def main(argv): serverlist = '' desynch = False try: opts, args = getopt.getopt(argv, 'hs:d', ['serverlist=', 'desynch']) except getopt.GetoptError as err: print err usage() sys.exit(2) for opt, arg in opts: if opt in ('-h', '--help'): usage() sys.exit() elif opt in ('-s', '--serverlist'): serverlist = arg elif opt in ('-d', '--desynch'): desynch = True servers = [] if len(serverlist) == 0: # read server list from config.py hosts = config.hosts for host in config.hosts: servers.append(["ssh %s" %(host[0])]) else: hosts = serverlist.split(',') for host in hosts: servers.append(["ssh %s" %(host)]) if desynch: smux.create(len(servers), servers) else: # synchronized panes smux.create(len(servers), servers, executeBeforeAttach=lambda : smux.tcmd("setw synchronize-panes on"))
def replay(file, numPanesPerWindow = 4, logDir = None, dryrun = False): serverAndCommand = [] guts = file.read() # Coordinator finding guts = findServerAndCommand(guts, 'Coordinator started on ([a-z0-9]+) .*\n' + 'Coordinator command line arguments (.*)', serverAndCommand) # Master finding guts = findServerAndCommand(guts, 'Server started on ([a-z0-9]+) at [^ ]+: (.*)$', serverAndCommand) # Client finding guts = findServerAndCommand(guts, 'Client [^ ]+ started on ([a-z0-9]+): (.*)$', serverAndCommand) # Modify the log directory serverAndCommand = adjustLogDirs(serverAndCommand, logDir) finalCommands = [['ssh ' + x, 'cd ' + os.getcwd(), y] for x,y in serverAndCommand] # Generate smux input instead if dryrun: SEP = '-' * 10 print "PANES_PER_WINDOW = %d" % numPanesPerWindow print SEP for server in finalCommands: for cmd in server: print cmd print SEP return smux.create(numPanesPerWindow, finalCommands)
def create(servers, superuser): commands = [] for server in servers: get_public_ip = "echo `geni-get manifest | grep %s | egrep -o \"ipv4=.*\" | cut -d'\"' -f2`" % server cmd = "ssh " if superuser: cmd += "root@" cmd += os.popen(get_public_ip).read() commands.append([cmd]) smux.create( len(commands), commands, executeBeforeAttach=lambda: smux.tcmd("setw synchronize-panes on"))
def create(serverRanges): if "," in serverRanges: serverRanges = serverRanges.split(",") else: serverRanges = [serverRanges] servers = [] for serverRange in serverRanges: if "-" in serverRange: left, right = serverRange.split("-") left, right = int(left), int(right) for x in xrange(left, right + 1): servers.append(["ssh rc%02d" % x]) else: x = int(serverRange) servers.append(["ssh rc%02d" % x]) smux.create(len(servers), servers, executeBeforeAttach=lambda: smux.tcmd("setw synchronize-panes on"))
def create(serverRanges): if "," in serverRanges: serverRanges = serverRanges.split(",") else: serverRanges = [serverRanges] servers = [] for serverRange in serverRanges: if '-' in serverRange: left, right = serverRange.split("-") left, right = int(left), int(right) for x in xrange(left, right + 1): servers.append(["ssh rc%02d" % x]) else: x = int(serverRange) servers.append(["ssh rc%02d" % x]) smux.create( len(servers), servers, executeBeforeAttach=lambda: smux.tcmd("setw synchronize-panes on"))
def replay(file, numPanesPerWindow=4, logDir=None, dryrun=False): serverAndCommand = [] guts = file.read() # Coordinator finding guts = findServerAndCommand( guts, 'Coordinator started on ([a-z0-9]+) .*\n' + 'Coordinator command line arguments (.*)', serverAndCommand) # Master finding guts = findServerAndCommand( guts, 'Server started on ([a-z0-9]+) at [^ ]+: (.*)$', serverAndCommand) # Client finding guts = findServerAndCommand(guts, 'Client [^ ]+ started on ([a-z0-9]+): (.*)$', serverAndCommand) # Modify the log directory serverAndCommand = adjustLogDirs(serverAndCommand, logDir) finalCommands = [['ssh ' + x, 'cd ' + os.getcwd(), y] for x, y in serverAndCommand] # Generate smux input instead if dryrun: SEP = '-' * 10 print "PANES_PER_WINDOW = %d" % numPanesPerWindow print SEP for server in finalCommands: for cmd in server: print cmd print SEP return smux.create(numPanesPerWindow, finalCommands)
def main(): description = """\ kmux.py takes a list of pods and a list of commands and generates their cross product in the form of interactive tmux panes, where each pane corresponds to a pod that each set of commands is executed against. The env variables POD, KUBE_CONTEXT will be set in each pane. KUBE_NAMESPACE will be set iff running kubectl against a namespaced context. The three core options are `--pods`, `--kube_context`, and the `commands_file` positional argument. All other options exist only for improved ergonomics, because one can always use an arbitrary program to compute a list of pods and pass them to kmux.py. Command line options can be specified in three locations; higher-numbered places override over-numbered locations. 1) The environmental varabie KMUX_ARGS. 2) Options passed to the current command directly on the command line. 3) When the input file contains a line that starts with `---`, the lines of the input file above that line are joined together by spaces an d treated as options. This option takes precedence over locations 1) and 2) so that commands in an input file can make strong assumptions about which environment they are running in, should they want to. In this section only, lines that start with `#` and blank lines are treated as comments and ignored. The options `--deployment`, `--pod_name_regex`, `--field_selector`, and `--label_selector` all function as filters that are ANDed together. If `--pods` is given, all four of the above options are ignored. """ def selector_join(*args): output = ','.join([x for x in args if x is not None]) return output if not output == '' else None parser = argparse.ArgumentParser( description=description, formatter_class=argparse.RawTextHelpFormatter) parser.add_argument( '--pods', '-p', metavar='PODS', type=str, help='Whitespace-separated list of pods. When given, -r is ignored.') parser.add_argument( '--kube_context', '-k', metavar='KUBE_CONTEXT', type=str, help= 'The Kubernetes context to pull pods from. Defaults to current context.' ) parser.add_argument('--namespace', '-n', metavar='NAMESPACE', type=str, help='Namespace to select K8 objects from.') parser.add_argument( '--pod_name_regex', '-r', metavar='POD_NAME_REGEX', type=re.compile, help='Regular expression matching a pod in the namespace.') parser.add_argument( '--deployment', '-d', metavar='DEPLOYMENT', type=str, help= 'Deployment to select pods from. Defaults to all pods when omitted.') parser.add_argument('--label_selector', '--selector', '-l', metavar='LABEL_SELECTOR', type=str, help='Equivalent to kubectl get pods --selector.') parser.add_argument( '--field_selector', '--field-selector', '-f', metavar='FIELD_SELECTOR', type=str, help='Equivalent to kubectl get pods --field-selector.') parser.add_argument('--all_namespaces', '--all-namespaces', action='store_true', help='Equivalent to kubectl get pods --all-namespaces') parser.add_argument( '--no_create', action='store_true', help= 'Do not create new tmux windows and panes. Run the commands in only the\n' + 'first found pod in the current window. One pane will be created if kmux\n' + 'is not started inside a tmux.') parser.add_argument( '--dry_run', '--dry-run', action='store_true', help= 'Do not run any tmux commands. Instead, write an smux file to stdout.') parser.add_argument( 'commands_file', nargs='?', type=argparse.FileType('r'), default=open(os.devnull, 'r'), help="A file containing shell commands to run in each pane.") # Check for environmental variable with args, command line overrides args = argv[1:] if 'KMUX_ARGS' in os.environ: args = shlex.split(os.environ['KMUX_ARGS']) + args options = parser.parse_args(args) commands = options.commands_file.read().strip().split("\n") # If commands include options `---`, add them and re-evaluate options. Note # that the positional argument `commands_file` is ignored if it appears in # the options at the top of the file. for i, command in enumerate(commands[:]): if command.startswith("---"): args += shlex.split(" ".join([ c.strip() for c in commands[:i] if not c.startswith("#") and c.strip() ])) options = parser.parse_args(args) commands = commands[i + 1:] # Deal with Python split producing an extra empty string at the end when # input is one or zero lines by removing it. if commands[-1] == '': commands = commands[:-1] ################################################################################ # Load K8 config and Initialize magic variables config.load_kube_config() contexts = config.list_kube_config_contexts()[0] current_context = config.list_kube_config_contexts()[1]['name'] current_namespace = None if 'namespace' in config.list_kube_config_contexts()[1]['context']: current_namespace = config.list_kube_config_contexts( )[1]['context']['namespace'] KUBE_CONTEXT = None KUBE_NAMESPACE = "default" if options.kube_context: KUBE_CONTEXT = options.kube_context # This variable is needed because we cannot rely on the value of # KUBE_NAMESPACE to determine whether we found the requested context, since # KUBE_NAMESPACE is None for non-namespaced K8 setups. foundContext = False for context in contexts: if context['name'] == KUBE_CONTEXT: if 'namespace' in context['context']: KUBE_NAMESPACE = context['context']['namespace'] foundContext = True break if not foundContext: print(f"Invalid context {KUBE_CONTEXT} given.") sys.exit(1) else: KUBE_CONTEXT = current_context KUBE_NAMESPACE = current_namespace # Override namespace here if it is explicitly specified. if options.namespace: KUBE_NAMESPACE = options.namespace # Namespace should never be None. Some contexts seem to be able to override # it to None, so we restore it here. if KUBE_NAMESPACE is None: KUBE_NAMESPACE = "default" # SDK wrappers for Kube APIs api_client = config.new_client_from_config(context=KUBE_CONTEXT) core_kube_client = client.CoreV1Api(api_client=api_client) apps_kube_client = client.AppsV1Api(api_client=api_client) if options.pods: pods = options.pods.split() if not options.all_namespaces: podObjects = core_kube_client.list_namespaced_pod( KUBE_NAMESPACE).items else: podObjects = core_kube_client.list_pod_for_all_namespaces().items podObjects = [pod for pod in podObjects if pod.metadata.name in pods] # Note that if a pod name appears in multiple namespaces, it will be # duplicated, which seems desirable since the namespace will be different. if len(podObjects) < len(pods): print( 'The following requested pods do not exist in either all namespaces ' + 'or specified namespace: ') print( set(pods).difference( set([pod.metadata.name for pod in podObjects]))) sys.exit(1) else: # Constrain the pods selected by deployment, if given. We first check for # the deployment option because it may introdue new label_selectors. # 1. Find all ReplicaSets that are owned by the target deployment. # 2. Find all pods owned by the target replica set or replica sets. deployment_label_selector = None if options.deployment: if not options.all_namespaces: deployments = apps_kube_client.list_namespaced_deployment( KUBE_NAMESPACE, field_selector=f"metadata.name={options.deployment}").items else: deployments = apps_kube_client.list_deployment_for_all_namespaces( field_selector=f"metadata.name={options.deployment}").items # Search for matching deployment. chosen_deployment = None for deployment in deployments: if deployment.metadata.name == options.deployment: chosen_deployment = deployment break if chosen_deployment is None: print( f'Deployment "{options.deployment}" does not exist in namespace ' + f'"{KUBE_NAMESPACE}" of context "{KUBE_CONTEXT}".') sys.exit(1) labels = chosen_deployment.spec.selector.match_labels deployment_label_selector = ",".join( [f"{key}={value}" for key, value in labels.items()]) if not options.all_namespaces: replica_sets = apps_kube_client.list_namespaced_replica_set( KUBE_NAMESPACE, label_selector=deployment_label_selector).items else: replica_sets = apps_kube_client.list_replica_set_for_all_namespaces( label_selector=deployment_label_selector).items replica_sets = [x for x in replica_sets if not x.spec.replicas == 0 and \ x.metadata.owner_references and \ x.metadata.owner_references[0].uid == chosen_deployment.metadata.uid] replica_set_uids = [x.metadata.uid for x in replica_sets] # Collect the pods matching the constraints. if not options.all_namespaces: podObjects = core_kube_client.list_namespaced_pod( KUBE_NAMESPACE, label_selector=selector_join(options.label_selector, deployment_label_selector), field_selector=options.field_selector).items else: podObjects = core_kube_client.list_pod_for_all_namespaces( label_selector=selector_join(options.label_selector, deployment_label_selector), field_selector=options.field_selector).items # Filter by deployments explicitly since selectors are not always reliable. if options.deployment: podObjects = [pod for pod in podObjects if pod.metadata.owner_references and \ pod.metadata.owner_references[0].uid in replica_set_uids] # Filter by regex if given if options.pod_name_regex: podObjects = [ pod for pod in podObjects if options.pod_name_regex.match(pod.metadata.name) ] if not podObjects: print("No pods selected.") return ################################################################################ # When all_namespaces is specified, we may have pods in different namespaces, # so it is more correct to get the namespace from the actual pods selected. pod_commands = [[ f'POD={pod.metadata.name}', f'KUBE_CONTEXT={KUBE_CONTEXT}', f'KUBE_NAMESPACE={pod.metadata.namespace}' ] + (commands) for pod in podObjects] if options.dry_run: if options.no_create: print("NO_CREATE") print("USE_THREADS") for pc in pod_commands: print("---") for c in pc: print(c) # Smux will ignore NO_CREATE if more than one command is passed. if options.no_create: break return # We hardcode useThreads to True because it is assumed that operations on # different pods are independent. This can be made an option if it seems # useful to be able to disable. smux.create( 1 if options.no_create else len(pod_commands), pod_commands[:1] if options.no_create else pod_commands, executeAfterCreate=lambda: smux.tcmd("setw synchronize-panes on"), noCreate=options.no_create, useThreads=True)