| import os, copy, logging, errno, fcntl, time, re, weakref, traceback |
| import tarfile |
| import cPickle as pickle |
| import tempfile |
| from autotest_lib.client.common_lib import autotemp, error, log |
| |
| |
| class job_directory(object): |
| """Represents a job.*dir directory.""" |
| |
| |
| class JobDirectoryException(error.AutotestError): |
| """Generic job_directory exception superclass.""" |
| |
| |
| class MissingDirectoryException(JobDirectoryException): |
| """Raised when a directory required by the job does not exist.""" |
| def __init__(self, path): |
| Exception.__init__(self, 'Directory %s does not exist' % path) |
| |
| |
| class UncreatableDirectoryException(JobDirectoryException): |
| """Raised when a directory required by the job is missing and cannot |
| be created.""" |
| def __init__(self, path, error): |
| msg = 'Creation of directory %s failed with exception %s' |
| msg %= (path, error) |
| Exception.__init__(self, msg) |
| |
| |
| class UnwritableDirectoryException(JobDirectoryException): |
| """Raised when a writable directory required by the job exists |
| but is not writable.""" |
| def __init__(self, path): |
| msg = 'Directory %s exists but is not writable' % path |
| Exception.__init__(self, msg) |
| |
| |
| def __init__(self, path, is_writable=False): |
| """ |
| Instantiate a job directory. |
| |
| @param path: The path of the directory. If None a temporary directory |
| will be created instead. |
| @param is_writable: If True, expect the directory to be writable. |
| |
| @raise MissingDirectoryException: raised if is_writable=False and the |
| directory does not exist. |
| @raise UnwritableDirectoryException: raised if is_writable=True and |
| the directory exists but is not writable. |
| @raise UncreatableDirectoryException: raised if is_writable=True, the |
| directory does not exist and it cannot be created. |
| """ |
| if path is None: |
| if is_writable: |
| self._tempdir = autotemp.tempdir(unique_id='autotest') |
| self.path = self._tempdir.name |
| else: |
| raise self.MissingDirectoryException(path) |
| else: |
| self._tempdir = None |
| self.path = path |
| self._ensure_valid(is_writable) |
| |
| |
| def _ensure_valid(self, is_writable): |
| """ |
| Ensure that this is a valid directory. |
| |
| Will check if a directory exists, can optionally also enforce that |
| it be writable. It can optionally create it if necessary. Creation |
| will still fail if the path is rooted in a non-writable directory, or |
| if a file already exists at the given location. |
| |
| @param dir_path A path where a directory should be located |
| @param is_writable A boolean indicating that the directory should |
| not only exist, but also be writable. |
| |
| @raises MissingDirectoryException raised if is_writable=False and the |
| directory does not exist. |
| @raises UnwritableDirectoryException raised if is_writable=True and |
| the directory is not wrtiable. |
| @raises UncreatableDirectoryException raised if is_writable=True, the |
| directory does not exist and it cannot be created |
| """ |
| # ensure the directory exists |
| if is_writable: |
| try: |
| os.makedirs(self.path) |
| except OSError, e: |
| if e.errno != errno.EEXIST or not os.path.isdir(self.path): |
| raise self.UncreatableDirectoryException(self.path, e) |
| elif not os.path.isdir(self.path): |
| raise self.MissingDirectoryException(self.path) |
| |
| # if is_writable=True, also check that the directory is writable |
| if is_writable and not os.access(self.path, os.W_OK): |
| raise self.UnwritableDirectoryException(self.path) |
| |
| |
| @staticmethod |
| def property_factory(attribute): |
| """ |
| Create a job.*dir -> job._*dir.path property accessor. |
| |
| @param attribute A string with the name of the attribute this is |
| exposed as. '_'+attribute must then be attribute that holds |
| either None or a job_directory-like object. |
| |
| @returns A read-only property object that exposes a job_directory path |
| """ |
| @property |
| def dir_property(self): |
| underlying_attribute = getattr(self, '_' + attribute) |
| if underlying_attribute is None: |
| return None |
| else: |
| return underlying_attribute.path |
| return dir_property |
| |
| |
| # decorator for use with job_state methods |
| def with_backing_lock(method): |
| """A decorator to perform a lock-*-unlock cycle. |
| |
| When applied to a method, this decorator will automatically wrap |
| calls to the method in a backing file lock and before the call |
| followed by a backing file unlock. |
| """ |
| def wrapped_method(self, *args, **dargs): |
| already_have_lock = self._backing_file_lock is not None |
| if not already_have_lock: |
| self._lock_backing_file() |
| try: |
| return method(self, *args, **dargs) |
| finally: |
| if not already_have_lock: |
| self._unlock_backing_file() |
| wrapped_method.__name__ = method.__name__ |
| wrapped_method.__doc__ = method.__doc__ |
| return wrapped_method |
| |
| |
| # decorator for use with job_state methods |
| def with_backing_file(method): |
| """A decorator to perform a lock-read-*-write-unlock cycle. |
| |
| When applied to a method, this decorator will automatically wrap |
| calls to the method in a lock-and-read before the call followed by a |
| write-and-unlock. Any operation that is reading or writing state |
| should be decorated with this method to ensure that backing file |
| state is consistently maintained. |
| """ |
| @with_backing_lock |
| def wrapped_method(self, *args, **dargs): |
| self._read_from_backing_file() |
| try: |
| return method(self, *args, **dargs) |
| finally: |
| self._write_to_backing_file() |
| wrapped_method.__name__ = method.__name__ |
| wrapped_method.__doc__ = method.__doc__ |
| return wrapped_method |
| |
| |
| |
| class job_state(object): |
| """A class for managing explicit job and user state, optionally persistent. |
| |
| The class allows you to save state by name (like a dictionary). Any state |
| stored in this class should be picklable and deep copyable. While this is |
| not enforced it is recommended that only valid python identifiers be used |
| as names. Additionally, the namespace 'stateful_property' is used for |
| storing the valued associated with properties constructed using the |
| property_factory method. |
| """ |
| |
| NO_DEFAULT = object() |
| PICKLE_PROTOCOL = 2 # highest protocol available in python 2.4 |
| |
| |
| def __init__(self): |
| """Initialize the job state.""" |
| self._state = {} |
| self._backing_file = None |
| self._backing_file_initialized = False |
| self._backing_file_lock = None |
| |
| |
| def _lock_backing_file(self): |
| """Acquire a lock on the backing file.""" |
| if self._backing_file: |
| self._backing_file_lock = open(self._backing_file, 'a') |
| fcntl.flock(self._backing_file_lock, fcntl.LOCK_EX) |
| |
| |
| def _unlock_backing_file(self): |
| """Release a lock on the backing file.""" |
| if self._backing_file_lock: |
| fcntl.flock(self._backing_file_lock, fcntl.LOCK_UN) |
| self._backing_file_lock.close() |
| self._backing_file_lock = None |
| |
| |
| def read_from_file(self, file_path, merge=True): |
| """Read in any state from the file at file_path. |
| |
| When merge=True, any state specified only in-memory will be preserved. |
| Any state specified on-disk will be set in-memory, even if an in-memory |
| setting already exists. |
| |
| @param file_path: The path where the state should be read from. It must |
| exist but it can be empty. |
| @param merge: If true, merge the on-disk state with the in-memory |
| state. If false, replace the in-memory state with the on-disk |
| state. |
| |
| @warning: This method is intentionally concurrency-unsafe. It makes no |
| attempt to control concurrent access to the file at file_path. |
| """ |
| |
| # we can assume that the file exists |
| if os.path.getsize(file_path) == 0: |
| on_disk_state = {} |
| else: |
| on_disk_state = pickle.load(open(file_path)) |
| |
| if merge: |
| # merge the on-disk state with the in-memory state |
| for namespace, namespace_dict in on_disk_state.iteritems(): |
| in_memory_namespace = self._state.setdefault(namespace, {}) |
| for name, value in namespace_dict.iteritems(): |
| if name in in_memory_namespace: |
| if in_memory_namespace[name] != value: |
| logging.info('Persistent value of %s.%s from %s ' |
| 'overridding existing in-memory ' |
| 'value', namespace, name, file_path) |
| in_memory_namespace[name] = value |
| else: |
| logging.debug('Value of %s.%s is unchanged, ' |
| 'skipping import', namespace, name) |
| else: |
| logging.debug('Importing %s.%s from state file %s', |
| namespace, name, file_path) |
| in_memory_namespace[name] = value |
| else: |
| # just replace the in-memory state with the on-disk state |
| self._state = on_disk_state |
| |
| # lock the backing file before we refresh it |
| with_backing_lock(self.__class__._write_to_backing_file)(self) |
| |
| |
| def write_to_file(self, file_path): |
| """Write out the current state to the given path. |
| |
| @param file_path: The path where the state should be written out to. |
| Must be writable. |
| |
| @warning: This method is intentionally concurrency-unsafe. It makes no |
| attempt to control concurrent access to the file at file_path. |
| """ |
| outfile = open(file_path, 'w') |
| try: |
| pickle.dump(self._state, outfile, self.PICKLE_PROTOCOL) |
| finally: |
| outfile.close() |
| |
| |
| def _read_from_backing_file(self): |
| """Refresh the current state from the backing file. |
| |
| If the backing file has never been read before (indicated by checking |
| self._backing_file_initialized) it will merge the file with the |
| in-memory state, rather than overwriting it. |
| """ |
| if self._backing_file: |
| merge_backing_file = not self._backing_file_initialized |
| self.read_from_file(self._backing_file, merge=merge_backing_file) |
| self._backing_file_initialized = True |
| |
| |
| def _write_to_backing_file(self): |
| """Flush the current state to the backing file.""" |
| if self._backing_file: |
| self.write_to_file(self._backing_file) |
| |
| |
| @with_backing_file |
| def _synchronize_backing_file(self): |
| """Synchronizes the contents of the in-memory and on-disk state.""" |
| # state is implicitly synchronized in _with_backing_file methods |
| pass |
| |
| |
| def set_backing_file(self, file_path): |
| """Change the path used as the backing file for the persistent state. |
| |
| When a new backing file is specified if a file already exists then |
| its contents will be added into the current state, with conflicts |
| between the file and memory being resolved in favor of the file |
| contents. The file will then be kept in sync with the (combined) |
| in-memory state. The syncing can be disabled by setting this to None. |
| |
| @param file_path: A path on the filesystem that can be read from and |
| written to, or None to turn off the backing store. |
| """ |
| self._synchronize_backing_file() |
| self._backing_file = file_path |
| self._backing_file_initialized = False |
| self._synchronize_backing_file() |
| |
| |
| @with_backing_file |
| def get(self, namespace, name, default=NO_DEFAULT): |
| """Returns the value associated with a particular name. |
| |
| @param namespace: The namespace that the property should be stored in. |
| @param name: The name the value was saved with. |
| @param default: A default value to return if no state is currently |
| associated with var. |
| |
| @return: A deep copy of the value associated with name. Note that this |
| explicitly returns a deep copy to avoid problems with mutable |
| values; mutations are not persisted or shared. |
| @raise KeyError: raised when no state is associated with var and a |
| default value is not provided. |
| """ |
| if self.has(namespace, name): |
| return copy.deepcopy(self._state[namespace][name]) |
| elif default is self.NO_DEFAULT: |
| raise KeyError('No key %s in namespace %s' % (name, namespace)) |
| else: |
| return default |
| |
| |
| @with_backing_file |
| def set(self, namespace, name, value): |
| """Saves the value given with the provided name. |
| |
| @param namespace: The namespace that the property should be stored in. |
| @param name: The name the value should be saved with. |
| @param value: The value to save. |
| """ |
| namespace_dict = self._state.setdefault(namespace, {}) |
| namespace_dict[name] = copy.deepcopy(value) |
| logging.debug('Persistent state %s.%s now set to %r', namespace, |
| name, value) |
| |
| |
| @with_backing_file |
| def has(self, namespace, name): |
| """Return a boolean indicating if namespace.name is defined. |
| |
| @param namespace: The namespace to check for a definition. |
| @param name: The name to check for a definition. |
| |
| @return: True if the given name is defined in the given namespace and |
| False otherwise. |
| """ |
| return namespace in self._state and name in self._state[namespace] |
| |
| |
| @with_backing_file |
| def discard(self, namespace, name): |
| """If namespace.name is a defined value, deletes it. |
| |
| @param namespace: The namespace that the property is stored in. |
| @param name: The name the value is saved with. |
| """ |
| if self.has(namespace, name): |
| del self._state[namespace][name] |
| if len(self._state[namespace]) == 0: |
| del self._state[namespace] |
| logging.debug('Persistent state %s.%s deleted', namespace, name) |
| else: |
| logging.debug( |
| 'Persistent state %s.%s not defined so nothing is discarded', |
| namespace, name) |
| |
| |
| @with_backing_file |
| def discard_namespace(self, namespace): |
| """Delete all defined namespace.* names. |
| |
| @param namespace: The namespace to be cleared. |
| """ |
| if namespace in self._state: |
| del self._state[namespace] |
| logging.debug('Persistent state %s.* deleted', namespace) |
| |
| |
| @staticmethod |
| def property_factory(state_attribute, property_attribute, default, |
| namespace='global_properties'): |
| """ |
| Create a property object for an attribute using self.get and self.set. |
| |
| @param state_attribute: A string with the name of the attribute on |
| job that contains the job_state instance. |
| @param property_attribute: A string with the name of the attribute |
| this property is exposed as. |
| @param default: A default value that should be used for this property |
| if it is not set. |
| @param namespace: The namespace to store the attribute value in. |
| |
| @return: A read-write property object that performs self.get calls |
| to read the value and self.set calls to set it. |
| """ |
| def getter(job): |
| state = getattr(job, state_attribute) |
| return state.get(namespace, property_attribute, default) |
| def setter(job, value): |
| state = getattr(job, state_attribute) |
| state.set(namespace, property_attribute, value) |
| return property(getter, setter) |
| |
| |
| class status_log_entry(object): |
| """Represents a single status log entry.""" |
| |
| RENDERED_NONE_VALUE = '----' |
| TIMESTAMP_FIELD = 'timestamp' |
| LOCALTIME_FIELD = 'localtime' |
| |
| # non-space whitespace is forbidden in any fields |
| BAD_CHAR_REGEX = re.compile(r'[\t\n\r\v\f]') |
| |
| def __init__(self, status_code, subdir, operation, message, fields, |
| timestamp=None): |
| """Construct a status.log entry. |
| |
| @param status_code: A message status code. Must match the codes |
| accepted by autotest_lib.common_lib.log.is_valid_status. |
| @param subdir: A valid job subdirectory, or None. |
| @param operation: Description of the operation, or None. |
| @param message: A printable string describing event to be recorded. |
| @param fields: A dictionary of arbitrary alphanumeric key=value pairs |
| to be included in the log, or None. |
| @param timestamp: An optional integer timestamp, in the same format |
| as a time.time() timestamp. If unspecified, the current time is |
| used. |
| |
| @raise ValueError: if any of the parameters are invalid |
| """ |
| |
| if not log.is_valid_status(status_code): |
| raise ValueError('status code %r is not valid' % status_code) |
| self.status_code = status_code |
| |
| if subdir and self.BAD_CHAR_REGEX.search(subdir): |
| raise ValueError('Invalid character in subdir string') |
| self.subdir = subdir |
| |
| if operation and self.BAD_CHAR_REGEX.search(operation): |
| raise ValueError('Invalid character in operation string') |
| self.operation = operation |
| |
| # break the message line into a single-line message that goes into the |
| # database, and a block of additional lines that goes into the status |
| # log but will never be parsed |
| message_lines = message.split('\n') |
| self.message = message_lines[0].replace('\t', ' ' * 8) |
| self.extra_message_lines = message_lines[1:] |
| if self.BAD_CHAR_REGEX.search(self.message): |
| raise ValueError('Invalid character in message %r' % self.message) |
| |
| if not fields: |
| self.fields = {} |
| else: |
| self.fields = fields.copy() |
| for key, value in self.fields.iteritems(): |
| if type(value) is int: |
| value = str(value) |
| if self.BAD_CHAR_REGEX.search(key + value): |
| raise ValueError('Invalid character in %r=%r field' |
| % (key, value)) |
| |
| # build up the timestamp |
| if timestamp is None: |
| timestamp = int(time.time()) |
| self.fields[self.TIMESTAMP_FIELD] = str(timestamp) |
| self.fields[self.LOCALTIME_FIELD] = time.strftime( |
| '%b %d %H:%M:%S', time.localtime(timestamp)) |
| |
| |
| def is_start(self): |
| """Indicates if this status log is the start of a new nested block. |
| |
| @return: A boolean indicating if this entry starts a new nested block. |
| """ |
| return self.status_code == 'START' |
| |
| |
| def is_end(self): |
| """Indicates if this status log is the end of a nested block. |
| |
| @return: A boolean indicating if this entry ends a nested block. |
| """ |
| return self.status_code.startswith('END ') |
| |
| |
| def render(self): |
| """Render the status log entry into a text string. |
| |
| @return: A text string suitable for writing into a status log file. |
| """ |
| # combine all the log line data into a tab-delimited string |
| subdir = self.subdir or self.RENDERED_NONE_VALUE |
| operation = self.operation or self.RENDERED_NONE_VALUE |
| extra_fields = ['%s=%s' % field for field in self.fields.iteritems()] |
| line_items = [self.status_code, subdir, operation] |
| line_items += extra_fields + [self.message] |
| first_line = '\t'.join(line_items) |
| |
| # append the extra unparsable lines, two-space indented |
| all_lines = [first_line] |
| all_lines += [' ' + line for line in self.extra_message_lines] |
| return '\n'.join(all_lines) |
| |
| |
| @classmethod |
| def parse(cls, line): |
| """Parse a status log entry from a text string. |
| |
| This method is the inverse of render; it should always be true that |
| parse(entry.render()) produces a new status_log_entry equivalent to |
| entry. |
| |
| @return: A new status_log_entry instance with fields extracted from the |
| given status line. If the line is an extra message line then None |
| is returned. |
| """ |
| # extra message lines are always prepended with two spaces |
| if line.startswith(' '): |
| return None |
| |
| line = line.lstrip('\t') # ignore indentation |
| entry_parts = line.split('\t') |
| if len(entry_parts) < 4: |
| raise ValueError('%r is not a valid status line' % line) |
| status_code, subdir, operation = entry_parts[:3] |
| if subdir == cls.RENDERED_NONE_VALUE: |
| subdir = None |
| if operation == cls.RENDERED_NONE_VALUE: |
| operation = None |
| message = entry_parts[-1] |
| fields = dict(part.split('=', 1) for part in entry_parts[3:-1]) |
| if cls.TIMESTAMP_FIELD in fields: |
| timestamp = int(fields[cls.TIMESTAMP_FIELD]) |
| else: |
| timestamp = None |
| return cls(status_code, subdir, operation, message, fields, timestamp) |
| |
| |
| class status_indenter(object): |
| """Abstract interface that a status log indenter should use.""" |
| |
| @property |
| def indent(self): |
| raise NotImplementedError |
| |
| |
| def increment(self): |
| """Increase indentation by one level.""" |
| raise NotImplementedError |
| |
| |
| def decrement(self): |
| """Decrease indentation by one level.""" |
| |
| |
| class status_logger(object): |
| """Represents a status log file. Responsible for translating messages |
| into on-disk status log lines. |
| |
| @property global_filename: The filename to write top-level logs to. |
| @property subdir_filename: The filename to write subdir-level logs to. |
| """ |
| def __init__(self, job, indenter, global_filename='status', |
| subdir_filename='status', record_hook=None, |
| tap_writer=None): |
| """Construct a logger instance. |
| |
| @param job: A reference to the job object this is logging for. Only a |
| weak reference to the job is held, to avoid a |
| status_logger <-> job circular reference. |
| @param indenter: A status_indenter instance, for tracking the |
| indentation level. |
| @param global_filename: An optional filename to initialize the |
| self.global_filename attribute. |
| @param subdir_filename: An optional filename to initialize the |
| self.subdir_filename attribute. |
| @param record_hook: An optional function to be called before an entry |
| is logged. The function should expect a single parameter, a |
| copy of the status_log_entry object. |
| @param tap_writer: An instance of the class TAPReport for addionally |
| writing TAP files |
| """ |
| self._jobref = weakref.ref(job) |
| self._indenter = indenter |
| self.global_filename = global_filename |
| self.subdir_filename = subdir_filename |
| self._record_hook = record_hook |
| if tap_writer is None: |
| self._tap_writer = TAPReport(None) |
| else: |
| self._tap_writer = tap_writer |
| |
| |
| def render_entry(self, log_entry): |
| """Render a status_log_entry as it would be written to a log file. |
| |
| @param log_entry: A status_log_entry instance to be rendered. |
| |
| @return: The status log entry, rendered as it would be written to the |
| logs (including indentation). |
| """ |
| if log_entry.is_end(): |
| indent = self._indenter.indent - 1 |
| else: |
| indent = self._indenter.indent |
| return '\t' * indent + log_entry.render().rstrip('\n') |
| |
| |
| def record_entry(self, log_entry, log_in_subdir=True): |
| """Record a status_log_entry into the appropriate status log files. |
| |
| @param log_entry: A status_log_entry instance to be recorded into the |
| status logs. |
| @param log_in_subdir: A boolean that indicates (when true) that subdir |
| logs should be written into the subdirectory status log file. |
| """ |
| # acquire a strong reference for the duration of the method |
| job = self._jobref() |
| if job is None: |
| logging.warning('Something attempted to write a status log entry ' |
| 'after its job terminated, ignoring the attempt.') |
| logging.warning(traceback.format_stack()) |
| return |
| |
| # call the record hook if one was given |
| if self._record_hook: |
| self._record_hook(log_entry) |
| |
| # figure out where we need to log to |
| log_files = [os.path.join(job.resultdir, self.global_filename)] |
| if log_in_subdir and log_entry.subdir: |
| log_files.append(os.path.join(job.resultdir, log_entry.subdir, |
| self.subdir_filename)) |
| |
| # write out to entry to the log files |
| log_text = self.render_entry(log_entry) |
| for log_file in log_files: |
| fileobj = open(log_file, 'a') |
| try: |
| print >> fileobj, log_text |
| finally: |
| fileobj.close() |
| |
| # write to TAPRecord instance |
| if log_entry.is_end() and self._tap_writer.do_tap_report: |
| self._tap_writer.record(log_entry, self._indenter.indent, log_files) |
| |
| # adjust the indentation if this was a START or END entry |
| if log_entry.is_start(): |
| self._indenter.increment() |
| elif log_entry.is_end(): |
| self._indenter.decrement() |
| |
| |
| class TAPReport(object): |
| """ |
| Deal with TAP reporting for the Autotest client. |
| """ |
| |
| job_statuses = { |
| "TEST_NA": False, |
| "ABORT": False, |
| "ERROR": False, |
| "FAIL": False, |
| "WARN": False, |
| "GOOD": True, |
| "START": True, |
| "END GOOD": True, |
| "ALERT": False, |
| "RUNNING": False, |
| "NOSTATUS": False |
| } |
| |
| |
| def __init__(self, enable, resultdir=None, global_filename='status'): |
| """ |
| @param enable: Set self.do_tap_report to trigger TAP reporting. |
| @param resultdir: Path where the TAP report files will be written. |
| @param global_filename: File name of the status files .tap extensions |
| will be appended. |
| """ |
| self.do_tap_report = enable |
| if resultdir is not None: |
| self.resultdir = os.path.abspath(resultdir) |
| self._reports_container = {} |
| self._keyval_container = {} # {'path1': [entries],} |
| self.global_filename = global_filename |
| |
| |
| @classmethod |
| def tap_ok(self, success, counter, message): |
| """ |
| return a TAP message string. |
| |
| @param success: True for positive message string. |
| @param counter: number of TAP line in plan. |
| @param message: additional message to report in TAP line. |
| """ |
| if success: |
| message = "ok %s - %s" % (counter, message) |
| else: |
| message = "not ok %s - %s" % (counter, message) |
| return message |
| |
| |
| def record(self, log_entry, indent, log_files): |
| """ |
| Append a job-level status event to self._reports_container. All |
| events will be written to TAP log files at the end of the test run. |
| Otherwise, it's impossilble to determine the TAP plan. |
| |
| @param log_entry: A string status code describing the type of status |
| entry being recorded. It must pass log.is_valid_status to be |
| considered valid. |
| @param indent: Level of the log_entry to determine the operation if |
| log_entry.operation is not given. |
| @param log_files: List of full path of files the TAP report will be |
| written to at the end of the test. |
| """ |
| for log_file in log_files: |
| log_file_path = os.path.dirname(log_file) |
| key = log_file_path.split(self.resultdir, 1)[1].strip(os.sep) |
| if not key: |
| key = 'root' |
| |
| if not self._reports_container.has_key(key): |
| self._reports_container[key] = [] |
| |
| if log_entry.operation: |
| operation = log_entry.operation |
| elif indent == 1: |
| operation = "job" |
| else: |
| operation = "unknown" |
| entry = self.tap_ok( |
| self.job_statuses.get(log_entry.status_code, False), |
| len(self._reports_container[key]) + 1, operation + "\n" |
| ) |
| self._reports_container[key].append(entry) |
| |
| |
| def record_keyval(self, path, dictionary, type_tag=None): |
| """ |
| Append a key-value pairs of dictionary to self._keyval_container in |
| TAP format. Once finished write out the keyval.tap file to the file |
| system. |
| |
| If type_tag is None, then the key must be composed of alphanumeric |
| characters (or dashes + underscores). However, if type-tag is not |
| null then the keys must also have "{type_tag}" as a suffix. At |
| the moment the only valid values of type_tag are "attr" and "perf". |
| |
| @param path: The full path of the keyval.tap file to be created |
| @param dictionary: The keys and values. |
| @param type_tag: The type of the values |
| """ |
| self._keyval_container.setdefault(path, [0, []]) |
| self._keyval_container[path][0] += 1 |
| |
| if type_tag is None: |
| key_regex = re.compile(r'^[-\.\w]+$') |
| else: |
| if type_tag not in ('attr', 'perf'): |
| raise ValueError('Invalid type tag: %s' % type_tag) |
| escaped_tag = re.escape(type_tag) |
| key_regex = re.compile(r'^[-\.\w]+\{%s\}$' % escaped_tag) |
| self._keyval_container[path][1].extend([ |
| self.tap_ok(True, self._keyval_container[path][0], "results"), |
| "\n ---\n", |
| ]) |
| try: |
| for key in sorted(dictionary.keys()): |
| if not key_regex.search(key): |
| raise ValueError('Invalid key: %s' % key) |
| self._keyval_container[path][1].append( |
| ' %s: %s\n' % (key.replace('{', '_').rstrip('}'), |
| dictionary[key]) |
| ) |
| finally: |
| self._keyval_container[path][1].append(" ...\n") |
| self._write_keyval() |
| |
| |
| def _write_reports(self): |
| """ |
| Write TAP reports to file. |
| """ |
| for key in self._reports_container.keys(): |
| if key == 'root': |
| sub_dir = '' |
| else: |
| sub_dir = key |
| tap_fh = open(os.sep.join( |
| [self.resultdir, sub_dir, self.global_filename] |
| ) + ".tap", 'w') |
| tap_fh.write('1..' + str(len(self._reports_container[key])) + '\n') |
| tap_fh.writelines(self._reports_container[key]) |
| tap_fh.close() |
| |
| |
| def _write_keyval(self): |
| """ |
| Write the self._keyval_container key values to a file. |
| """ |
| for path in self._keyval_container.keys(): |
| tap_fh = open(path + ".tap", 'w') |
| tap_fh.write('1..' + str(self._keyval_container[path][0]) + '\n') |
| tap_fh.writelines(self._keyval_container[path][1]) |
| tap_fh.close() |
| |
| |
| def write(self): |
| """ |
| Write the TAP reports to files. |
| """ |
| self._write_reports() |
| |
| |
| def _write_tap_archive(self): |
| """ |
| Write a tar archive containing all the TAP files and |
| a meta.yml containing the file names. |
| """ |
| os.chdir(self.resultdir) |
| tap_files = [] |
| for rel_path, d, files in os.walk('.'): |
| tap_files.extend(["/".join( |
| [rel_path, f]) for f in files if f.endswith('.tap')]) |
| meta_yaml = open('meta.yml', 'w') |
| meta_yaml.write('file_order:\n') |
| tap_tar = tarfile.open(self.resultdir + '/tap.tar.gz', 'w:gz') |
| for f in tap_files: |
| meta_yaml.write(" - " + f.lstrip('./') + "\n") |
| tap_tar.add(f) |
| meta_yaml.close() |
| tap_tar.add('meta.yml') |
| tap_tar.close() |
| |
| |
| class base_job(object): |
| """An abstract base class for the various autotest job classes. |
| |
| @property autodir: The top level autotest directory. |
| @property clientdir: The autotest client directory. |
| @property serverdir: The autotest server directory. [OPTIONAL] |
| @property resultdir: The directory where results should be written out. |
| [WRITABLE] |
| |
| @property pkgdir: The job packages directory. [WRITABLE] |
| @property tmpdir: The job temporary directory. [WRITABLE] |
| @property testdir: The job test directory. [WRITABLE] |
| @property site_testdir: The job site test directory. [WRITABLE] |
| |
| @property bindir: The client bin/ directory. |
| @property profdir: The client profilers/ directory. |
| @property toolsdir: The client tools/ directory. |
| |
| @property conmuxdir: The conmux directory. [OPTIONAL] |
| |
| @property control: A path to the control file to be executed. [OPTIONAL] |
| @property hosts: A set of all live Host objects currently in use by the |
| job. Code running in the context of a local client can safely assume |
| that this set contains only a single entry. |
| @property machines: A list of the machine names associated with the job. |
| @property user: The user executing the job. |
| @property tag: A tag identifying the job. Often used by the scheduler to |
| give a name of the form NUMBER-USERNAME/HOSTNAME. |
| @property test_retry: The number of times to retry a test if the test did |
| not complete successfully. |
| @property args: A list of addtional miscellaneous command-line arguments |
| provided when starting the job. |
| |
| @property last_boot_tag: The label of the kernel from the last reboot. |
| [OPTIONAL,PERSISTENT] |
| @property automatic_test_tag: A string which, if set, will be automatically |
| added to the test name when running tests. |
| |
| @property default_profile_only: A boolean indicating the default value of |
| profile_only used by test.execute. [PERSISTENT] |
| @property drop_caches: A boolean indicating if caches should be dropped |
| before each test is executed. |
| @property drop_caches_between_iterations: A boolean indicating if caches |
| should be dropped before each test iteration is executed. |
| @property run_test_cleanup: A boolean indicating if test.cleanup should be |
| run by default after a test completes, if the run_cleanup argument is |
| not specified. [PERSISTENT] |
| |
| @property num_tests_run: The number of tests run during the job. [OPTIONAL] |
| @property num_tests_failed: The number of tests failed during the job. |
| [OPTIONAL] |
| |
| @property bootloader: An instance of the boottool class. May not be |
| available on job instances where access to the bootloader is not |
| available (e.g. on the server running a server job). [OPTIONAL] |
| @property harness: An instance of the client test harness. Only available |
| in contexts where client test execution happens. [OPTIONAL] |
| @property logging: An instance of the logging manager associated with the |
| job. |
| @property profilers: An instance of the profiler manager associated with |
| the job. |
| @property sysinfo: An instance of the sysinfo object. Only available in |
| contexts where it's possible to collect sysinfo. |
| @property warning_manager: A class for managing which types of WARN |
| messages should be logged and which should be supressed. [OPTIONAL] |
| @property warning_loggers: A set of readable streams that will be monitored |
| for WARN messages to be logged. [OPTIONAL] |
| |
| Abstract methods: |
| _find_base_directories [CLASSMETHOD] |
| Returns the location of autodir, clientdir and serverdir |
| |
| _find_resultdir |
| Returns the location of resultdir. Gets a copy of any parameters |
| passed into base_job.__init__. Can return None to indicate that |
| no resultdir is to be used. |
| |
| _get_status_logger |
| Returns a status_logger instance for recording job status logs. |
| """ |
| |
| # capture the dependency on several helper classes with factories |
| _job_directory = job_directory |
| _job_state = job_state |
| |
| |
| # all the job directory attributes |
| autodir = _job_directory.property_factory('autodir') |
| clientdir = _job_directory.property_factory('clientdir') |
| serverdir = _job_directory.property_factory('serverdir') |
| resultdir = _job_directory.property_factory('resultdir') |
| pkgdir = _job_directory.property_factory('pkgdir') |
| tmpdir = _job_directory.property_factory('tmpdir') |
| testdir = _job_directory.property_factory('testdir') |
| site_testdir = _job_directory.property_factory('site_testdir') |
| bindir = _job_directory.property_factory('bindir') |
| profdir = _job_directory.property_factory('profdir') |
| toolsdir = _job_directory.property_factory('toolsdir') |
| conmuxdir = _job_directory.property_factory('conmuxdir') |
| |
| |
| # all the generic persistent properties |
| tag = _job_state.property_factory('_state', 'tag', '') |
| test_retry = _job_state.property_factory('_state', 'test_retry', 0) |
| default_profile_only = _job_state.property_factory( |
| '_state', 'default_profile_only', False) |
| run_test_cleanup = _job_state.property_factory( |
| '_state', 'run_test_cleanup', True) |
| last_boot_tag = _job_state.property_factory( |
| '_state', 'last_boot_tag', None) |
| automatic_test_tag = _job_state.property_factory( |
| '_state', 'automatic_test_tag', None) |
| |
| # the use_sequence_number property |
| _sequence_number = _job_state.property_factory( |
| '_state', '_sequence_number', None) |
| def _get_use_sequence_number(self): |
| return bool(self._sequence_number) |
| def _set_use_sequence_number(self, value): |
| if value: |
| self._sequence_number = 1 |
| else: |
| self._sequence_number = None |
| use_sequence_number = property(_get_use_sequence_number, |
| _set_use_sequence_number) |
| |
| # parent job id is passed in from autoserv command line. It's only used in |
| # server job. The property is added here for unittest |
| # (base_job_unittest.py) to be consistent on validating public properties of |
| # a base_job object. |
| parent_job_id = None |
| |
| def __init__(self, *args, **dargs): |
| # initialize the base directories, all others are relative to these |
| autodir, clientdir, serverdir = self._find_base_directories() |
| self._autodir = self._job_directory(autodir) |
| self._clientdir = self._job_directory(clientdir) |
| # TODO(scottz): crosbug.com/38259, needed to pass unittests for now. |
| self.label = None |
| if serverdir: |
| self._serverdir = self._job_directory(serverdir) |
| else: |
| self._serverdir = None |
| |
| # initialize all the other directories relative to the base ones |
| self._initialize_dir_properties() |
| self._resultdir = self._job_directory( |
| self._find_resultdir(*args, **dargs), True) |
| self._execution_contexts = [] |
| |
| # initialize all the job state |
| self._state = self._job_state() |
| |
| # initialize tap reporting |
| if dargs.has_key('options'): |
| self._tap = self._tap_init(dargs['options'].tap_report) |
| else: |
| self._tap = self._tap_init(False) |
| |
| @classmethod |
| def _find_base_directories(cls): |
| raise NotImplementedError() |
| |
| |
| def _initialize_dir_properties(self): |
| """ |
| Initializes all the secondary self.*dir properties. Requires autodir, |
| clientdir and serverdir to already be initialized. |
| """ |
| # create some stubs for use as shortcuts |
| def readonly_dir(*args): |
| return self._job_directory(os.path.join(*args)) |
| def readwrite_dir(*args): |
| return self._job_directory(os.path.join(*args), True) |
| |
| # various client-specific directories |
| self._bindir = readonly_dir(self.clientdir, 'bin') |
| self._profdir = readonly_dir(self.clientdir, 'profilers') |
| self._pkgdir = readwrite_dir(self.clientdir, 'packages') |
| self._toolsdir = readonly_dir(self.clientdir, 'tools') |
| |
| # directories which are in serverdir on a server, clientdir on a client |
| # tmp tests, and site_tests need to be read_write for client, but only |
| # read for server. |
| if self.serverdir: |
| root = self.serverdir |
| r_or_rw_dir = readonly_dir |
| else: |
| root = self.clientdir |
| r_or_rw_dir = readwrite_dir |
| self._testdir = r_or_rw_dir(root, 'tests') |
| self._site_testdir = r_or_rw_dir(root, 'site_tests') |
| |
| # various server-specific directories |
| if self.serverdir: |
| self._tmpdir = readwrite_dir(tempfile.gettempdir()) |
| self._conmuxdir = readonly_dir(self.autodir, 'conmux') |
| else: |
| self._tmpdir = readwrite_dir(root, 'tmp') |
| self._conmuxdir = None |
| |
| |
| def _find_resultdir(self, *args, **dargs): |
| raise NotImplementedError() |
| |
| |
| def push_execution_context(self, resultdir): |
| """ |
| Save off the current context of the job and change to the given one. |
| |
| In practice method just changes the resultdir, but it may become more |
| extensive in the future. The expected use case is for when a child |
| job needs to be executed in some sort of nested context (for example |
| the way parallel_simple does). The original context can be restored |
| with a pop_execution_context call. |
| |
| @param resultdir: The new resultdir, relative to the current one. |
| """ |
| new_dir = self._job_directory( |
| os.path.join(self.resultdir, resultdir), True) |
| self._execution_contexts.append(self._resultdir) |
| self._resultdir = new_dir |
| |
| |
| def pop_execution_context(self): |
| """ |
| Reverse the effects of the previous push_execution_context call. |
| |
| @raise IndexError: raised when the stack of contexts is empty. |
| """ |
| if not self._execution_contexts: |
| raise IndexError('No old execution context to restore') |
| self._resultdir = self._execution_contexts.pop() |
| |
| |
| def get_state(self, name, default=_job_state.NO_DEFAULT): |
| """Returns the value associated with a particular name. |
| |
| @param name: The name the value was saved with. |
| @param default: A default value to return if no state is currently |
| associated with var. |
| |
| @return: A deep copy of the value associated with name. Note that this |
| explicitly returns a deep copy to avoid problems with mutable |
| values; mutations are not persisted or shared. |
| @raise KeyError: raised when no state is associated with var and a |
| default value is not provided. |
| """ |
| try: |
| return self._state.get('public', name, default=default) |
| except KeyError: |
| raise KeyError(name) |
| |
| |
| def set_state(self, name, value): |
| """Saves the value given with the provided name. |
| |
| @param name: The name the value should be saved with. |
| @param value: The value to save. |
| """ |
| self._state.set('public', name, value) |
| |
| |
| def _build_tagged_test_name(self, testname, dargs): |
| """Builds the fully tagged testname and subdirectory for job.run_test. |
| |
| @param testname: The base name of the test |
| @param dargs: The ** arguments passed to run_test. And arguments |
| consumed by this method will be removed from the dictionary. |
| |
| @return: A 3-tuple of the full name of the test, the subdirectory it |
| should be stored in, and the full tag of the subdir. |
| """ |
| tag_parts = [] |
| |
| # build up the parts of the tag used for the test name |
| master_testpath = dargs.get('master_testpath', "") |
| base_tag = dargs.pop('tag', None) |
| if base_tag: |
| tag_parts.append(str(base_tag)) |
| if self.use_sequence_number: |
| tag_parts.append('_%02d_' % self._sequence_number) |
| self._sequence_number += 1 |
| if self.automatic_test_tag: |
| tag_parts.append(self.automatic_test_tag) |
| full_testname = '.'.join([testname] + tag_parts) |
| |
| # build up the subdir and tag as well |
| subdir_tag = dargs.pop('subdir_tag', None) |
| if subdir_tag: |
| tag_parts.append(subdir_tag) |
| subdir = '.'.join([testname] + tag_parts) |
| subdir = os.path.join(master_testpath, subdir) |
| tag = '.'.join(tag_parts) |
| |
| return full_testname, subdir, tag |
| |
| |
| def _make_test_outputdir(self, subdir): |
| """Creates an output directory for a test to run it. |
| |
| @param subdir: The subdirectory of the test. Generally computed by |
| _build_tagged_test_name. |
| |
| @return: A job_directory instance corresponding to the outputdir of |
| the test. |
| @raise TestError: If the output directory is invalid. |
| """ |
| # explicitly check that this subdirectory is new |
| path = os.path.join(self.resultdir, subdir) |
| if os.path.exists(path): |
| msg = ('%s already exists; multiple tests cannot run with the ' |
| 'same subdirectory' % subdir) |
| raise error.TestError(msg) |
| |
| # create the outputdir and raise a TestError if it isn't valid |
| try: |
| outputdir = self._job_directory(path, True) |
| return outputdir |
| except self._job_directory.JobDirectoryException, e: |
| logging.exception('%s directory creation failed with %s', |
| subdir, e) |
| raise error.TestError('%s directory creation failed' % subdir) |
| |
| def _tap_init(self, enable): |
| """Initialize TAP reporting |
| """ |
| return TAPReport(enable, resultdir=self.resultdir) |
| |
| |
| def record(self, status_code, subdir, operation, status='', |
| optional_fields=None): |
| """Record a job-level status event. |
| |
| Logs an event noteworthy to the Autotest job as a whole. Messages will |
| be written into a global status log file, as well as a subdir-local |
| status log file (if subdir is specified). |
| |
| @param status_code: A string status code describing the type of status |
| entry being recorded. It must pass log.is_valid_status to be |
| considered valid. |
| @param subdir: A specific results subdirectory this also applies to, or |
| None. If not None the subdirectory must exist. |
| @param operation: A string describing the operation that was run. |
| @param status: An optional human-readable message describing the status |
| entry, for example an error message or "completed successfully". |
| @param optional_fields: An optional dictionary of addtional named fields |
| to be included with the status message. Every time timestamp and |
| localtime entries are generated with the current time and added |
| to this dictionary. |
| """ |
| entry = status_log_entry(status_code, subdir, operation, status, |
| optional_fields) |
| self.record_entry(entry) |
| |
| |
| def record_entry(self, entry, log_in_subdir=True): |
| """Record a job-level status event, using a status_log_entry. |
| |
| This is the same as self.record but using an existing status log |
| entry object rather than constructing one for you. |
| |
| @param entry: A status_log_entry object |
| @param log_in_subdir: A boolean that indicates (when true) that subdir |
| logs should be written into the subdirectory status log file. |
| """ |
| self._get_status_logger().record_entry(entry, log_in_subdir) |