ceph/teuthology/suite.py
2014-07-07 15:35:45 -06:00

644 lines
21 KiB
Python

# this file is responsible for submitting tests into the queue
# by generating combinations of facets found in
# https://github.com/ceph/ceph-qa-suite.git
import copy
from datetime import datetime
import itertools
import logging
import os
import requests
import pwd
import subprocess
import smtplib
import sys
import yaml
from email.mime.text import MIMEText
from tempfile import NamedTemporaryFile
import teuthology
from . import lock
from .config import config
from .repo_utils import enforce_repo_state, BranchNotFoundError
log = logging.getLogger(__name__)
def main(args):
verbose = args['--verbose']
if verbose:
teuthology.log.setLevel(logging.DEBUG)
dry_run = args['--dry-run']
base_yaml_paths = args['<config_yaml>']
suite = args['--suite']
nice_suite = suite.replace('/', ':')
ceph_branch = args['--ceph']
kernel_branch = args['--kernel']
kernel_flavor = args['--flavor']
teuthology_branch = args['--teuthology-branch']
machine_type = args['--machine-type']
distro = args['--distro']
suite_branch = args['--suite-branch'] or ceph_branch
suite_dir = args['--suite-dir']
limit = int(args['--limit'])
priority = int(args['--priority'])
num = int(args['--num'])
owner = args['--owner']
email = args['--email']
if email:
config.results_email = email
timeout = args['--timeout']
name = make_run_name(nice_suite, ceph_branch, kernel_branch, kernel_flavor,
machine_type)
if suite_dir:
suite_repo_path = suite_dir
else:
suite_repo_path = fetch_suite_repo(suite_branch, test_name=name)
config_string = create_initial_config(nice_suite, ceph_branch,
teuthology_branch, kernel_branch,
kernel_flavor, distro, machine_type)
with NamedTemporaryFile(prefix='schedule_suite_',
delete=False) as base_yaml:
base_yaml.write(config_string)
base_yaml_path = base_yaml.name
base_yaml_paths.insert(0, base_yaml_path)
prepare_and_schedule(owner=owner,
name=name,
suite=suite,
machine_type=machine_type,
suite_repo_path=suite_repo_path,
base_yaml_paths=base_yaml_paths,
priority=priority,
limit=limit,
num=num,
timeout=timeout,
dry_run=dry_run,
verbose=verbose,
)
os.remove(base_yaml_path)
def make_run_name(suite, ceph_branch, kernel_branch, kernel_flavor,
machine_type, user=None, timestamp=None):
"""
Generate a run name based on the parameters. A run name looks like:
teuthology-2014-06-23_19:00:37-rados-dumpling-testing-basic-plana
"""
if not user:
user = pwd.getpwuid(os.getuid()).pw_name
# We assume timestamp is a datetime.datetime object
if not timestamp:
timestamp = datetime.now().strftime('%Y-%m-%d_%H:%M:%S')
worker = get_worker(machine_type)
return '-'.join(
[user, str(timestamp), suite, ceph_branch,
kernel_branch, kernel_flavor, worker]
)
def fetch_suite_repo(branch, test_name):
"""
Fetch the suite repo (and also the teuthology repo) so that we can use it
to build jobs. Repos are stored in ~/src/.
The reason the teuthology repo is also fetched is that currently we use
subprocess to call teuthology-schedule to schedule jobs so we need to make
sure it is up-to-date. For that reason we always fetch the master branch
for test scheduling, regardless of what teuthology branch is requested for
testing.
:returns: The path to the repo on disk
"""
src_base_path = os.path.expanduser('~/src')
if not os.path.exists(src_base_path):
os.mkdir(src_base_path)
suite_repo_path = os.path.join(src_base_path,
'ceph-qa-suite_' + branch)
try:
# When a user is scheduling a test run from their own copy of
# teuthology, let's not wreak havoc on it.
if config.automated_scheduling:
enforce_repo_state(
repo_url=os.path.join(config.ceph_git_base_url,
'teuthology.git'),
dest_path=os.path.join(src_base_path, 'teuthology'),
branch='master',
remove_on_error=False,
)
enforce_repo_state(
repo_url=os.path.join(config.ceph_git_base_url,
'ceph-qa-suite.git'),
dest_path=suite_repo_path,
branch=branch,
)
except BranchNotFoundError as exc:
schedule_fail(message=str(exc), name=test_name)
return suite_repo_path
def create_initial_config(nice_suite, ceph_branch, teuthology_branch,
kernel_branch, kernel_flavor, distro, machine_type):
"""
Put together the config file used as the basis for each job in the run.
Grabs hashes for the latest ceph, kernel and teuthology versions in the
branches specified and specifies them so we know exactly what we're
testing.
:returns: A yaml-formatted string
"""
# Put together a stanza specifying the kernel hash
if kernel_branch == 'distro':
kernel_hash = 'distro'
# Skip the stanza if the branch passed is '-'
elif kernel_branch == '-':
kernel_hash = None
else:
kernel_hash = get_hash('kernel', kernel_branch, kernel_flavor,
machine_type)
if not kernel_hash:
schedule_fail(message="Kernel branch '{branch}' not found".format(
branch=kernel_branch))
if kernel_hash:
log.info("kernel sha1: {hash}".format(hash=kernel_hash))
kernel_dict = dict(kernel=dict(kdb=True, sha1=kernel_hash))
kernel_stanza = yaml.dump(kernel_dict,
default_flow_style=False).strip()
else:
kernel_stanza = ''
# Get the ceph hash
ceph_hash = get_hash('ceph', ceph_branch, kernel_flavor, machine_type)
if not ceph_hash:
schedule_fail("Ceph branch '{branch}' not found".format(
branch=ceph_branch))
log.info("ceph sha1: {hash}".format(hash=ceph_hash))
# Get the ceph package version
ceph_version = package_version_for_hash(ceph_hash, kernel_flavor,
machine_type)
if not ceph_version:
schedule_fail("Packages for ceph version '{ver}' not found".format(
ver=ceph_version))
log.info("ceph version: {ver}".format(ver=ceph_version))
# Decide what branch of s3-tests to use
if get_branch_info('s3-tests', ceph_branch):
s3_branch = ceph_branch
else:
log.info("branch {0} not in s3-tests.git; will use master for"
"s3-tests".format(ceph_branch))
s3_branch = 'master'
log.info("s3-tests branch: %s", s3_branch)
if not teuthology_branch:
# Decide what branch of teuthology to use
if get_branch_info('teuthology', ceph_branch):
teuthology_branch = ceph_branch
else:
log.info("branch {0} not in teuthology.git; will use master for"
"teuthology".format(ceph_branch))
teuthology_branch = 'master'
log.info("teuthology branch: %s", teuthology_branch)
config_input = dict(
nice_suite=nice_suite,
ceph_branch=ceph_branch,
ceph_hash=ceph_hash,
teuthology_branch=teuthology_branch,
machine_type=machine_type,
kernel_stanza=kernel_stanza,
distro=distro,
s3_branch=s3_branch,
)
return config_template.format(**config_input)
def prepare_and_schedule(owner, name, suite, machine_type, suite_repo_path,
base_yaml_paths, priority, limit, num, timeout,
dry_run, verbose):
"""
Puts together some "base arguments" with which to execute
teuthology-schedule for each job, then passes them and other parameters to
schedule_suite(). Finally, schedules a "last-in-suite" job that sends an
email to the specified address (if one is configured).
"""
email = config.results_email
arch = get_arch(machine_type)
base_args = [
os.path.join(os.path.dirname(sys.argv[0]), 'teuthology-schedule'),
'--name', name,
'--num', str(num),
'--worker', get_worker(machine_type),
]
if priority:
base_args.extend(['--priority', str(priority)])
if verbose:
base_args.append('-v')
if owner:
base_args.extend(['--owner', owner])
suite_path = os.path.join(suite_repo_path, 'suites', suite)
num_jobs = schedule_suite(
name=suite,
path=suite_path,
base_yamls=base_yaml_paths,
base_args=base_args,
arch=arch,
machine_type=machine_type,
limit=limit,
dry_run=dry_run,
)
if email and num_jobs:
arg = copy.deepcopy(base_args)
arg.append('--last-in-suite')
arg.extend(['--email', email])
if timeout:
arg.extend(['--timeout', timeout])
if dry_run:
log.info('dry-run: %s' % ' '.join(arg))
else:
subprocess.check_call(
args=arg,
)
def schedule_fail(message, name=''):
"""
If an email address has been specified anywhere, send an alert there. Then
raise a ScheduleFailError.
"""
email = config.results_email
if email:
subject = "Failed to schedule {name}".format(name=name)
msg = MIMEText(message)
msg['Subject'] = subject
msg['From'] = config.results_sending_email
msg['To'] = email
smtp = smtplib.SMTP('localhost')
smtp.sendmail(msg['From'], [msg['To']], msg.as_string())
smtp.quit()
raise ScheduleFailError(message, name)
class ScheduleFailError(RuntimeError):
def __init__(self, message, name=None):
self.message = message
self.name = name
def __str__(self):
return "Job scheduling {name} failed: '{msg}'".format(
name=self.name,
msg=self.message,
).replace(' ', ' ')
def get_worker(machine_type):
"""
Map a given machine_type to a beanstalkd worker. If machine_type mentions
multiple machine types - e.g. 'plana,mira', then this returns 'multi'.
Otherwise it returns what was passed.
"""
if ',' in machine_type:
return 'multi'
else:
return machine_type
def get_hash(project='ceph', branch='master', flavor='basic',
distro='ubuntu', machine_type='plana'):
"""
Find the hash representing the head of the project's repository via
querying a gitbuilder repo.
Will return None in the case of a 404 or any other HTTP error.
"""
# Alternate method for github-hosted projects - left here for informational
# purposes
# resp = requests.get(
# 'https://api.github.com/repos/ceph/ceph/git/refs/heads/master')
# hash = .json()['object']['sha']
(arch, release, pkg_type) = get_distro_defaults(distro, machine_type)
base_url = get_gitbuilder_url(project, release, pkg_type, arch, flavor)
url = os.path.join(base_url, 'ref', branch, 'sha1')
resp = requests.get(url)
if not resp.ok:
return None
return str(resp.text.strip())
def get_distro_defaults(distro, machine_type):
"""
Given a distro (e.g. 'ubuntu') and machine type, return:
(arch, release, pkg_type)
This is mainly used to default to:
('x86_64', 'precise', 'deb') when passed 'ubuntu' and 'plana'
And ('armv7l', 'saucy', 'deb') when passed 'ubuntu' and 'saya'
And ('x86_64', 'centos6', 'rpm') when passed anything non-ubuntu
"""
if distro == 'ubuntu':
if machine_type == 'saya':
arch = 'armv7l'
release = 'saucy'
pkg_type = 'deb'
else:
arch = 'x86_64'
release = 'precise'
pkg_type = 'deb'
else:
arch = 'x86_64'
release = 'centos6'
pkg_type = 'rpm'
log.debug(
"Defaults for machine_type %s: arch=%s, release=%s, pkg_type=%s)",
machine_type, arch, release, pkg_type)
return (
arch,
release,
pkg_type,
)
def get_gitbuilder_url(project, distro, pkg_type, arch, kernel_flavor):
"""
Return a base URL like:
http://gitbuilder.ceph.com/ceph-deb-squeeze-x86_64-basic/
:param project: 'ceph' or 'kernel'
:param distro: A distro-ish string like 'trusty' or 'fedora20'
:param pkg_type: Probably 'rpm' or 'deb'
:param arch: A string like 'x86_64'
:param kernel_flavor: A string like 'basic'
"""
templ = 'http://gitbuilder.ceph.com/{proj}-{pkg}-{distro}-{arch}-{flav}/'
return templ.format(proj=project, pkg=pkg_type, distro=distro, arch=arch,
flav=kernel_flavor)
def package_version_for_hash(hash, kernel_flavor='basic',
distro='ubuntu', machine_type='plana'):
"""
Does what it says on the tin. Uses gitbuilder repos.
:returns: a string.
"""
(arch, release, pkg_type) = get_distro_defaults(distro, machine_type)
base_url = get_gitbuilder_url('ceph', release, pkg_type, arch,
kernel_flavor)
url = os.path.join(base_url, 'sha1', hash, 'version')
resp = requests.get(url)
if resp.ok:
return resp.text.strip()
def get_branch_info(project, branch, project_owner='ceph'):
"""
Use the GitHub API to query a project's branch. Returns:
{u'object': {u'sha': <a_sha_string>,
u'type': <string>,
u'url': <url_to_commit>},
u'ref': u'refs/heads/<branch>',
u'url': <url_to_branch>}
We mainly use this to check if a branch exists.
"""
url_templ = 'https://api.github.com/repos/{project_owner}/{project}/git/refs/heads/{branch}' # noqa
url = url_templ.format(project_owner=project_owner, project=project,
branch=branch)
resp = requests.get(url)
if resp.ok:
return resp.json()
def schedule_suite(name,
path,
base_yamls,
base_args,
arch,
machine_type,
limit=0,
dry_run=True,
):
"""
schedule one suite.
returns number of jobs scheduled
"""
count = 0
log.debug('Suite %s in %s' % (name, path))
configs = [(combine_path(name, item[0]), item[1]) for item in
build_matrix(path)]
job_count = len(configs)
log.info('Suite %s in %s generated %d jobs' % (
name, path, len(configs)))
for description, fragment_paths in configs:
if limit > 0 and count >= limit:
log.info(
'Stopped after {limit} jobs due to --limit={limit}'.format(
limit=limit))
break
raw_yaml = '\n'.join([file(a, 'r').read() for a in fragment_paths])
parsed_yaml = yaml.load(raw_yaml)
os_type = parsed_yaml.get('os_type')
exclude_arch = parsed_yaml.get('exclude_arch')
exclude_os_type = parsed_yaml.get('exclude_os_type')
if exclude_arch and exclude_arch == arch:
log.info('Skipping due to excluded_arch: %s facets %s',
exclude_arch, description)
continue
if exclude_os_type and exclude_os_type == os_type:
log.info('Skipping due to excluded_os_type: %s facets %s',
exclude_os_type, description)
continue
# We should not run multiple tests (changing distros) unless the
# machine is a VPS.
# Re-imaging baremetal is not yet supported.
if machine_type != 'vps' and os_type and os_type != 'ubuntu':
log.info(
'Skipping due to non-ubuntu on baremetal facets %s',
description)
continue
log.info(
'Scheduling %s', description
)
arg = copy.deepcopy(base_args)
arg.extend([
'--description', description,
'--',
])
arg.extend(base_yamls)
arg.extend(fragment_paths)
if dry_run:
# Quote any individual args so that individual commands can be
# copied and pasted in order to execute them individually.
printable_args = []
for item in arg:
if ' ' in item:
printable_args.append("'%s'" % item)
else:
printable_args.append(item)
log.info('dry-run: %s' % ' '.join(printable_args))
else:
subprocess.check_call(
args=arg,
)
count += 1
return job_count
def combine_path(left, right):
"""
os.path.join(a, b) doesn't like it when b is None
"""
if right:
return os.path.join(left, right)
return left
def build_matrix(path):
"""
Return a list of items describe by path
The input is just a path. The output is an array of (description,
[file list]) tuples.
For a normal file we generate a new item for the result list.
For a directory, we (recursively) generate a new item for each
file/dir.
For a directory with a magic '+' file, we generate a single item
that concatenates all files/subdirs.
For a directory with a magic '%' file, we generate a result set
for each item in the directory, and then do a product to generate
a result list with all combinations.
The final description (after recursion) for each item will look
like a relative path. If there was a % product, that path
component will appear as a file with braces listing the selection
of chosen subitems.
"""
if os.path.isfile(path):
if path.endswith('.yaml'):
return [(None, [path])]
if os.path.isdir(path):
files = sorted(os.listdir(path))
if '+' in files:
# concatenate items
files.remove('+')
raw = []
for fn in files:
raw.extend(build_matrix(os.path.join(path, fn)))
out = [(
'{' + ' '.join(files) + '}',
[a[1][0] for a in raw]
)]
return out
elif '%' in files:
# convolve items
files.remove('%')
sublists = []
for fn in files:
raw = build_matrix(os.path.join(path, fn))
sublists.append([(combine_path(fn, item[0]), item[1])
for item in raw])
out = []
if sublists:
for sublist in itertools.product(*sublists):
name = '{' + ' '.join([item[0] for item in sublist]) + '}'
val = []
for item in sublist:
val.extend(item[1])
out.append((name, val))
return out
else:
# list items
out = []
for fn in files:
raw = build_matrix(os.path.join(path, fn))
out.extend([(combine_path(fn, item[0]), item[1])
for item in raw])
return out
return []
def get_arch(machine_type):
"""
Based on a given machine_type, return its architecture by querying the lock
server. Sound expensive? It is!
:returns: A string or None
"""
locks = lock.list_locks()
for machine in locks:
if machine['type'] == machine_type:
arch = machine['arch']
return arch
return None
# yaml template for the config that becomes the base for each generated job
# config
config_template = """
teuthology_branch: {teuthology_branch}
{kernel_stanza}
nuke-on-error: true
machine_type: {machine_type}
os_type: {distro}
branch: {ceph_branch}
suite: {nice_suite}
tasks:
- chef:
- clock.check:
overrides:
workunit:
sha1: {ceph_hash}
s3tests:
branch: {s3_branch}
install:
ceph:
sha1: {ceph_hash}
ceph:
sha1: {ceph_hash}
conf:
mon:
debug ms: 1
debug mon: 20
debug paxos: 20
osd:
debug ms: 1
debug osd: 20
debug filestore: 20
debug journal: 20
log-whitelist:
- slow request
ceph-deploy:
branch:
dev: {ceph_branch}
conf:
mon:
osd default pool size: 2
debug mon: 1
debug paxos: 20
debug ms: 20
client:
log file: /var/log/ceph/ceph-$name.$pid.log
admin_socket:
branch: {ceph_branch}
""".strip()