xref: /external/autotest/client/common_lib/base_job.py

import os, copy, logging, errno, fcntl, time, re, weakref, traceback
import cPickle as pickle

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 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):
        """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.
        """
        self._jobref = weakref.ref(job)
        self._indenter = indenter
        self.global_filename = global_filename
        self.subdir_filename = subdir_filename
        self._record_hook = record_hook


    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()

        # 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 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 configdir: The client config/ 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 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')
    configdir = _job_directory.property_factory('configdir')
    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', '')
    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)


    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)
        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()


    @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._configdir = readonly_dir(self.clientdir, 'config')
        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
        if self.serverdir:
            root = self.serverdir
        else:
            root = self.clientdir
        self._tmpdir = readwrite_dir(root, 'tmp')
        self._testdir = readwrite_dir(root, 'tests')
        self._site_testdir = readwrite_dir(root, 'site_tests')

        # various server-specific directories
        if self.serverdir:
            self._conmuxdir = readonly_dir(self.autodir, 'conmux')
        else:
            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
        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)
        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 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)