ceph/teuthology/misc.py

from cStringIO import StringIO

import os
import logging
import configobj
import getpass
import paramiko
import socket
import time
import urllib2
import urlparse
import yaml

from .orchestra import run

log = logging.getLogger(__name__)

def get_ceph_binary_url(branch=None, tag=None, sha1=None, flavor=None):
    if flavor is None:
        flavor = ''
    else:
        # TODO hardcoding amd64 here for simplicity; clients will try
        # to fetch the tarball matching their arch, non-x86_64 just
        # won't find anything and the test will fail. trying to
        # support cross-arch clusters is messy because nothing
        # guarantees the same sha1 of "master" has been built for all
        # of them. hoping for yagni.
        flavor = '-{flavor}-amd64'.format(flavor=flavor)
    BASE = 'http://ceph.newdream.net/gitbuilder{flavor}/output/'.format(flavor=flavor)

    if sha1 is not None:
        assert branch is None, "cannot set both sha1 and branch"
        assert tag is None, "cannot set both sha1 and tag"
    else:
        # gitbuilder uses remote-style ref names for branches, mangled to
        # have underscores instead of slashes; e.g. origin_master
        if tag is not None:
            ref = tag
            assert branch is None, "cannot set both branch and tag"
        else:
            if branch is None:
                branch = 'master'
            ref = branch

        sha1_url = urlparse.urljoin(BASE, 'ref/{ref}/sha1'.format(ref=ref))
        log.debug('Translating ref to sha1 using url %s', sha1_url)
        sha1_fp = urllib2.urlopen(sha1_url)
        sha1 = sha1_fp.read().rstrip('\n')
        sha1_fp.close()

    log.debug('Using ceph sha1 %s', sha1)
    bindir_url = urlparse.urljoin(BASE, 'sha1/{sha1}/'.format(sha1=sha1))
    return (sha1, bindir_url)

def feed_many_stdins(fp, processes):
    while True:
        data = fp.read(8192)
        if not data:
            break
        for proc in processes:
            proc.stdin.write(data)

def feed_many_stdins_and_close(fp, processes):
    feed_many_stdins(fp, processes)
    for proc in processes:
        proc.stdin.close()

def get_mons(roles, ips):
    mons = {}
    mon_id = 0
    for idx, roles in enumerate(roles):
        for role in roles:
            if not role.startswith('mon.'):
                continue
            addr = '{ip}:{port}'.format(
                ip=ips[idx],
                port=6789+mon_id,
                )
            mon_id += 1
            mons[role] = addr
    assert mons
    return mons

def generate_caps(type_):
    defaults = dict(
        osd=dict(
            mon='allow *',
            osd='allow *',
            ),
        mds=dict(
            mon='allow *',
            osd='allow *',
            mds='allow',
            ),
        client=dict(
            mon='allow rw',
            osd='allow rwx pool=data,rbd',
            mds='allow',
            ),
        )
    for subsystem, capability in defaults[type_].items():
        yield '--cap'
        yield subsystem
        yield capability

def skeleton_config(roles, ips):
    """
    Returns a ConfigObj that's prefilled with a skeleton config.

    Use conf[section][key]=value or conf.merge to change it.

    Use conf.write to write it out, override .filename first if you want.
    """
    path = os.path.join(os.path.dirname(__file__), 'ceph.conf')
    conf = configobj.ConfigObj(path, file_error=True)
    mons = get_mons(roles=roles, ips=ips)
    for role, addr in mons.iteritems():
        conf.setdefault(role, {})
        conf[role]['mon addr'] = addr
    # set up standby mds's
    for roles in roles:
        for role in roles:
            if role.startswith('mds.') and role.endswith('-s'):
                conf.setdefault(role, {})
                conf[role]['mds standby for name'] = role[:-2]
    return conf

def roles_of_type(roles_for_host, type_):
    prefix = '{type}.'.format(type=type_)
    for name in roles_for_host:
        if not name.startswith(prefix):
            continue
        id_ = name[len(prefix):]
        yield id_

def all_roles_of_type(cluster, type_):
    prefix = '{type}.'.format(type=type_)
    for _, roles_for_host in cluster.remotes.iteritems():
        for name in roles_for_host:
            if not name.startswith(prefix):
                continue
            id_ = name[len(prefix):]
            yield id_

def is_type(type_):
    """
    Returns a matcher function for whether role is of type given.
    """
    prefix = '{type}.'.format(type=type_)
    def _is_type(role):
        return role.startswith(prefix)
    return _is_type

def num_instances_of_type(cluster, type_):
    remotes_and_roles = cluster.remotes.items()
    roles = [roles for (remote, roles) in remotes_and_roles]
    prefix = '{type}.'.format(type=type_)
    num = sum(sum(1 for role in hostroles if role.startswith(prefix)) for hostroles in roles)
    return num

def create_simple_monmap(remote, conf):
    """
    Writes a simple monmap based on current ceph.conf into <tmpdir>/monmap.

    Assumes ceph_conf is up to date.

    Assumes mon sections are named "mon.*", with the dot.
    """
    def gen_addresses():
        for section, data in conf.iteritems():
            PREFIX = 'mon.'
            if not section.startswith(PREFIX):
                continue
            name = section[len(PREFIX):]
            addr = data['mon addr']
            yield (name, addr)

    addresses = list(gen_addresses())
    assert addresses, "There are no monitors in config!"
    log.debug('Ceph mon addresses: %s', addresses)

    args = [
        '/tmp/cephtest/enable-coredump',
        '/tmp/cephtest/binary/usr/local/bin/ceph-coverage',
        '/tmp/cephtest/archive/coverage',
        '/tmp/cephtest/binary/usr/local/bin/monmaptool',
        '--create',
        '--clobber',
        ]
    for (name, addr) in addresses:
        args.extend(('--add', name, addr))
    args.extend([
            '--print',
            '/tmp/cephtest/monmap',
            ])
    remote.run(
        args=args,
        )

def write_file(remote, path, data):
    remote.run(
        args=[
            'python',
            '-c',
            'import shutil, sys; shutil.copyfileobj(sys.stdin, file(sys.argv[1], "wb"))',
            path,
            ],
        stdin=data,
        )

def sudo_write_file(remote, path, data):
    remote.run(
        args=[
            'sudo',
            'python',
            '-c',
            'import shutil, sys; shutil.copyfileobj(sys.stdin, file(sys.argv[1], "wb"))',
            path,
            ],
        stdin=data,
        )

def get_file(remote, path):
    """
    Read a file from remote host into memory.
    """
    proc = remote.run(
        args=[
            'cat',
            '--',
            path,
            ],
        stdout=StringIO(),
        )
    data = proc.stdout.getvalue()
    return data

def get_scratch_devices(remote):
    """
    Read the scratch disk list from remote host
    """
    devs = []
    try:
        file_data = get_file(remote, "/scratch_devs")
        devs = file_data.split()
    except:
        devs = ['/dev/sdb']

    retval = []
    for dev in devs:
        try:
            remote.run(
                args=[
                    'stat',
                    dev
                    ]
                )
            retval.append(dev)
        except:
            pass
    return retval

def wait_until_healthy(remote):
    """Wait until a Ceph cluster is healthy."""
    while True:
        r = remote.run(
            args=[
                '/tmp/cephtest/enable-coredump',
                '/tmp/cephtest/binary/usr/local/bin/ceph-coverage',
                '/tmp/cephtest/archive/coverage',
                '/tmp/cephtest/binary/usr/local/bin/ceph',
                '-c', '/tmp/cephtest/ceph.conf',
                'health',
                '--concise',
                ],
            stdout=StringIO(),
            logger=log.getChild('health'),
            )
        out = r.stdout.getvalue()
        log.debug('Ceph health: %s', out.rstrip('\n'))
        if out.split(None, 1)[0] == 'HEALTH_OK':
            break
        time.sleep(1)

def wait_until_fuse_mounted(remote, fuse, mountpoint):
    while True:
        proc = remote.run(
            args=[
                'stat',
                '--file-system',
                '--printf=%T\n',
                '--',
                mountpoint,
                ],
            stdout=StringIO(),
            )
        fstype = proc.stdout.getvalue().rstrip('\n')
        if fstype == 'fuseblk':
            break
        log.debug('cfuse not yet mounted, got fs type {fstype!r}'.format(fstype=fstype))

        # it shouldn't have exited yet; exposes some trivial problems
        assert not fuse.exitstatus.ready()

        time.sleep(5)
    log.info('cfuse is mounted on %s', mountpoint)

def reconnect(ctx, timeout):
    """
    Connect to all the machines in ctx.cluster.

    Presumably, some of them won't be up. Handle this
    by waiting for them, unless the wait time exceeds
    the specified timeout.

    ctx needs to contain the cluster of machines you
    wish it to try and connect to, as well as a config
    holding the ssh keys for each of them. As long as it
    contains this data, you can construct a context
    that is a subset of your full cluster.
    """
    import errno
    log.info('Re-opening connections...')
    starttime = time.time()
    need_reconnect = ctx.cluster.remotes.keys()
    while True:
        for remote in list(need_reconnect):
            try:
                log.info('trying to connect to %s', remote.name)
                from .orchestra import connection
                remote.ssh = connection.connect(
                    user_at_host=remote.name,
                    host_key=ctx.config['targets'][remote.name],
                    keep_alive=True,
                    )
            except socket.timeout:
                pass
            except socket.error as e:
                if hasattr(e, '__getitem__'):
                    if e[0] not in [errno.ECONNREFUSED, errno.ETIMEDOUT,
                                errno.EHOSTUNREACH, errno.EHOSTDOWN]:
                        log.exception('unknown socket error: %s', repr(e))
                        raise
                    else:
                        if time.time() - starttime > timeout:
                            raise
                else:
                    log.exception('weird socket error without error code')
                    raise
            except paramiko.SSHException:
                if time.time() - starttime > timeout:
                    raise
            else:
                need_reconnect.remove(remote)

        if not need_reconnect:
            break
        log.debug('waited {elapsed}'.format(elapsed=str(time.time() - starttime)))
        time.sleep(1)

def write_secret_file(remote, role, filename):
    remote.run(
        args=[
            '/tmp/cephtest/enable-coredump',
            '/tmp/cephtest/binary/usr/local/bin/ceph-coverage',
            '/tmp/cephtest/archive/coverage',
            '/tmp/cephtest/binary/usr/local/bin/ceph-authtool',
            '--name={role}'.format(role=role),
            '--print-key',
            '/tmp/cephtest/data/{role}.keyring'.format(role=role),
            run.Raw('>'),
            filename,
            ],
        )

def get_clients(ctx, roles):
    for role in roles:
        assert isinstance(role, basestring)
        PREFIX = 'client.'
        assert role.startswith(PREFIX)
        id_ = role[len(PREFIX):]
        (remote,) = ctx.cluster.only(role).remotes.iterkeys()
        yield (id_, remote)

def get_user():
    return getpass.getuser() + '@' + socket.gethostname()

def read_config(ctx):
    filename = os.path.join(os.environ['HOME'], '.teuthology.yaml')
    ctx.teuthology_config = {}
    with file(filename) as f:
        g = yaml.safe_load_all(f)
        for new in g:
            ctx.teuthology_config.update(new)

# return the "first" mon (alphanumerically, for lack of anything better)
def get_first_mon(ctx, config):
    mons = []
    for remote, roles in ctx.cluster.remotes.items():
        for role in roles:
            if not role.startswith('mon.'):
                continue
            mons.append(role)
            break
    firstmon = sorted(mons)[0]
    assert firstmon
    return firstmon

def replace_all_with_clients(cluster, config):
    """
    Converts a dict containing a key all to one
    mapping all clients to the value of config['all']
    """
    assert isinstance(config, dict), 'config must be a dict'
    if 'all' not in config:
        return config
    norm_config = {}
    assert len(config) == 1, \
        "config cannot have 'all' and specific clients listed"
    for client in all_roles_of_type(cluster, 'client'):
        norm_config['client.{id}'.format(id=client)] = config['all']
    return norm_config