X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=system%2FTestPlc.py;h=88a4a4146bbf884173fc4e168969da71f6ca2f2d;hb=6466c59ed2ca5112c7915b1a98eeb4525b4bc993;hp=fb259f5839f912daa927bf84c2f3678158e3329a;hpb=ae3850f83f338b0545a24b73d856743f7f1d09cb;p=tests.git diff --git a/system/TestPlc.py b/system/TestPlc.py index fb259f5..88a4a41 100644 --- a/system/TestPlc.py +++ b/system/TestPlc.py @@ -3,9 +3,9 @@ import os, os.path import datetime import time import sys -import datetime import traceback from types import StringTypes +import socket import utils from TestSite import TestSite @@ -37,17 +37,62 @@ def standby_generic (func): def node_mapper (method): def actual(self): overall=True + node_method = TestNode.__dict__[method.__name__] for site_spec in self.plc_spec['sites']: test_site = TestSite (self,site_spec) for node_spec in site_spec['nodes']: test_node = TestNode (self,test_site,node_spec) - node_method = TestNode.__dict__[method.__name__] if not node_method(test_node): overall=False return overall return actual +def slice_mapper_options (method): + def actual(self): + overall=True + slice_method = TestSlice.__dict__[method.__name__] + for slice_spec in self.plc_spec['slices']: + site_spec = self.locate_site (slice_spec['sitename']) + test_site = TestSite(self,site_spec) + test_slice=TestSlice(self,test_site,slice_spec) + if not slice_method(test_slice,self.options): overall=False + return overall + return actual + +SEP='' + class TestPlc: + default_steps = [ + 'display','uninstall','install','install_rpm', + 'configure', 'start', 'fetch_keys', SEP, + 'store_keys', 'clear_known_hosts', 'initscripts', SEP, + 'sites', 'nodes', 'slices', 'nodegroups', SEP, + 'init_node','bootcd', 'configure_qemu', 'export_qemu', + 'kill_all_qemus', 'reinstall_node','start_node', SEP, + # better use of time: do this now that the nodes are taking off + 'plcsh_stress_test', SEP, + 'nodes_ssh_debug', 'nodes_ssh_boot', 'check_slice', 'check_initscripts', SEP, + 'check_tcp', SEP, + 'force_gather_logs', 'force_kill_qemus', 'force_record_tracker','force_free_tracker', + ] + other_steps = [ + 'stop_all_vservers','fresh_install', 'cache_rpm', 'stop', 'vs_start', SEP, + 'check_sanity', SEP, + 'clean_initscripts', 'clean_nodegroups','clean_all_sites', SEP, + 'clean_sites', 'clean_nodes', + 'clean_slices', 'clean_keys', SEP, + 'show_boxes', 'list_all_qemus', 'list_qemus', SEP, + 'db_dump' , 'db_restore', 'cleanup_trackers', 'cleanup_all_trackers', + 'standby_1 through 20', + ] + + @staticmethod + def printable_steps (list): + return " ".join(list).replace(" "+SEP+" "," \\\n") + @staticmethod + def valid_step (step): + return step != SEP + def __init__ (self,plc_spec,options): self.plc_spec=plc_spec self.options=options @@ -58,17 +103,12 @@ class TestPlc: self.url="https://%s:443/PLCAPI/"%plc_spec['vserverip'] self.vserver=True except: - self.vserver=False - self.url="https://%s:443/PLCAPI/"%plc_spec['hostname'] -# utils.header('Using API url %s'%self.url) + raise Exception,'chroot-based myplc testing is deprecated' self.apiserver=TestApiserver(self.url,options.dry_run) def name(self): name=self.plc_spec['name'] - if self.vserver: - return name+".vserver.%s"%self.vservername - else: - return name+".chroot" + return "%s.%s"%(name,self.vservername) def hostname(self): return self.plc_spec['hostname'] @@ -84,40 +124,26 @@ class TestPlc: def actual_command_in_guest (self,command): return self.test_ssh.actual_command(self.host_to_guest(command)) + def start_guest (self): + return utils.system(self.test_ssh.actual_command(self.start_guest_in_host())) + def run_in_guest (self,command): return utils.system(self.actual_command_in_guest(command)) def run_in_host (self,command): return self.test_ssh.run_in_buildname(command) - #command gets run in the chroot/vserver + #command gets run in the vserver def host_to_guest(self,command): - if self.vserver: - return "vserver %s exec %s"%(self.vservername,command) - else: - return "chroot /plc/root %s"%TestSsh.backslash_shell_specials(command) + return "vserver %s exec %s"%(self.vservername,command) + + #command gets run in the vserver + def start_guest_in_host(self): + return "vserver %s start"%(self.vservername) - # copy a file to the myplc root image - pass in_data=True if the file must go in /plc/data - def copy_in_guest (self, localfile, remotefile, in_data=False): - if in_data: - chroot_dest="/plc/data" - else: - chroot_dest="/plc/root" - if self.is_local(): - if not self.vserver: - utils.system("cp %s %s/%s"%(localfile,chroot_dest,remotefile)) - else: - utils.system("cp %s /vservers/%s/%s"%(localfile,self.vservername,remotefile)) - else: - if not self.vserver: - utils.system("scp %s %s:%s/%s"%(localfile,self.hostname(),chroot_dest,remotefile)) - else: - utils.system("scp %s %s@/vservers/%s/%s"%(localfile,self.hostname(),self.vservername,remotefile)) - - # xxx quick n dirty def run_in_guest_piped (self,local,remote): - return utils.system(local+" | "+self.test_ssh.actual_command(self.host_to_guest(remote))) + return utils.system(local+" | "+self.test_ssh.actual_command(self.host_to_guest(remote),keep_stdin=True)) def auth_root (self): return {'Username':self.plc_spec['PLC_ROOT_USER'], @@ -159,6 +185,37 @@ class TestPlc: return slice raise Exception,"Cannot locate slice %s"%slicename + def all_sliver_objs (self): + result=[] + for slice_spec in self.plc_spec['slices']: + slicename = slice_spec['slice_fields']['name'] + for nodename in slice_spec['nodenames']: + result.append(self.locate_sliver_obj (nodename,slicename)) + return result + + def locate_sliver_obj (self,nodename,slicename): + (site,node) = self.locate_node(nodename) + slice = self.locate_slice (slicename) + # build objects + test_site = TestSite (self, site) + test_node = TestNode (self, test_site,node) + # xxx the slice site is assumed to be the node site - mhh - probably harmless + test_slice = TestSlice (self, test_site, slice) + return TestSliver (self, test_node, test_slice) + + def locate_first_node(self): + nodename=self.plc_spec['slices'][0]['nodenames'][0] + (site,node) = self.locate_node(nodename) + test_site = TestSite (self, site) + test_node = TestNode (self, test_site,node) + return test_node + + def locate_first_sliver (self): + slice_spec=self.plc_spec['slices'][0] + slicename=slice_spec['slice_fields']['name'] + nodename=slice_spec['nodenames'][0] + return self.locate_sliver_obj(nodename,slicename) + # all different hostboxes used in this plc def gather_hostBoxes(self): # maps on sites and nodes, return [ (host_box,test_node) ] @@ -169,7 +226,7 @@ class TestPlc: test_node = TestNode (self, test_site, node_spec) if not test_node.is_real(): tuples.append( (test_node.host_box(),test_node) ) - # transform into a dict { 'host_box' -> [ hostnames .. ] } + # transform into a dict { 'host_box' -> [ test_node .. ] } result = {} for (box,node) in tuples: if not result.has_key(box): @@ -186,9 +243,11 @@ class TestPlc: # make this a valid step def kill_all_qemus(self): + # this is the brute force version, kill all qemus on that host box for (box,nodes) in self.gather_hostBoxes().iteritems(): - # this is the brute force version, kill all qemus on that host box - TestBox(box,self.options.buildname).kill_all_qemus() + # pass the first nodename, as we don't push template-qemu on testboxes + nodedir=nodes[0].nodedir() + TestBox(box,self.options.buildname).kill_all_qemus(nodedir) return True # make this a valid step @@ -214,89 +273,242 @@ class TestPlc: node.kill_qemu() return True - #################### step methods - - ### uninstall - def uninstall_chroot(self): - self.run_in_host('service plc safestop') - #####detecting the last myplc version installed and remove it - self.run_in_host('rpm -e myplc') - ##### Clean up the /plc directory - self.run_in_host('rm -rf /plc/data') - ##### stop any running vservers - self.run_in_host('for vserver in $(ls -d /vservers/* | sed -e s,/vservers/,,) ; do case $vserver in vtest*) echo Shutting down vserver $vserver ; vserver $vserver stop ;; esac ; done') + #################### display config + def display (self): + self.display_pass (1) + self.display_pass (2) return True - def uninstall_vserver(self): - self.run_in_host("vserver --silent %s delete"%self.vservername) + # entry point + def display_pass (self,passno): + for (key,val) in self.plc_spec.iteritems(): + if passno == 2: + if key == 'sites': + for site in val: + self.display_site_spec(site) + for node in site['nodes']: + self.display_node_spec(node) + elif key=='initscripts': + for initscript in val: + self.display_initscript_spec (initscript) + elif key=='slices': + for slice in val: + self.display_slice_spec (slice) + elif key=='keys': + for key in val: + self.display_key_spec (key) + elif passno == 1: + if key not in ['sites','initscripts','slices','keys']: + print '* ',key,':',val + + def display_site_spec (self,site): + print '* ======== site',site['site_fields']['name'] + for (k,v) in site.iteritems(): + if k=='nodes': + if v: + print '* ','nodes : ', + for node in v: + print node['node_fields']['hostname'],'', + print '' + elif k=='users': + if v: + print '* users : ', + for user in v: + print user['name'],'', + print '' + elif k == 'site_fields': + print '* login_base',':',v['login_base'] + elif k == 'address_fields': + pass + else: + print '* ',k, + PrettyPrinter(indent=8,depth=2).pprint(v) + + def display_initscript_spec (self,initscript): + print '* ======== initscript',initscript['initscript_fields']['name'] + + def display_key_spec (self,key): + print '* ======== key',key['name'] + + def display_slice_spec (self,slice): + print '* ======== slice',slice['slice_fields']['name'] + for (k,v) in slice.iteritems(): + if k=='nodenames': + if v: + print '* nodes : ', + for nodename in v: + print nodename,'', + print '' + elif k=='usernames': + if v: + print '* users : ', + for username in v: + print username,'', + print '' + elif k=='slice_fields': + print '* fields',':', + print 'max_nodes=',v['max_nodes'], + print '' + else: + print '* ',k,v + + def display_node_spec (self,node): + print "* node",node['name'],"host_box=",node['host_box'], + print "hostname=",node['node_fields']['hostname'], + print "ip=",node['interface_fields']['ip'] + + + # another entry point for just showing the boxes involved + def display_mapping (self): + TestPlc.display_mapping_plc(self.plc_spec) return True - def uninstall(self): - # if there's a chroot-based myplc running, and then a native-based myplc is being deployed - # it sounds safer to have the former uninstalled too - # now the vserver method cannot be invoked for chroot instances as vservername is required - if self.vserver: - self.uninstall_vserver() - self.uninstall_chroot() - else: - self.uninstall_chroot() + @staticmethod + def display_mapping_plc (plc_spec): + print '* MyPLC',plc_spec['name'] + print '*\tvserver address = root@%s:/vservers/%s'%(plc_spec['hostname'],plc_spec['vservername']) + print '*\tIP = %s/%s'%(plc_spec['PLC_API_HOST'],plc_spec['vserverip']) + for site_spec in plc_spec['sites']: + for node_spec in site_spec['nodes']: + TestPlc.display_mapping_node(node_spec) + + @staticmethod + def display_mapping_node (node_spec): + print '* NODE %s'%(node_spec['name']) + print '*\tqemu box %s'%node_spec['host_box'] + print '*\thostname=%s'%node_spec['node_fields']['hostname'] + + ### utility methods for handling the pool of IP addresses allocated to plcs + # Logic + # (*) running plcs are recorded in the file named ~/running-test-plcs + # (*) this file contains a line for each running plc, older first + # (*) each line contains the vserver name + the hostname of the (vserver) testbox where it sits + # (*) the free_tracker method performs a vserver stop on the oldest entry + # (*) the record_tracker method adds an entry at the bottom of the file + # (*) the cleanup_tracker method stops all known vservers and removes the tracker file + + TRACKER_FILE=os.environ['HOME']+"/running-test-plcs" + # how many concurrent plcs are we keeping alive - adjust with the IP pool size + TRACKER_KEEP_VSERVERS = 12 + + def record_tracker (self): + try: + lines=file(TestPlc.TRACKER_FILE).readlines() + except: + lines=[] + + this_line="%s %s\n"%(self.vservername,self.test_ssh.hostname) + for line in lines: + if line==this_line: + print 'this vserver is already included in %s'%TestPlc.TRACKER_FILE + return True + if self.options.dry_run: + print 'dry_run: record_tracker - skipping tracker update' + return True + tracker=file(TestPlc.TRACKER_FILE,"w") + for line in lines+[this_line]: + tracker.write(line) + tracker.close() + print "Recorded %s in running plcs on host %s"%(self.vservername,self.test_ssh.hostname) return True - ### install - def install_chroot(self): - # nothing to do + def free_tracker (self, keep_vservers=None): + if not keep_vservers: keep_vservers=TestPlc.TRACKER_KEEP_VSERVERS + try: + lines=file(TestPlc.TRACKER_FILE).readlines() + except: + print 'dry_run: free_tracker - skipping tracker update' + return True + how_many = len(lines) - keep_vservers + # nothing todo until we have more than keep_vservers in the tracker + if how_many <= 0: + print 'free_tracker : limit %d not reached'%keep_vservers + return True + to_stop = lines[:how_many] + to_keep = lines[how_many:] + for line in to_stop: + print '>%s<'%line + [vname,hostname]=line.split() + command=TestSsh(hostname).actual_command("vserver --silent %s stop"%vname) + utils.system(command) + if self.options.dry_run: + print 'dry_run: free_tracker would stop %d vservers'%len(to_stop) + for line in to_stop: print line, + print 'dry_run: free_tracker would keep %d vservers'%len(to_keep) + for line in to_keep: print line, + return True + print "Storing %d remaining vservers in %s"%(len(to_keep),TestPlc.TRACKER_FILE) + tracker=open(TestPlc.TRACKER_FILE,"w") + for line in to_keep: + tracker.write(line) + tracker.close() return True - def install_vserver(self): - # we need build dir for vtest-init-vserver + # this should/could stop only the ones in TRACKER_FILE if that turns out to be reliable + def cleanup_trackers (self): + try: + for line in TestPlc.TRACKER_FILE.readlines(): + [vname,hostname]=line.split() + stop="vserver --silent %s stop"%vname + command=TestSsh(hostname).actual_command(stop) + utils.system(command) + clean_tracker = "rm -f %s"%TestPlc.TRACKER_FILE + utils.system(self.test_ssh.actual_command(clean_tracker)) + except: + return True + + # this should/could stop only the ones in TRACKER_FILE if that turns out to be reliable + def cleanup_all_trackers (self): + stop_all = "cd /vservers ; for i in * ; do vserver --silent $i stop ; done" + utils.system(self.test_ssh.actual_command(stop_all)) + clean_tracker = "rm -f %s"%TestPlc.TRACKER_FILE + utils.system(self.test_ssh.actual_command(clean_tracker)) + return True + + def uninstall(self): + self.run_in_host("vserver --silent %s delete"%self.vservername) + return True + + ### install + def install(self): if self.is_local(): # a full path for the local calls - build_dir=os.path(sys.argv[0])+"/build" + build_dir=os.path.dirname(sys.argv[0]) + # sometimes this is empty - set to "." in such a case + if not build_dir: build_dir="." + build_dir += "/build" else: - # use a standard name - will be relative to HOME - build_dir="options.buildname" + # use a standard name - will be relative to remote buildname + build_dir="build" # run checkout in any case - would do an update if already exists build_checkout = "svn checkout %s %s"%(self.options.build_url,build_dir) if self.run_in_host(build_checkout) != 0: return False - # the repo url is taken from myplc-url - # with the last two steps (i386/myplc...) removed - repo_url = self.options.myplc_url - for level in [ 'rpmname','arch' ]: + # the repo url is taken from arch-rpms-url + # with the last step (i386.) removed + repo_url = self.options.arch_rpms_url + for level in [ 'arch' ]: repo_url = os.path.dirname(repo_url) - create_vserver="%s/vtest-init-vserver.sh %s %s -- --interface eth0:%s"%\ - (build_dir,self.vservername,repo_url,self.vserverip) + # pass the vbuild-nightly options to vtest-init-vserver + test_env_options="" + test_env_options += " -p %s"%self.options.personality + test_env_options += " -d %s"%self.options.pldistro + test_env_options += " -f %s"%self.options.fcdistro + script="vtest-init-vserver.sh" + vserver_name = self.vservername + vserver_options="--netdev eth0 --interface %s"%self.vserverip + try: + vserver_hostname=socket.gethostbyaddr(self.vserverip)[0] + vserver_options += " --hostname %s"%vserver_hostname + except: + pass + create_vserver="%(build_dir)s/%(script)s %(test_env_options)s %(vserver_name)s %(repo_url)s -- %(vserver_options)s"%locals() return self.run_in_host(create_vserver) == 0 - def install(self): - if self.vserver: - return self.install_vserver() - else: - return self.install_chroot() - - ### install_rpm - make this an optional step - def cache_rpm(self): - url = self.options.myplc_url - rpm = os.path.basename(url) - cache_fetch="pwd;if [ -f %(rpm)s ] ; then echo Using cached rpm %(rpm)s ; else echo Fetching %(url)s ; curl -O %(url)s; fi"%locals() - return self.run_in_host(cache_fetch)==0 - - def install_rpm_chroot(self): - url = self.options.myplc_url - rpm = os.path.basename(url) - if not self.cache_rpm(): - return False - utils.header('Installing the : %s'%rpm) - return self.run_in_host('rpm -Uvh '+rpm)==0 and self.run_in_host('service plc mount')==0 - - def install_rpm_vserver(self): - return self.run_in_guest("yum -y install myplc-native")==0 - + ### install_rpm def install_rpm(self): - if self.vserver: - return self.install_rpm_vserver() - else: - return self.install_rpm_chroot() + return self.run_in_guest("yum -y install myplc-native")==0 \ + and self.run_in_guest("yum -y install noderepo-%s-%s"%(self.options.pldistro,self.options.arch))==0 ### def configure(self): @@ -322,22 +534,19 @@ class TestPlc: utils.system('rm %s'%tmpname) return True - # the chroot install is slightly different to this respect def start(self): - if self.vserver: - self.run_in_guest('service plc start') - else: - self.run_in_host('service plc start') + self.run_in_guest('service plc start') return True - + def stop(self): - if self.vserver: - self.run_in_guest('service plc stop') - else: - self.run_in_host('service plc stop') + self.run_in_guest('service plc stop') return True - # could use a TestKey class + def vs_start (self): + self.start_guest() + return True + + # stores the keys from the config for further use def store_keys(self): for key_spec in self.plc_spec['keys']: TestKey(self,key_spec).store_key() @@ -346,6 +555,26 @@ class TestPlc: def clean_keys(self): utils.system("rm -rf %s/keys/"%os.path(sys.argv[0])) + # fetches the ssh keys in the plc's /etc/planetlab and stores them in keys/ + # for later direct access to the nodes + def fetch_keys(self): + dir="./keys" + if not os.path.isdir(dir): + os.mkdir(dir) + vservername=self.vservername + overall=True + prefix = 'root_ssh_key' + for ext in [ 'pub', 'rsa' ] : + src="/vservers/%(vservername)s/etc/planetlab/%(prefix)s.%(ext)s"%locals() + dst="keys/%(vservername)s.%(ext)s"%locals() + if self.test_ssh.fetch(src,dst) != 0: overall=False + prefix = 'debug_ssh_key' + for ext in [ 'pub', 'rsa' ] : + src="/vservers/%(vservername)s/etc/planetlab/%(prefix)s.%(ext)s"%locals() + dst="keys/%(vservername)s-debug.%(ext)s"%locals() + if self.test_ssh.fetch(src,dst) != 0: overall=False + return overall + def sites (self): return self.do_sites() @@ -367,6 +596,13 @@ class TestPlc: test_site.create_users() return True + def clean_all_sites (self): + print 'auth_root',self.auth_root() + site_ids = [s['site_id'] for s in self.apiserver.GetSites(self.auth_root(), {}, ['site_id'])] + for site_id in site_ids: + print 'Deleting site_id',site_id + self.apiserver.DeleteSite(self.auth_root(),site_id) + def nodes (self): return self.do_nodes() def clean_nodes (self): @@ -389,9 +625,13 @@ class TestPlc: test_node.create_node () return True - # create nodegroups if needed, and populate - # no need for a clean_nodegroups if we are careful enough def nodegroups (self): + return self.do_nodegroups("add") + def clean_nodegroups (self): + return self.do_nodegroups("delete") + + # create nodegroups if needed, and populate + def do_nodegroups (self, action="add"): # 1st pass to scan contents groups_dict = {} for site_spec in self.plc_spec['sites']: @@ -407,14 +647,54 @@ class TestPlc: groups_dict[nodegroupname]=[] groups_dict[nodegroupname].append(test_node.name()) auth=self.auth_root() + overall = True for (nodegroupname,group_nodes) in groups_dict.iteritems(): - try: - self.apiserver.GetNodeGroups(auth,{'name':nodegroupname})[0] - except: - self.apiserver.AddNodeGroup(auth,{'name':nodegroupname}) - for node in group_nodes: - self.apiserver.AddNodeToNodeGroup(auth,node,nodegroupname) - return True + if action == "add": + print 'nodegroups:','dealing with nodegroup',nodegroupname,'on nodes',group_nodes + # first, check if the nodetagtype is here + tag_types = self.apiserver.GetTagTypes(auth,{'tagname':nodegroupname}) + if tag_types: + tag_type_id = tag_types[0]['tag_type_id'] + else: + tag_type_id = self.apiserver.AddTagType(auth, + {'tagname':nodegroupname, + 'description': 'for nodegroup %s'%nodegroupname, + 'category':'test', + 'min_role_id':10}) + print 'located tag (type)',nodegroupname,'as',tag_type_id + # create nodegroup + nodegroups = self.apiserver.GetNodeGroups (auth, {'groupname':nodegroupname}) + if not nodegroups: + self.apiserver.AddNodeGroup(auth, nodegroupname, tag_type_id, 'yes') + print 'created nodegroup',nodegroupname,'from tagname',nodegroupname,'and value','yes' + # set node tag on all nodes, value='yes' + for nodename in group_nodes: + try: + self.apiserver.AddNodeTag(auth, nodename, nodegroupname, "yes") + except: + traceback.print_exc() + print 'node',nodename,'seems to already have tag',nodegroupname + # check anyway + try: + expect_yes = self.apiserver.GetNodeTags(auth, + {'hostname':nodename, + 'tagname':nodegroupname}, + ['tagvalue'])[0]['tagvalue'] + if expect_yes != "yes": + print 'Mismatch node tag on node',nodename,'got',expect_yes + overall=False + except: + if not self.options.dry_run: + print 'Cannot find tag',nodegroupname,'on node',nodename + overall = False + else: + try: + print 'cleaning nodegroup',nodegroupname + self.apiserver.DeleteNodeGroup(auth,nodegroupname) + except: + traceback.print_exc() + overall=False + return overall def all_hostnames (self) : hostnames = [] @@ -423,14 +703,14 @@ class TestPlc: for node_spec in site_spec['nodes'] ] return hostnames - # gracetime : during the first minutes nothing gets printed - def do_nodes_booted (self, minutes, gracetime=2): + # silent_minutes : during the first minutes nothing gets printed + def nodes_check_boot_state (self, target_boot_state, timeout_minutes, silent_minutes,period=15): if self.options.dry_run: print 'dry_run' return True # compute timeout - timeout = datetime.datetime.now()+datetime.timedelta(minutes=minutes) - graceout = datetime.datetime.now()+datetime.timedelta(minutes=gracetime) + timeout = datetime.datetime.now()+datetime.timedelta(minutes=timeout_minutes) + graceout = datetime.datetime.now()+datetime.timedelta(minutes=silent_minutes) # the nodes that haven't checked yet - start with a full list and shrink over time tocheck = self.all_hostnames() utils.header("checking nodes %r"%tocheck) @@ -443,21 +723,21 @@ class TestPlc: for array in tocheck_status: hostname=array['hostname'] boot_state=array['boot_state'] - if boot_state == 'boot': - utils.header ("%s has reached the 'boot' state"%hostname) + if boot_state == target_boot_state: + utils.header ("%s has reached the %s state"%(hostname,target_boot_state)) else: # if it's a real node, never mind (site_spec,node_spec)=self.locate_hostname(hostname) if TestNode.is_real_model(node_spec['node_fields']['model']): utils.header("WARNING - Real node %s in %s - ignored"%(hostname,boot_state)) # let's cheat - boot_state = 'boot' - if datetime.datetime.now() > graceout: + boot_state = target_boot_state + elif datetime.datetime.now() > graceout: utils.header ("%s still in '%s' state"%(hostname,boot_state)) graceout=datetime.datetime.now()+datetime.timedelta(1) status[hostname] = boot_state # refresh tocheck - tocheck = [ hostname for (hostname,boot_state) in status.iteritems() if boot_state != 'boot' ] + tocheck = [ hostname for (hostname,boot_state) in status.iteritems() if boot_state != target_boot_state ] if not tocheck: return True if datetime.datetime.now() > timeout: @@ -465,27 +745,47 @@ class TestPlc: utils.header("FAILURE due to %s in '%s' state"%(hostname,status[hostname])) return False # otherwise, sleep for a while - time.sleep(15) + time.sleep(period) # only useful in empty plcs return True def nodes_booted(self): - return self.do_nodes_booted(minutes=0) - + return self.nodes_check_boot_state('boot',timeout_minutes=20,silent_minutes=15) - def do_nodes_ssh(self,minutes): + def check_nodes_ssh(self,debug,timeout_minutes,silent_minutes,period=20): # compute timeout - timeout = datetime.datetime.now()+datetime.timedelta(minutes=minutes) + timeout = datetime.datetime.now()+datetime.timedelta(minutes=timeout_minutes) + graceout = datetime.datetime.now()+datetime.timedelta(minutes=silent_minutes) + vservername=self.vservername + if debug: + message="debug" + local_key = "keys/%(vservername)s-debug.rsa"%locals() + else: + message="boot" + local_key = "keys/%(vservername)s.rsa"%locals() tocheck = self.all_hostnames() -# self.scan_publicKeys(tocheck) - utils.header("checking Connectivity on nodes %r"%tocheck) + utils.header("checking ssh access (expected in %s mode) to nodes %r"%(message,tocheck)) + utils.header("max timeout is %d minutes, silent for %d minutes (period is %s)"%\ + (timeout_minutes,silent_minutes,period)) while tocheck: for hostname in tocheck: - # try to ssh in nodes - node_test_ssh = TestSsh (hostname,key="/etc/planetlab/root_ssh_key.rsa") - access=self.run_in_guest(node_test_ssh.actual_command("date")) - if not access: - utils.header('The node %s is sshable -->'%hostname) + # try to run 'hostname' in the node + command = TestSsh (hostname,key=local_key).actual_command("hostname;uname -a") + # don't spam logs - show the command only after the grace period + if datetime.datetime.now() > graceout: + success=utils.system(command) + else: + # truly silent, just print out a dot to show we're alive + print '.', + sys.stdout.flush() + command += " 2>/dev/null" + if self.options.dry_run: + print 'dry_run',command + success=0 + else: + success=os.system(command) + if success==0: + utils.header('Successfully entered root@%s (%s)'%(hostname,message)) # refresh tocheck tocheck.remove(hostname) else: @@ -501,12 +801,15 @@ class TestPlc: utils.header("FAILURE to ssh into %s"%hostname) return False # otherwise, sleep for a while - time.sleep(15) + time.sleep(period) # only useful in empty plcs return True - def nodes_ssh(self): - return self.do_nodes_ssh(minutes=2) + def nodes_ssh_debug(self): + return self.check_nodes_ssh(debug=True,timeout_minutes=30,silent_minutes=10) + + def nodes_ssh_boot(self): + return self.check_nodes_ssh(debug=False,timeout_minutes=30,silent_minutes=10) @node_mapper def init_node (self): pass @@ -514,7 +817,21 @@ class TestPlc: def bootcd (self): pass @node_mapper def configure_qemu (self): pass + @node_mapper + def reinstall_node (self): pass + @node_mapper + def export_qemu (self): pass + ### check sanity : invoke scripts from qaapi/qa/tests/{node,slice} + def check_sanity_node (self): + return self.locate_first_node().check_sanity() + def check_sanity_sliver (self) : + return self.locate_first_sliver().check_sanity() + + def check_sanity (self): + return self.check_sanity_node() and self.check_sanity_sliver() + + ### initscripts def do_check_initscripts(self): overall = True for slice_spec in self.plc_spec['slices']: @@ -541,6 +858,18 @@ class TestPlc: self.apiserver.AddInitScript(self.auth_root(),initscript['initscript_fields']) return True + def clean_initscripts (self): + for initscript in self.plc_spec['initscripts']: + initscript_name = initscript['initscript_fields']['name'] + print('Attempting to delete %s in plc %s'%(initscript_name,self.plc_spec['name'])) + try: + self.apiserver.DeleteInitScript(self.auth_root(),initscript_name) + print initscript_name,'deleted' + except: + print 'deletion went wrong - probably did not exist' + return True + + ### manage slices def slices (self): return self.do_slices() @@ -561,38 +890,14 @@ class TestPlc: utils.header('Created Slice %s'%slice['slice_fields']['name']) return True - def check_slices(self): - for slice_spec in self.plc_spec['slices']: - site_spec = self.locate_site (slice_spec['sitename']) - test_site = TestSite(self,site_spec) - test_slice=TestSlice(self,test_site,slice_spec) - status=test_slice.do_check_slice(self.options) - if (not status): - return False - return status - - def start_nodes (self): - utils.header("Starting nodes") - for site_spec in self.plc_spec['sites']: - TestSite(self,site_spec).start_nodes (self.options) - return True - - - def locate_first_sliver (self): - slice_spec = self.plc_spec['slices'][0] - slicename = slice_spec['slice_fields']['name'] - nodename = slice_spec['nodenames'][0] - return self.locate_sliver_obj(nodename,slicename) + @slice_mapper_options + def check_slice(self): pass - def locate_sliver_obj (self,nodename,slicename): - (site,node) = self.locate_node(nodename) - slice = self.locate_slice (slicename) - # build objects - test_site = TestSite (self, site) - test_node = TestNode (self, test_site,node) - # xxx the slice site is assumed to be the node site - mhh - probably harmless - test_slice = TestSlice (self, test_site, slice) - return TestSliver (self, test_node, test_slice) + @node_mapper + def clear_known_hosts (self): pass + + @node_mapper + def start_node (self) : pass def check_tcp (self): specs = self.plc_spec['tcp_test'] @@ -610,16 +915,30 @@ class TestPlc: if not c_test_sliver.run_tcp_client(s_test_sliver.test_node.name(),port): overall=False return overall - + + def plcsh_stress_test (self): + # install the stress-test in the plc image + location = "/usr/share/plc_api/plcsh-stress-test.py" + remote="/vservers/%s/%s"%(self.vservername,location) + self.test_ssh.copy_abs("plcsh-stress-test.py",remote) + command = location + command += " -- --check" + if self.options.size == 1: + command += " --tiny" + return ( self.run_in_guest(command) == 0) def gather_logs (self): - # (1) get the plc's /var/log and store it locally in logs/-var-log/* - # (2) get all the nodes qemu log and store it as logs/-qemu.log - # (3) get the nodes /var/log and store is as logs/-var-log/* - # (4) as far as possible get the slice's /var/log as logs/--var-log/* - # (1) + # (1.a) get the plc's /var/log/ and store it locally in logs/myplc.var-log./* + # (1.b) get the plc's /var/lib/pgsql/data/pg_log/ -> logs/myplc.pgsql-log./* + # (2) get all the nodes qemu log and store it as logs/node.qemu..log + # (3) get the nodes /var/log and store is as logs/node.var-log./* + # (4) as far as possible get the slice's /var/log as logs/sliver.var-log./* + # (1.a) print "-------------------- TestPlc.gather_logs : PLC's /var/log" self.gather_var_logs () + # (1.b) + print "-------------------- TestPlc.gather_logs : PLC's /var/lib/psql/data/pg_log/" + self.gather_pgsql_logs () # (2) print "-------------------- TestPlc.gather_logs : nodes's QEMU logs" for site_spec in self.plc_spec['sites']: @@ -632,24 +951,29 @@ class TestPlc: self.gather_nodes_var_logs() # (4) print "-------------------- TestPlc.gather_logs : sample sliver's /var/log" - self.gather_first_sliver_logs() + self.gather_slivers_var_logs() return True - def gather_first_sliver_logs(self): - try: - test_sliver = self.locate_first_sliver() + def gather_slivers_var_logs(self): + for test_sliver in self.all_sliver_objs(): remote = test_sliver.tar_var_logs() - utils.system("mkdir -p logs/%s-var-log"%test_sliver.name()) - command = remote + " | tar -C logs/%s-var-log -xf -"%test_sliver.name() + utils.system("mkdir -p logs/sliver.var-log.%s"%test_sliver.name()) + command = remote + " | tar -C logs/sliver.var-log.%s -xf -"%test_sliver.name() utils.system(command) - except Exception,e: - print 'Cannot locate first sliver - giving up',e return True def gather_var_logs (self): + utils.system("mkdir -p logs/myplc.var-log.%s"%self.name()) to_plc = self.actual_command_in_guest("tar -C /var/log/ -cf - .") - command = to_plc + "| tar -C logs/%s-var-log -xf -"%self.name() - utils.system("mkdir -p logs/%s-var-log"%self.name()) + command = to_plc + "| tar -C logs/myplc.var-log.%s -xf -"%self.name() + utils.system(command) + command = "chmod a+r,a+x logs/myplc.var-log.%s/httpd"%self.name() + utils.system(command) + + def gather_pgsql_logs (self): + utils.system("mkdir -p logs/myplc.pgsql-log.%s"%self.name()) + to_plc = self.actual_command_in_guest("tar -C /var/lib/pgsql/data/pg_log/ -cf - .") + command = to_plc + "| tar -C logs/myplc.pgsql-log.%s -xf -"%self.name() utils.system(command) def gather_nodes_var_logs (self): @@ -659,8 +983,8 @@ class TestPlc: test_node=TestNode(self,test_site,node_spec) test_ssh = TestSsh (test_node.name(),key="/etc/planetlab/root_ssh_key.rsa") to_plc = self.actual_command_in_guest ( test_ssh.actual_command("tar -C /var/log -cf - .")) - command = to_plc + "| tar -C logs/%s-var-log -xf -"%test_node.name() - utils.system("mkdir -p logs/%s-var-log"%test_node.name()) + command = to_plc + "| tar -C logs/node.var-log.%s -xf -"%test_node.name() + utils.system("mkdir -p logs/node.var-log.%s"%test_node.name()) utils.system(command)