)
            done = [not t.isAlive() for t in threads]




    def __call__(self, parent, aid, user, rspec, pubkey, secretkey, master,
            ename, stagingdir, tmpdir, certfile, certpw, export_certfile, topo,
            connInfo, services, 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.
        """

        def random_slicename(user):
            slicename = user
            for i in range(0,5):
                slicename += random.choice(string.ascii_letters)
            return slicename

        host = parent.staging_host
        ctxt = fedd_ssl_context(my_cert=certfile, password=certpw)
        # Local software dir
        lsoftdir = "%s/software" % tmpdir

        # Open up a temporary file to contain a script for setting up the
        # filespace for the new experiment.
        self.log.info("[start_segment]: creating script file")
        try:
            sf, scriptname = tempfile.mkstemp()
            scriptfile = os.fdopen(sf, 'w')
        except IOError:
            return False

        scriptbase = os.path.basename(scriptname)

        # Script the filesystem changes
        print >>scriptfile, "/bin/rm -rf %s" % stagingdir
        print >>scriptfile, 'mkdir -p %s' % stagingdir
        print >>scriptfile, "rm -f %s" % scriptbase
        scriptfile.close()

        # Move the script to the remote machine
        # XXX: could collide tempfile names on the remote host
        if self.scp_file(scriptname, user, host, scriptbase):
            os.remove(scriptname)
        else:
            return False

        # Execute the script (and the script's last line deletes it)
        if not self.ssh_cmd(user, host, "sh -x %s" % scriptbase):
            return False

        try:
            gcred = self.pg_call(self.sa_url, 'GetCredential', {}, ctxt)
        except self.ProtoGENIError, e:
            raise service_error(service_error.federant,
                    "ProtoGENI: %s" % e)
        # Find a slicename not in use
        slicename = "fabereGpgL"
        while True:
            slicename = random_slicename(user)
            try:
                param = {
                        'credential': gcred, 
                        'hrn': slicename,
                        'type': 'Slice'
                        }
                self.pg_call(self.sa_url, 'Resolve', param, ctxt)
            except self.ProtoGENIError, e:
                print e
                break

        self.log.info("Creating %s" % slicename)
        f = open("./rspec", "w")
        print >>f, "%s" % rspec
        f.close()
        # Create the slice and allocate resources.  If any of this stuff fails,
        # the allocations will time out on PG in short order, so we just raise
        # the service_error.
        try:
            param = {
                    'credential': gcred, 
                    'hrn': slicename,
                    'type': 'Slice'
                    }
            slice_cred = self.pg_call(self.sa_url, 'Register', param, ctxt)
            f = open("./slice_cred", "w")
            print >>f, slice_cred
            f.close()
            # Populate the ssh keys (let PG format them)
            param = {
                    'credential': gcred, 
                    }
            keys =  self.pg_call(self.sa_url, 'GetKeys', param, ctxt)
            # Grab and redeem a ticket
            param = {
                    'credential': slice_cred, 
                    'rspec': rspec,
                    }
            ticket = self.pg_call(self.cm_url, 'GetTicket', param, ctxt)
            f = open("./ticket", "w")
            print >>f, ticket
            f.close()
            param = { 
                    'credential': slice_cred, 
                    'keys': keys,
                    'ticket': ticket,
                    }
            sliver_cred, manifest = self.pg_call(self.cm_url, 
                    'RedeemTicket', param, ctxt)
            f = open("./sliver_cred", "w")
            print >>f, sliver_cred
            f.close()
            f = open("./manifest", "w")
            print >>f, manifest
            f.close()
            # start 'em up
            param = { 
                    'credential': sliver_cred,
                    }
            self.pg_call(self.cm_url, 'StartSliver', param, ctxt)
        except self.ProtoGENIError, e:
            raise service_error(service_error.federant,
                    "ProtoGENI: %s %s" % (e.code, e))

        # With manifest in hand, we can export the portal node names.
        nodes = self.manifest_to_dict(manifest, ignore_debug=True)
        self.export_store_info(export_certfile, nodes, parent.ssh_port,
                connInfo)
        self.generate_portal_configs(parent, topo, pubkey, secretkey, tmpdir, 
                master, ename, connInfo, services, nodes)

        # Copy software to the staging machine (done after generation to copy
        # those, too)
        for d in (tmpdir, lsoftdir):
            if os.path.isdir(d):
                for f in os.listdir(d):
                    if not os.path.isdir("%s/%s" % (d, f)):
                        if not self.scp_file("%s/%s" % (d, f), 
                                user, host, "%s/%s" % (stagingdir, f)):
                            self.log.error("Scp failed")
                            return False


        # Now we wait for the nodes to start on PG
        status = 'notready'
        try:
            while status == 'notready':
                param = { 
                        'credential': slice_cred
                        }
                r = self.pg_call(self.cm_url, 'SliceStatus', param, ctxt)
                print r
                status = r.get('status', 'notready')
                if status == 'notready':
                    time.sleep(30)
        except self.ProtoGENIError, e:
            raise service_error(service_error.federant,
                    "ProtoGENI: %s %s" % (e.code, e))

        if status == 'failed':
            self.log.error('Sliver failed to start on ProtoGENI')
            try:
                param = { 
                        'credential': slice_cred
                        }
                self.pg_call(self.cm_url, 'DeleteSliver', param, ctxt)
            except self.ProtoGENIError, e:
                raise service_error(service_error.federant,
                    "ProtoGENI: %s" % e)
            return False
        else:
            parent.state_lock.acquire()
            parent.allocation[aid]['slice_name'] = slicename
            parent.allocation[aid]['slice_credential'] = slice_cred
            parent.allocation[aid]['sliver_credential'] = sliver_cred
            parent.allocation[aid]['manifest'] = manifest
            parent.allocation[aid]['certfile'] = certfile
            parent.allocation[aid]['certpw'] = certpw
            parent.write_state()
            parent.state_lock.release()

        # The startcmds for portals and standard nodes (the Master Slave
        # distinction is going away)
        gate_cmd = parent.attrs.get('SlaveConnectorStartCmd', '/bin/true')
        node_cmd = parent.attrs.get('SlaveNodeStartCmd', 'bin/true')

        # Now we have configuration to do for ProtoGENI
        self.configure_nodes(topo, nodes, user, parent.staging_host,
                parent.sshd, parent.sshd_config, gate_cmd, node_cmd, 
                pubkey, secretkey, stagingdir, tmpdir)

        self.start_nodes(user, parent.staging_host, 
                [ n['hostname'] for n in nodes.values()])

        # Everything has gone OK.
        return True

class stop_segment(segment_base):
    def __init__(self, log=None, keyfile=None, debug=False, 
            ch_url=None, sa_url=None, cm_url=None):
        segment_base.__init__(self, log=log, keyfile=keyfile, debug=debug,
                ch_url=cm_url, sa_url=sa_url, cm_url=cm_url)

    def __call__(self, parent, user, stagingdir, slice_cred, certfile, certpw):
        """
        Stop a sub experiment by calling swapexp on the federant
        """
        host = parent.staging_host
        rv = False
        try:
            # Clean out tar files: we've gone over quota in the past
            if stagingdir:
                self.ssh_cmd(user, host, "rm -rf %s" % stagingdir)
            if slice_cred:
                self.log.error('Removing Sliver on ProtoGENI')
                ctxt = fedd_ssl_context(my_cert=certfile, password=certpw)
                try:
                    param = { 
                            'credential': slice_cred
                            }
                    self.pg_call(self.cm_url, 'DeleteSlice', param, ctxt)
                except self.ProtoGENIError, e:
                    raise service_error(service_error.federant,
                        "ProtoGENI: %s" % e)
            return True
        except self.ssh_cmd_timeout:
            rv = False
        return rv

class renew_segment(segment_base):
    def __init__(self, log=None, debug=False, keyfile=None,
            ch_url=None, sa_url=None, cm_url=None):
        segment_base.__init__(self, log=log, keyfile=keyfile, debug=debug,
                ch_url=cm_url, sa_url=sa_url, cm_url=cm_url)

    def __call__(self, name, scred, interval, certfile, certpw):
        ctxt = fedd_ssl_context(my_cert=certfile, password=certpw)
        try:
            expiration = time.strftime("%Y%m%dT%H:%M:%S",
                    time.gmtime(time.time() + interval))
            cred = self.pg_call(self.sa_url, 'GetCredential', {}, ctxt)

            param = {
                    'credential': scred,
                    'expiration': expiration
                    }
            r = self.pg_call(self.sa_url, 'RenewSlice', param, ctxt)
            param = {
                    'credential': cred,
                    'hrn': name,
                    'type': 'Slice',
                    }
            slice = self.pg_call(self.sa_url, 'Resolve', param, ctxt)
            uuid = slice.get('uuid', None)
            if uuid == None:
                sys.exit('No uuid for %s' % slicename)

            print 'Calling GetCredential (uuid)'
            param = {
                    'credential': cred,
                    'uuid': uuid,
                    'type': 'Slice',
                    }
            new_scred = self.pg_call(self.sa_url, 'GetCredential', param, ctxt)
            f = open('./new_slice_cred', 'w')
            print >>f, new_scred
            f.close()

        except self.ProtoGENIError, e:
            self.log.error("Failed to extend slice %s: %s" % (name, e))
            return None
        try:
            print 'Calling RenewSlice (CM)'
            param = {
                    'credential': new_scred,
                    }
            r = self.pg_call(self.cm_url, 'RenewSlice', param, ctxt)
        except self.ProtoGENIError, e:
            self.log.warn("Failed to renew sliver for %s: %s" % (name, e))

        return new_scred