def partition_stable(self, nodes, timeout=None): watchpats = [ ] watchpats.append("Current ping state: S_IDLE") watchpats.append(self.templates["Pat:DC_IDLE"]) self.debug("Waiting for cluster stability...") if timeout == None: timeout = self.Env["DeadTime"] if len(nodes) < 3: self.debug("Cluster is inactive") return 1 idle_watch = LogWatcher(self.Env["LogFileName"], watchpats, "ClusterStable", timeout, hosts=nodes.split(), kind=self.Env["LogWatcher"]) idle_watch.setwatch() for node in nodes.split(): # have each node dump its current state self.rsh(node, self.templates["StatusCmd"] % node, 1) ret = idle_watch.look() while ret: self.debug(ret) for node in nodes.split(): if re.search(node, ret): return 1 ret = idle_watch.look() self.debug("Warn: Partition %s not IDLE after %ds" % (repr(nodes), timeout)) return None
def prepare_fencing_watcher(self, name): # If we don't have quorum now but get it as a result of starting this node, # then a bunch of nodes might get fenced upnode = None if self.HasQuorum(None): self.debug("Have quorum") return None if not self.templates["Pat:Fencing_start"]: print("No start pattern") return None if not self.templates["Pat:Fencing_ok"]: print("No ok pattern") return None stonith = None stonithPats = [] for peer in self.Env["nodes"]: if self.ShouldBeStatus[peer] != "up": stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) elif self.Env["Stack"] == "corosync (cman)": # There is a delay between gaining quorum and CMAN starting fencing # This can mean that even nodes that are fully up get fenced # There is no use fighting it, just look for everyone so that CTS doesn't get confused stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) stonith.setwatch() return stonith
def prepare_fencing_watcher(self, name): # If we don't have quorum now but get it as a result of starting this node, # then a bunch of nodes might get fenced upnode = None if self.HasQuorum(None): self.debug("Have quorum") return None if not self.templates["Pat:Fencing_start"]: print("No start pattern") return None if not self.templates["Pat:Fencing_ok"]: print("No ok pattern") return None stonith = None stonithPats = [] for peer in self.Env["nodes"]: if self.ShouldBeStatus[peer] != "up": stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) stonith.setwatch() return stonith
def setup_scenario(self, cluster_manager): cluster_manager.log("Reusing cluster") # Disable STONITH by default. A dedicated ScenarioComponent # is in charge of enabling it if requested self.rsh_check(self.Env["nodes"][0], "pcs property set stonith-enabled=false") # Stop and remove galera if it exists # Note1: in order to avoid error when stopping the resource while # in unknown state, we first reprobe the resource state. # Note2: if you clean and delete before pacemaker had a # chance to re-probe state, it will consider resource is stopped # and will happily delete the resource from the cib even if # galera is still running! # Note3: after a cleanup, pacemaker may log a warning log # if it finds the resource is still running. This does not # count as an error for the CTS test target=self.Env["nodes"][0] rc = self.rsh(target, "pcs resource unmanage galera") if rc == 0: patterns = [r"crmd.*:\s*Initiating action.*: probe_complete probe_complete-%s on %s"%(n,n) \ for n in self.Env["nodes"]] watch=LogWatcher(self.Env["LogFileName"], patterns, None, self.Env["DeadTime"], kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) watch.setwatch() self.rsh(target, "pcs resource cleanup galera") watch.lookforall() assert not watch.unmatched, watch.unmatched self.rsh(target, "pcs resource disable galera") self.rsh(target, "pcs resource manage galera") self.rsh(target, "pcs resource delete galera --wait")
def setup_scenario(self, cluster_manager): # mysql setup self.init_and_setup_mysql_defaults() self.setup_galera_config() remote_authkey = "/etc/pacemaker/authkey" if not self.rsh.exists_on_all(remote_authkey, self.Env["nodes"]): self.log("Creating auth key for communication with pacemaker remote") with tempfile.NamedTemporaryFile() as tmp: tmp.write(os.urandom(4096)) tmp.flush() self.copy_to_nodes([(tmp.name, remote_authkey)], True, "root:haclient", "440") # cluster_manager.prepare() # stop cluster if previously running, failure is not fatal for node in self.Env["nodes"]: self.rsh(node, "pcs cluster destroy") self.rsh(node, "systemctl stop pacemaker_remote") self.rsh(node, "systemctl enable pacemaker") # reconfigure cluster for 2-nodes + one remote arbitrator self.Env["arb"]=self.Env["nodes"][-1] self.Env["nodes"]=self.Env["nodes"][:-1] self.rsh_check(self.Env["nodes"][0], "pcs cluster setup --force --name ratester %s %s" % \ (self.Env["nodes"][0],self.Env["nodes"][1])) # note: setting up cluster disable pacemaker service. re-enable it self.rsh_check(self.Env["nodes"][0], "systemctl enable pacemaker") self.rsh_check(self.Env["nodes"][0], "pcs cluster start --all") # TODO: better way to wait until cluster is started time.sleep(8) # Disable STONITH by default. A dedicated ScenarioComponent # is in charge of enabling it if requested self.rsh_check(self.Env["nodes"][0], "pcs property set stonith-enabled=false") for node in self.Env["nodes"]: self.rsh_check(node, "pcs property set --node %s osprole=controller"%node) # cluster_manager.prepare() # pacemaker remote to host garbd res=self.rsh_check(self.Env["arb"], "systemctl disable pacemaker") res=self.rsh_check(self.Env["arb"], "systemctl enable pacemaker_remote") res=self.rsh_check(self.Env["arb"], "systemctl start pacemaker_remote") remote_ok_pat = self.ratemplates.build("Pat:RscRemoteOp", "start", "arb", "\S+", "ok") watch=LogWatcher(self.Env["LogFileName"], [remote_ok_pat], None, self.Env["DeadTime"], kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) # watch = self.create_watch([remote_ok_pat], self.Env["DeadTime"]) watch.setwatch() res=self.rsh_check(self.Env["nodes"][0], "pcs resource create arb ocf:pacemaker:remote server=%s reconnect_interval=60 op monitor interval=20"%self.Env["arb"]) watch.lookforall() assert not watch.unmatched, watch.unmatched self.rsh_check(self.Env["nodes"][0], "pcs property set --node arb osprole=arbitrator") # there's no selinux context for garbd currently res=self.rsh_check(self.Env["arb"], "test -x /usr/sbin/setenforce && setenforce 0 || true")
def startall(self, nodelist=None, verbose=False, quick=False): '''Start the cluster manager on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist = self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": self.ns.WaitForAllNodesToComeUp(nodelist, 300) if not quick: # This is used for "basic sanity checks", so only start one node ... if not self.StartaCM(node, verbose=verbose): return 0 return 1 # Approximation of SimulStartList for --boot watchpats = [] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in nodelist: watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) watchpats.append(self.templates["Pat:Local_started"] % node) watchpats.append(self.templates["Pat:They_up"] % (nodelist[0], node)) # Start all the nodes - at about the same time... watch = LogWatcher(self.Env["LogFileName"], watchpats, "fast-start", self.Env["DeadTime"] + 10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) watch.setwatch() if not self.StartaCM(nodelist[0], verbose=verbose): return 0 for node in nodelist: self.StartaCMnoBlock(node, verbose=verbose) watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log("Warn: Startup pattern not found: %s" % (regex)) if not self.cluster_stable(): self.logger.log("Cluster did not stabilize") return 0 return 1
def setup_scenario(self, cluster_manager): # pre-requisites prerequisite = ["/usr/bin/gdb", "/usr/bin/screen", "/usr/bin/dig"] missing_reqs = False for req in prerequisite: if not self.rsh.exists_on_all(req, self.Env["nodes"]): self.log("error: %s could not be found on remote nodes. " "Please install the necessary package to run the tests"% req) missing_reqs = True assert not missing_reqs # galera-specific data test_scripts = ["kill-during-txn.gdb", "slow_down_sst.sh"] for node in self.Env["nodes"]: for script in test_scripts: src = os.path.join(os.path.dirname(os.path.abspath(__file__)), script) rc = self.rsh.cp(src, "root@%s:/tmp/%s" % (node, script)) assert rc == 0, \ "failed to copy data \"%s\" on remote node \"%s\"" % \ (src, node) # mysql setup self.init_and_setup_mysql_defaults() self.setup_galera_config() # clean up any traffic control on target network interface for node in self.Env["nodes"]: self.rsh(node, "/tmp/slow_down_sst.sh -n %s off"%node) # stop cluster if previously running, failure is not fatal for node in self.Env["nodes"]: self.rsh(node, "pcs cluster destroy") self.rsh(node, "systemctl enable pacemaker") self.rsh(node, "systemctl stop pacemaker_remote") self.rsh(node, "systemctl disable pacemaker_remote") # create a new cluster # note: setting up cluster disable pacemaker service. re-enable it patterns = [r"crmd.*:\s*notice:\sState\stransition\sS_STARTING(\s->.*origin=do_started)?", r"crmd.*:\s*notice:\sState\stransition\s.*->\sS_IDLE(\s.*origin=notify_crmd)?"] watch = LogWatcher(self.Env["LogFileName"], patterns, None, self.Env["DeadTime"], kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) watch.setwatch() self.rsh_check(self.Env["nodes"][0], "pcs cluster setup --force --name ratester %s" % \ " ".join(self.Env["nodes"])) self.rsh_check(self.Env["nodes"][0], "systemctl enable pacemaker") self.rsh_check(self.Env["nodes"][0], "pcs cluster start --all") # Disable STONITH by default. A dedicated ScenarioComponent # is in charge of enabling it if requested self.rsh_check(self.Env["nodes"][0], "pcs property set stonith-enabled=false") watch.lookforall() assert not watch.unmatched, watch.unmatched
def test_node_CM(self, node): '''Report the status of the cluster manager on a given node''' watchpats = [] watchpats.append("Current ping state: (S_IDLE|S_NOT_DC)") watchpats.append(self.templates["Pat:NonDC_started"] % node) watchpats.append(self.templates["Pat:DC_started"] % node) idle_watch = LogWatcher(self.Env["LogFileName"], watchpats, "ClusterIdle", hosts=[node], kind=self.Env["LogWatcher"]) idle_watch.setwatch() out = self.rsh(node, self.templates["StatusCmd"] % node, 1) self.debug("Node %s status: '%s'" % (node, out)) if not out or (out.find('ok') < 0): if self.ShouldBeStatus[node] == "up": self.log( "Node status for %s is %s but we think it should be %s" % (node, "down", self.ShouldBeStatus[node])) self.ShouldBeStatus[node] = "down" return 0 if self.ShouldBeStatus[node] == "down": self.log( "Node status for %s is %s but we think it should be %s: %s" % (node, "up", self.ShouldBeStatus[node], out)) self.ShouldBeStatus[node] = "up" # check the output first - because syslog-ng loses messages if out.find('S_NOT_DC') != -1: # Up and stable return 2 if out.find('S_IDLE') != -1: # Up and stable return 2 # fall back to syslog-ng and wait if not idle_watch.look(): # just up self.debug("Warn: Node %s is unstable: %s" % (node, out)) return 1 # Up and stable return 2
def startall(self, nodelist=None, verbose=False, quick=False): '''Start the cluster manager on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist = self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": self.ns.WaitForAllNodesToComeUp(nodelist, 300) if not quick: # This is used for "basic sanity checks", so only start one node ... if not self.StartaCM(node, verbose=verbose): return 0 return 1 # Approximation of SimulStartList for --boot watchpats = [ ] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in nodelist: watchpats.append(self.templates["Pat:Local_started"] % node) watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... watch = LogWatcher(self.Env["LogFileName"], watchpats, "fast-start", self.Env["DeadTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) watch.setwatch() if not self.StartaCM(nodelist[0], verbose=verbose): return 0 for node in nodelist: self.StartaCMnoBlock(node, verbose=verbose) watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" % (regex)) if not self.cluster_stable(): self.logger.log("Cluster did not stabilize") return 0 return 1
def test_node_CM(self, node): '''Report the status of the cluster manager on a given node''' watchpats = [ ] watchpats.append("Current ping state: (S_IDLE|S_NOT_DC)") watchpats.append(self.templates["Pat:Slave_started"]%node) watchpats.append(self.templates["Pat:Master_started"]%node) idle_watch = LogWatcher(self.Env["LogFileName"], watchpats, "ClusterIdle", hosts=[node], kind=self.Env["LogWatcher"]) idle_watch.setwatch() out = self.rsh(node, self.templates["StatusCmd"]%node, 1) self.debug("Node %s status: '%s'" %(node, out)) if not out or string.find(out, 'ok') < 0: if self.ShouldBeStatus[node] == "up": self.log( "Node status for %s is %s but we think it should be %s" % (node, "down", self.ShouldBeStatus[node])) self.ShouldBeStatus[node] = "down" return 0 if self.ShouldBeStatus[node] == "down": self.log( "Node status for %s is %s but we think it should be %s: %s" % (node, "up", self.ShouldBeStatus[node], out)) self.ShouldBeStatus[node] = "up" # check the output first - because syslog-ng looses messages if string.find(out, 'S_NOT_DC') != -1: # Up and stable return 2 if string.find(out, 'S_IDLE') != -1: # Up and stable return 2 # fall back to syslog-ng and wait if not idle_watch.look(): # just up self.debug("Warn: Node %s is unstable: %s" % (node, out)) return 1 # Up and stable return 2
def setup_scenario(self, cluster_manager): # consider cluster has 2-nodes + one remote arbitrator cluster_manager.log("Reusing cluster") target=self.Env["nodes"][0] self.Env["arb"]=self.Env["nodes"][-1] self.rsh_check(target, "pcs property set --node arb osprole=arbitrator") # attempt at cleaning up and remove garbd if it exists rc = self.rsh(target, "pcs resource unmanage garbd") if rc == 0: self.rsh(target, "pcs resource cleanup garbd") self.rsh(target, "pcs resource disable garbd") self.rsh(target, "pcs resource manage garbd") self.rsh(target, "pcs resource delete garbd --wait") self.Env["nodes"]=self.Env["nodes"][:-1] for node in self.Env["nodes"]: self.rsh_check(node, "pcs property set --node %s osprole=controller"%node) # Stop and remove galera if it exists # Note1: in order to avoid error when stopping the resource while # in unknown state, we first reprobe the resource state. # Note2: if you clean and delete before pacemaker had a # chance to re-probe state, it will consider resource is stopped # and will happily delete the resource from the cib even if # galera is still running! # Note3: after a cleanup, pacemaker may log a warning log # if it finds the resource is still running. This does not # count as an error for the CTS test rc = self.rsh(target, "pcs resource unmanage galera") if rc == 0: patterns = [r"crmd.*:\s*Initiating action.*: probe_complete probe_complete-%s on %s"%(n,n) \ for n in self.Env["nodes"]] watch=LogWatcher(self.Env["LogFileName"], patterns, None, self.Env["DeadTime"], kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) watch.setwatch() self.rsh(target, "pcs resource cleanup galera") watch.lookforall() assert not watch.unmatched, watch.unmatched self.rsh(target, "pcs resource disable galera") self.rsh(target, "pcs resource manage galera") self.rsh(target, "pcs resource delete galera --wait")
def prepare_fencing_watcher(self, node): # If we don't have quorum now but get it as a result of starting this node, # then a bunch of nodes might get fenced upnode = None if self.HasQuorum(None): return None if not self.has_key("Pat:Fencing_start"): return None if not self.has_key("Pat:Fencing_ok"): return None stonith = None stonithPats = [] for peer in self.Env["nodes"]: if peer != node and self.ShouldBeStatus[peer] != "up": stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) elif self.Env["Stack"] == "corosync (cman)": # There is a delay between gaining quorum and CMAN starting fencing # This can mean that even nodes that are fully up get fenced # There is no use fighting it, just look for everyone so that CTS doesn't get confused stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) if peer != node and not upnode and self.ShouldBeStatus[peer] == "up": upnode = peer # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status if not upnode: return None stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=[upnode], kind=self.Env["LogWatcher"]) stonith.setwatch() return stonith
def StartaCM(self, node, verbose=False): '''Start up the cluster manager on a given node''' if verbose: self.logger.log("Starting %s on node %s" % (self.templates["Name"], node)) else: self.debug("Starting %s on node %s" % (self.templates["Name"], node)) ret = 1 if not node in self.ShouldBeStatus: self.ShouldBeStatus[node] = "down" if self.ShouldBeStatus[node] != "down": return 1 patterns = [] # Technically we should always be able to notice ourselves starting patterns.append(self.templates["Pat:Local_started"] % node) if self.upcount() == 0: patterns.append(self.templates["Pat:DC_started"] % node) else: patterns.append(self.templates["Pat:NonDC_started"] % node) watch = LogWatcher(self.Env["LogFileName"], patterns, "StartaCM", self.Env["StartTime"] + 10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) self.install_config(node) self.ShouldBeStatus[node] = "any" if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.logger.log("%s was already started" % (node)) return 1 stonith = self.prepare_fencing_watcher(node) watch.setwatch() if self.rsh(node, self.templates["StartCmd"]) != 0: self.logger.log("Warn: Start command failed on node %s" % (node)) self.fencing_cleanup(node, stonith) return None self.ShouldBeStatus[node] = "up" watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log("Warn: Startup pattern not found: %s" % (regex)) if watch_result and self.cluster_stable(self.Env["DeadTime"]): #self.debug("Found match: "+ repr(watch_result)) self.fencing_cleanup(node, stonith) return 1 elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.fencing_cleanup(node, stonith) return 1 self.logger.log("Warn: Start failed for node %s" % (node)) return None
class Scenario: ('''The basic idea of a scenario is that of an ordered list of ScenarioComponent objects. Each ScenarioComponent is SetUp() in turn, and then after the tests have been run, they are torn down using TearDown() (in reverse order). A Scenario is applicable to a particular cluster manager iff each ScenarioComponent is applicable. A partially set up scenario is torn down if it fails during setup. ''') def __init__(self, ClusterManager, Components, Audits, Tests): "Initialize the Scenario from the list of ScenarioComponents" self.ClusterManager = ClusterManager self.Components = Components self.Audits = Audits self.Tests = Tests self.BadNews = None self.TestSets = [] self.Stats = {"success": 0, "failure": 0, "BadNews": 0, "skipped": 0} self.Sets = [] #self.ns=CTS.NodeStatus(self.Env) for comp in Components: if not issubclass(comp.__class__, ScenarioComponent): raise ValueError( "Init value must be subclass of ScenarioComponent") for audit in Audits: if not issubclass(audit.__class__, ClusterAudit): raise ValueError("Init value must be subclass of ClusterAudit") for test in Tests: if not issubclass(test.__class__, CTSTest): raise ValueError("Init value must be a subclass of CTSTest") def IsApplicable(self): ('''A Scenario IsApplicable() iff each of its ScenarioComponents IsApplicable() ''') for comp in self.Components: if not comp.IsApplicable(): return None return 1 def SetUp(self): '''Set up the Scenario. Return TRUE on success.''' self.ClusterManager.prepare() self.audit() # Also detects remote/local log config self.ClusterManager.StatsMark(0) self.ClusterManager.ns.WaitForAllNodesToComeUp( self.ClusterManager.Env["nodes"]) self.audit() if self.ClusterManager.Env["valgrind-tests"]: self.ClusterManager.install_helper("cts.supp") self.BadNews = LogWatcher(self.ClusterManager.Env["LogFileName"], self.ClusterManager.templates.get_patterns( self.ClusterManager.Env["Name"], "BadNews"), "BadNews", 0, kind=self.ClusterManager.Env["LogWatcher"], hosts=self.ClusterManager.Env["nodes"]) self.BadNews.setwatch( ) # Call after we've figured out what type of log watching to do in LogAudit j = 0 while j < len(self.Components): if not self.Components[j].SetUp(self.ClusterManager): # OOPS! We failed. Tear partial setups down. self.audit() self.ClusterManager.log("Tearing down partial setup") self.TearDown(j) return None j = j + 1 self.audit() return 1 def TearDown(self, max=None): '''Tear Down the Scenario - in reverse order.''' if max == None: max = len(self.Components) - 1 j = max while j >= 0: self.Components[j].TearDown(self.ClusterManager) j = j - 1 self.audit() self.ClusterManager.StatsExtract() def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not name in self.Stats: self.Stats[name] = 0 self.Stats[name] = self.Stats[name] + 1 def run(self, Iterations): self.ClusterManager.oprofileStart() try: self.run_loop(Iterations) self.ClusterManager.oprofileStop() except: self.ClusterManager.oprofileStop() raise def run_loop(self, Iterations): raise ValueError("Abstract Class member (run_loop)") def run_test(self, test, testcount): nodechoice = self.ClusterManager.Env.RandomNode() ret = 1 where = "" did_run = 0 self.ClusterManager.StatsMark(testcount) self.ClusterManager.instance_errorstoignore_clear() self.ClusterManager.log(("Running test %s" % test.name).ljust(35) + (" (%s) " % nodechoice).ljust(15) + "[" + ("%d" % testcount).rjust(3) + "]") starttime = test.set_timer() if not test.setup(nodechoice): self.ClusterManager.log("Setup failed") ret = 0 elif not test.canrunnow(nodechoice): self.ClusterManager.log("Skipped") test.skipped() else: did_run = 1 ret = test(nodechoice) if not test.teardown(nodechoice): self.ClusterManager.log("Teardown failed") answer = raw_input('Continue? [nY] ') if answer and answer == "n": raise ValueError("Teardown of %s on %s failed" % (test.name, nodechoice)) ret = 0 stoptime = time.time() self.ClusterManager.oprofileSave(testcount) elapsed_time = stoptime - starttime test_time = stoptime - test.get_timer() if not test["min_time"]: test["elapsed_time"] = elapsed_time test["min_time"] = test_time test["max_time"] = test_time else: test["elapsed_time"] = test["elapsed_time"] + elapsed_time if test_time < test["min_time"]: test["min_time"] = test_time if test_time > test["max_time"]: test["max_time"] = test_time if ret: self.incr("success") test.log_timer() else: self.incr("failure") self.ClusterManager.statall() did_run = 1 # Force the test count to be incrimented anyway so test extraction works self.audit(test.errorstoignore()) return did_run def summarize(self): self.ClusterManager.log("****************") self.ClusterManager.log("Overall Results:" + repr(self.Stats)) self.ClusterManager.log("****************") stat_filter = { "calls": 0, "failure": 0, "skipped": 0, "auditfail": 0, } self.ClusterManager.log("Test Summary") for test in self.Tests: for key in list(stat_filter.keys()): stat_filter[key] = test.Stats[key] self.ClusterManager.log(("Test %s: " % test.name).ljust(25) + " %s" % repr(stat_filter)) self.ClusterManager.debug("Detailed Results") for test in self.Tests: self.ClusterManager.debug(("Test %s: " % test.name).ljust(25) + " %s" % repr(test.Stats)) self.ClusterManager.log("<<<<<<<<<<<<<<<< TESTS COMPLETED") def audit(self, LocalIgnore=[]): errcount = 0 ignorelist = [] ignorelist.append("CTS:") ignorelist.extend(LocalIgnore) ignorelist.extend(self.ClusterManager.errorstoignore()) ignorelist.extend(self.ClusterManager.instance_errorstoignore()) # This makes sure everything is stabilized before starting... failed = 0 for audit in self.Audits: if not audit(): self.ClusterManager.log("Audit " + audit.name() + " FAILED.") failed += 1 else: self.ClusterManager.debug("Audit " + audit.name() + " passed.") while errcount < 1000: match = None if self.BadNews: match = self.BadNews.look(0) if match: add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): add_err = 0 if add_err == 1: self.ClusterManager.log("BadNews: " + match) self.incr("BadNews") errcount = errcount + 1 else: break else: answer = raw_input('Big problems. Continue? [nY]') if answer and answer == "n": self.ClusterManager.log("Shutting down.") self.summarize() self.TearDown() raise ValueError("Looks like we hit a BadNews jackpot!") if self.BadNews: self.BadNews.end() return failed
def StartaCM(self, node, verbose=False): '''Start up the cluster manager on a given node''' if verbose: self.logger.log("Starting %s on node %s" % (self.templates["Name"], node)) else: self.debug("Starting %s on node %s" % (self.templates["Name"], node)) ret = 1 if not node in self.ShouldBeStatus: self.ShouldBeStatus[node] = "down" if self.ShouldBeStatus[node] != "down": return 1 patterns = [] # Technically we should always be able to notice ourselves starting patterns.append(self.templates["Pat:Local_started"] % node) if self.upcount() == 0: patterns.append(self.templates["Pat:Master_started"] % node) else: patterns.append(self.templates["Pat:Slave_started"] % node) watch = LogWatcher( self.Env["LogFileName"], patterns, "StartaCM", self.Env["StartTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) self.install_config(node) self.ShouldBeStatus[node] = "any" if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.logger.log ("%s was already started" % (node)) return 1 # Clear out the host cache so autojoin can be exercised if self.clear_cache: self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") if not(self.Env["valgrind-tests"]): startCmd = self.templates["StartCmd"] else: if self.Env["valgrind-prefix"]: prefix = self.Env["valgrind-prefix"] else: prefix = "cts" startCmd = """G_SLICE=always-malloc HA_VALGRIND_ENABLED='%s' VALGRIND_OPTS='%s --log-file=/tmp/%s-%s.valgrind' %s""" % ( self.Env["valgrind-procs"], self.Env["valgrind-opts"], prefix, """%p""", self.templates["StartCmd"]) stonith = self.prepare_fencing_watcher(node) watch.setwatch() if self.rsh(node, startCmd) != 0: self.logger.log ("Warn: Start command failed on node %s" % (node)) self.fencing_cleanup(node, stonith) return None self.ShouldBeStatus[node] = "up" watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" % (regex)) if watch_result and self.cluster_stable(self.Env["DeadTime"]): #self.debug("Found match: "+ repr(watch_result)) self.fencing_cleanup(node, stonith) return 1 elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.fencing_cleanup(node, stonith) return 1 self.logger.log ("Warn: Start failed for node %s" % (node)) return None
class Scenario: ( '''The basic idea of a scenario is that of an ordered list of ScenarioComponent objects. Each ScenarioComponent is SetUp() in turn, and then after the tests have been run, they are torn down using TearDown() (in reverse order). A Scenario is applicable to a particular cluster manager iff each ScenarioComponent is applicable. A partially set up scenario is torn down if it fails during setup. ''') def __init__(self, ClusterManager, Components, Audits, Tests): "Initialize the Scenario from the list of ScenarioComponents" self.ClusterManager = ClusterManager self.Components = Components self.Audits = Audits self.Tests = Tests self.BadNews = None self.TestSets = [] self.Stats = {"success":0, "failure":0, "BadNews":0, "skipped":0} self.Sets = [] #self.ns=CTS.NodeStatus(self.Env) for comp in Components: if not issubclass(comp.__class__, ScenarioComponent): raise ValueError("Init value must be subclass of ScenarioComponent") for audit in Audits: if not issubclass(audit.__class__, ClusterAudit): raise ValueError("Init value must be subclass of ClusterAudit") for test in Tests: if not issubclass(test.__class__, CTSTest): raise ValueError("Init value must be a subclass of CTSTest") def IsApplicable(self): ( '''A Scenario IsApplicable() iff each of its ScenarioComponents IsApplicable() ''' ) for comp in self.Components: if not comp.IsApplicable(): return None return 1 def SetUp(self): '''Set up the Scenario. Return TRUE on success.''' self.ClusterManager.prepare() self.audit() # Also detects remote/local log config self.ClusterManager.StatsMark(0) self.ClusterManager.ns.WaitForAllNodesToComeUp(self.ClusterManager.Env["nodes"]) self.audit() if self.ClusterManager.Env["valgrind-tests"]: self.ClusterManager.install_helper("cts.supp") self.BadNews = LogWatcher(self.ClusterManager.Env["LogFileName"], self.ClusterManager.templates.get_patterns( self.ClusterManager.Env["Name"], "BadNews"), "BadNews", 0, kind=self.ClusterManager.Env["LogWatcher"], hosts=self.ClusterManager.Env["nodes"]) self.BadNews.setwatch() # Call after we've figured out what type of log watching to do in LogAudit j = 0 while j < len(self.Components): if not self.Components[j].SetUp(self.ClusterManager): # OOPS! We failed. Tear partial setups down. self.audit() self.ClusterManager.log("Tearing down partial setup") self.TearDown(j) return None j = j + 1 self.audit() return 1 def TearDown(self, max=None): '''Tear Down the Scenario - in reverse order.''' if max == None: max = len(self.Components)-1 j = max while j >= 0: self.Components[j].TearDown(self.ClusterManager) j = j - 1 self.audit() self.ClusterManager.StatsExtract() def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not name in self.Stats: self.Stats[name] = 0 self.Stats[name] = self.Stats[name]+1 def run(self, Iterations): self.ClusterManager.oprofileStart() try: self.run_loop(Iterations) self.ClusterManager.oprofileStop() except: self.ClusterManager.oprofileStop() raise def run_loop(self, Iterations): raise ValueError("Abstract Class member (run_loop)") def run_test(self, test, testcount): nodechoice = self.ClusterManager.Env.RandomNode() ret = 1 where = "" did_run = 0 self.ClusterManager.StatsMark(testcount) self.ClusterManager.instance_errorstoignore_clear() self.ClusterManager.log(("Running test %s" % test.name).ljust(35) + (" (%s) " % nodechoice).ljust(15) + "[" + ("%d" % testcount).rjust(3) + "]") starttime = test.set_timer() if not test.setup(nodechoice): self.ClusterManager.log("Setup failed") ret = 0 elif not test.canrunnow(nodechoice): self.ClusterManager.log("Skipped") test.skipped() else: did_run = 1 ret = test(nodechoice) if not test.teardown(nodechoice): self.ClusterManager.log("Teardown failed") if self.ClusterManager.Env["continue"] == 1: answer = "Y" else: try: answer = raw_input('Continue? [nY]') except EOFError as e: answer = "n" if answer and answer == "n": raise ValueError("Teardown of %s on %s failed" % (test.name, nodechoice)) ret = 0 stoptime = time.time() self.ClusterManager.oprofileSave(testcount) elapsed_time = stoptime - starttime test_time = stoptime - test.get_timer() if not test["min_time"]: test["elapsed_time"] = elapsed_time test["min_time"] = test_time test["max_time"] = test_time else: test["elapsed_time"] = test["elapsed_time"] + elapsed_time if test_time < test["min_time"]: test["min_time"] = test_time if test_time > test["max_time"]: test["max_time"] = test_time if ret: self.incr("success") test.log_timer() else: self.incr("failure") self.ClusterManager.statall() did_run = 1 # Force the test count to be incremented anyway so test extraction works self.audit(test.errorstoignore()) return did_run def summarize(self): self.ClusterManager.log("****************") self.ClusterManager.log("Overall Results:" + repr(self.Stats)) self.ClusterManager.log("****************") stat_filter = { "calls":0, "failure":0, "skipped":0, "auditfail":0, } self.ClusterManager.log("Test Summary") for test in self.Tests: for key in list(stat_filter.keys()): stat_filter[key] = test.Stats[key] self.ClusterManager.log(("Test %s: "%test.name).ljust(25) + " %s"%repr(stat_filter)) self.ClusterManager.debug("Detailed Results") for test in self.Tests: self.ClusterManager.debug(("Test %s: "%test.name).ljust(25) + " %s"%repr(test.Stats)) self.ClusterManager.log("<<<<<<<<<<<<<<<< TESTS COMPLETED") def audit(self, LocalIgnore=[]): errcount = 0 ignorelist = [] ignorelist.append("CTS:") ignorelist.extend(LocalIgnore) ignorelist.extend(self.ClusterManager.errorstoignore()) ignorelist.extend(self.ClusterManager.instance_errorstoignore()) # This makes sure everything is stabilized before starting... failed = 0 for audit in self.Audits: if not audit(): self.ClusterManager.log("Audit " + audit.name() + " FAILED.") failed += 1 else: self.ClusterManager.debug("Audit " + audit.name() + " passed.") while errcount < 1000: match = None if self.BadNews: match = self.BadNews.look(0) if match: add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): add_err = 0 if add_err == 1: self.ClusterManager.log("BadNews: " + match) self.incr("BadNews") errcount = errcount + 1 else: break else: if self.ClusterManager.Env["continue"] == 1: answer = "Y" else: try: answer = raw_input('Big problems. Continue? [nY]') except EOFError as e: answer = "n" if answer and answer == "n": self.ClusterManager.log("Shutting down.") self.summarize() self.TearDown() raise ValueError("Looks like we hit a BadNews jackpot!") if self.BadNews: self.BadNews.end() return failed
class Scenario: ('''The basic idea of a scenario is that of an ordered list of ScenarioComponent objects. Each ScenarioComponent is SetUp() in turn, and then after the tests have been run, they are torn down using TearDown() (in reverse order). A Scenario is applicable to a particular cluster manager iff each ScenarioComponent is applicable. A partially set up scenario is torn down if it fails during setup. ''') def __init__(self, ClusterManager, Components, Audits, Tests): "Initialize the Scenario from the list of ScenarioComponents" self.ClusterManager = ClusterManager self.Components = Components self.Audits = Audits self.Tests = Tests self.BadNews = None self.TestSets = [] self.Stats = {"success": 0, "failure": 0, "BadNews": 0, "skipped": 0} self.Sets = [] #self.ns=CTS.NodeStatus(self.Env) for comp in Components: if not issubclass(comp.__class__, ScenarioComponent): raise ValueError( "Init value must be subclass of ScenarioComponent") for audit in Audits: if not issubclass(audit.__class__, ClusterAudit): raise ValueError("Init value must be subclass of ClusterAudit") for test in Tests: if not issubclass(test.__class__, CTSTest): raise ValueError("Init value must be a subclass of CTSTest") def IsApplicable(self): ('''A Scenario IsApplicable() iff each of its ScenarioComponents IsApplicable() ''') for comp in self.Components: if not comp.IsApplicable(): return None return 1 def SetUp(self): '''Set up the Scenario. Return TRUE on success.''' self.ClusterManager.prepare() self.audit() # Also detects remote/local log config self.ClusterManager.StatsMark(0) self.ClusterManager.ns.WaitForAllNodesToComeUp( self.ClusterManager.Env["nodes"]) self.audit() if self.ClusterManager.Env["valgrind-tests"]: self.ClusterManager.install_helper("cts.supp") self.BadNews = LogWatcher(self.ClusterManager.Env["LogFileName"], self.ClusterManager.templates.get_patterns( self.ClusterManager.Env["Name"], "BadNews"), "BadNews", 0, kind=self.ClusterManager.Env["LogWatcher"], hosts=self.ClusterManager.Env["nodes"]) self.BadNews.setwatch( ) # Call after we've figured out what type of log watching to do in LogAudit j = 0 while j < len(self.Components): if not self.Components[j].SetUp(self.ClusterManager): # OOPS! We failed. Tear partial setups down. self.audit() self.ClusterManager.log("Tearing down partial setup") self.TearDown(j) return None j = j + 1 self.audit() return 1 def TearDown(self, max=None): '''Tear Down the Scenario - in reverse order.''' if max == None: max = len(self.Components) - 1 j = max while j >= 0: self.Components[j].TearDown(self.ClusterManager) j = j - 1 self.audit() self.ClusterManager.StatsExtract() def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not name in self.Stats: self.Stats[name] = 0 self.Stats[name] = self.Stats[name] + 1 def run(self, Iterations): self.ClusterManager.oprofileStart() try: self.run_loop(Iterations) self.ClusterManager.oprofileStop() except: self.ClusterManager.oprofileStop() raise def run_loop(self, Iterations): raise ValueError("Abstract Class member (run_loop)") def run_test(self, test, testcount): nodechoice = self.ClusterManager.Env.RandomNode() ret = 1 where = "" did_run = 0 self.ClusterManager.StatsMark(testcount) self.ClusterManager.instance_errorstoignore_clear() self.ClusterManager.log(("Running test %s" % test.name).ljust(35) + (" (%s) " % nodechoice).ljust(15) + "[" + ("%d" % testcount).rjust(3) + "]") starttime = test.set_timer() if not test.setup(nodechoice): self.ClusterManager.log("Setup failed") ret = 0 elif not test.canrunnow(nodechoice): self.ClusterManager.log("Skipped") test.skipped() else: did_run = 1 ret = test(nodechoice) if not test.teardown(nodechoice): self.ClusterManager.log("Teardown failed") if self.ClusterManager.Env["continue"] == 1: answer = "Y" else: try: answer = raw_input('Continue? [nY]') except EOFError, e: answer = "n" if answer and answer == "n": raise ValueError("Teardown of %s on %s failed" % (test.name, nodechoice)) ret = 0 stoptime = time.time() self.ClusterManager.oprofileSave(testcount) elapsed_time = stoptime - starttime test_time = stoptime - test.get_timer() if not test["min_time"]: test["elapsed_time"] = elapsed_time test["min_time"] = test_time test["max_time"] = test_time else: test["elapsed_time"] = test["elapsed_time"] + elapsed_time if test_time < test["min_time"]: test["min_time"] = test_time if test_time > test["max_time"]: test["max_time"] = test_time if ret: self.incr("success") test.log_timer() else: self.incr("failure") self.ClusterManager.statall() did_run = 1 # Force the test count to be incremented anyway so test extraction works self.audit(test.errorstoignore()) return did_run