def stop(self, host):
"""
Stop Collectd service on the agent host.
:param host: the name of the agent host. Could be a list.
"""
log, barreleye_instance = init_env(self._bac_config_fpath,
self._bac_logdir,
self._bac_log_to_file)
host = cmd_general.check_argument_str(log, "host", host)
hostnames = cmd_general.parse_list_string(log, host)
if hostnames is None:
log.cl_error("host list [%s] is invalid", host)
cmd_general.cmd_exit(log, -1)
ret = barreleye_instance.bei_stop_agents(log, hostnames)
cmd_general.cmd_exit(log, ret)
def build(log,
source_dir,
workspace,
cache=constant.CORAL_BUILD_CACHE,
lustre_rpms_dir=None,
e2fsprogs_rpms_dir=None,
collectd=None,
enable_zfs=False,
enable_devel=False,
disable_plugin=None,
tsinghua_mirror=False):
"""
Build the Coral ISO.
"""
# pylint: disable=too-many-locals,too-many-branches
# pylint: disable=too-many-statements
if disable_plugin is None:
disabled_plugins = []
else:
disabled_plugins = cmd_general.parse_list_string(log, disable_plugin)
if disabled_plugins is None:
log.cl_error("invalid option [%s] of --disable_plugin",
disable_plugin)
return -1
plugins = list(build_common.CORAL_RELEASE_PLUGIN_DICT.values())
if enable_devel:
plugins += list(build_common.CORAL_DEVEL_PLUGIN_DICT.values())
sync_cache_back = True
disable_plugins_str = ""
for plugin_name in disabled_plugins:
if plugin_name not in build_common.CORAL_PLUGIN_DICT:
log.cl_error("unknown plugin [%s] of --disable_plugin",
plugin_name)
log.cl_error("possible plugins are %s",
list(build_common.CORAL_PLUGIN_DICT.keys()))
return -1
if ((not enable_devel) and
(plugin_name not in build_common.CORAL_RELEASE_PLUGIN_DICT)):
log.cl_info(
"plugin [%s] will not be included in release "
"ISO anyway", plugin_name)
continue
sync_cache_back = False
plugin = build_common.CORAL_PLUGIN_DICT[plugin_name]
if plugin in plugins:
plugins.remove(plugin)
disable_plugins_str += " --disable-%s" % plugin_name
if len(plugins) == 0:
log.cl_error("everything has been disabled, nothing to build")
return -1
need_lustre_rpms = False
need_collectd = False
install_lustre = False
enabled_plugin_str = ""
for plugin in plugins:
if enabled_plugin_str == "":
enabled_plugin_str = plugin.cpt_plugin_name
else:
enabled_plugin_str += ", " + plugin.cpt_plugin_name
if plugin.cpt_need_lustre_rpms:
need_lustre_rpms = True
if plugin.cpt_need_collectd:
need_collectd = True
if plugin.cpt_install_lustre:
install_lustre = True
type_fname = constant.CORAL_BUILD_CACHE_TYPE_OPEN
if type_fname == constant.CORAL_BUILD_CACHE_TYPE_OPEN:
log.cl_info("building ISO with %s", enabled_plugin_str)
local_host = ssh_host.get_local_host(ssh=False)
distro = local_host.sh_distro(log)
if distro not in (ssh_host.DISTRO_RHEL7, ssh_host.DISTRO_RHEL8):
log.cl_error("build on distro [%s] is not supported yet", distro)
return -1
shared_cache = cache.rstrip("/")
# Shared cache for this build type
shared_type_cache = shared_cache + "/" + type_fname
# Extra RPMs to download
extra_rpm_names = []
# Extra RPM file names under package directory
extra_package_fnames = []
# Extra file names under ISO directory
extra_iso_fnames = []
enable_zfs_string = ""
if enable_zfs:
enable_zfs_string = ", ZFS support disabled"
type_cache = workspace + "/" + type_fname
build_pip_dir = type_cache + "/" + constant.BUILD_PIP
iso_cache = type_cache + "/" + constant.ISO_CACHE_FNAME
# Directory path of package under ISO cache
packages_dir = iso_cache + "/" + constant.BUILD_PACKAGES
default_lustre_rpms_dir = (iso_cache + "/" +
constant.LUSTRE_RPM_DIR_BASENAME)
default_e2fsprogs_rpms_dir = (iso_cache + "/" +
constant.E2FSPROGS_RPM_DIR_BASENAME)
if not need_collectd:
if collectd is not None:
log.cl_warning(
"option [--collectd %s] has been ignored since "
"no need to have Collectd RPMs", collectd)
elif collectd is not None:
sync_cache_back = False
if not need_lustre_rpms and not install_lustre:
if lustre_rpms_dir is not None:
log.cl_warning(
"option [--lustre %s] has been ignored since "
"no need to have Lustre RPMs", lustre_rpms_dir)
if e2fsprogs_rpms_dir is not None:
log.cl_warning(
"option [--e2fsprogs %s] has been ignored since "
"no need to have Lustre RPMs", e2fsprogs_rpms_dir)
else:
if lustre_rpms_dir is None:
lustre_rpms_dir = default_lustre_rpms_dir
if e2fsprogs_rpms_dir is None:
e2fsprogs_rpms_dir = default_e2fsprogs_rpms_dir
command = ("mkdir -p %s" % workspace)
retval = local_host.sh_run(log, command)
if retval.cr_exit_status:
log.cl_error(
"failed to run command [%s] on host [%s], "
"ret = [%d], stdout = [%s], stderr = [%s]", command,
local_host.sh_hostname, retval.cr_exit_status, retval.cr_stdout,
retval.cr_stderr)
return -1
ret = get_shared_build_cache(log, local_host, workspace, shared_type_cache)
if ret:
log.cl_error("failed to get shared build cache")
return -1
target_cpu = local_host.sh_target_cpu(log)
if target_cpu is None:
log.cl_error("failed to get the target cpu on host [%s]",
local_host.sh_hostname)
return -1
ret = install_build_dependency(log,
workspace,
local_host,
distro,
target_cpu,
type_cache,
plugins,
build_pip_dir,
tsinghua_mirror=tsinghua_mirror)
if ret:
log.cl_error("failed to install dependency for building")
return -1
command = ("mkdir -p %s" % (packages_dir))
retval = local_host.sh_run(log, command)
if retval.cr_exit_status:
log.cl_error(
"failed to run command [%s] on host [%s], "
"ret = [%d], stdout = [%s], stderr = [%s]", command,
local_host.sh_hostname, retval.cr_exit_status, retval.cr_stdout,
retval.cr_stderr)
return -1
for plugin in plugins:
ret = plugin.cpt_build(log, workspace, local_host, source_dir,
target_cpu, type_cache, iso_cache, packages_dir,
extra_iso_fnames, extra_package_fnames,
extra_rpm_names, collectd)
if ret:
log.cl_error("failed to build plugin [%s]", plugin.cpt_plugin_name)
return -1
ret = download_dependent_rpms(log, workspace, local_host, distro,
target_cpu, packages_dir, type_cache,
extra_package_fnames, extra_rpm_names)
if ret:
log.cl_error("failed to download dependent rpms")
return -1
pip_dir = iso_cache + "/" + constant.BUILD_PIP
command = ("mkdir -p %s" % (pip_dir))
retval = local_host.sh_run(log, command)
if retval.cr_exit_status:
log.cl_error(
"failed to run command [%s] on host [%s], "
"ret = [%d], stdout = [%s], stderr = [%s]", command,
local_host.sh_hostname, retval.cr_exit_status, retval.cr_stdout,
retval.cr_stderr)
return -1
ret = install_common.download_pip3_packages(
log,
local_host,
pip_dir,
constant.CORAL_DEPENDENT_PIPS,
tsinghua_mirror=tsinghua_mirror)
if ret:
log.cl_error("failed to download pip3 packages")
return -1
lustre_distribution = None
if need_lustre_rpms or install_lustre:
if lustre_distribution is None:
log.cl_error("Lustre distribution is needed unexpectedly")
return -1
contents = ([constant.BUILD_PACKAGES, constant.BUILD_PIP] +
extra_iso_fnames)
ret = local_host.sh_check_dir_content(log,
iso_cache,
contents,
cleanup=True)
if ret:
log.cl_error("directory [%s] does not have expected content",
iso_cache)
return -1
log.cl_info("generating Coral ISO")
enable_zfs_string = ""
if enable_zfs:
enable_zfs_string = " --enable-zfs"
enable_devel_string = ""
if enable_devel:
enable_devel_string = " --enable-devel"
command = ("cd %s && rm coral-*.tar.bz2 coral-*.tar.gz -f && "
"sh autogen.sh && "
"./configure --with-iso-cache=%s%s%s%s && "
"make -j8 && "
"make iso" % (source_dir, iso_cache, enable_zfs_string,
enable_devel_string, disable_plugins_str))
retval = local_host.sh_watched_run(log,
command,
None,
None,
return_stdout=False,
return_stderr=False)
if retval.cr_exit_status:
log.cl_error("failed to run command [%s] on host [%s]", command,
local_host.sh_hostname)
return -1
# If there is any plugin disabled or Collectd is special, the local cache
# might have some things missing thus should not be used by other build.
if sync_cache_back:
ret = sync_shared_build_cache(log, local_host, type_cache,
shared_cache)
if ret:
log.cl_error("failed to sync to shared build cache")
return -1
log.cl_info("Built Coral ISO successfully")
return 0
def barrele_init_instance(log, workspace, config, config_fpath, log_to_file,
logdir_is_default):
"""
Parse the config and init the instance
"""
# pylint: disable=too-many-locals,too-many-branches,too-many-statements
collect_interval = utils.config_value(
config, barrele_constant.BRL_COLLECT_INTERVAL)
if collect_interval is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [%s]", barrele_constant.BRL_COLLECT_INTERVAL,
config_fpath, BARRELE_COLLECT_INTERVAL)
collect_interval = BARRELE_COLLECT_INTERVAL
continuous_query_periods = utils.config_value(
config, barrele_constant.BRL_CONTINUOUS_QUERY_PERIODS)
if continuous_query_periods is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [%s]",
barrele_constant.BRL_CONTINUOUS_QUERY_PERIODS, config_fpath,
BARRELE_CONTINUOUS_QUERY_PERIODS)
continuous_query_periods = BARRELE_CONTINUOUS_QUERY_PERIODS
jobstat_pattern = utils.config_value(config,
barrele_constant.BRL_JOBSTAT_PATTERN)
if jobstat_pattern is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [%s]", barrele_constant.BRL_JOBSTAT_PATTERN,
config_fpath, barrele_constant.BARRELE_JOBSTAT_PATTERN_UNKNOWN)
jobstat_pattern = barrele_constant.BARRELE_JOBSTAT_PATTERN_UNKNOWN
if jobstat_pattern not in barrele_constant.BARRELE_JOBSTAT_PATTERNS:
log.cl_error("unsupported jobstat_pattern [%s], supported: %s",
jobstat_pattern,
barrele_constant.BARRELE_JOBSTAT_PATTERNS)
return None
lustre_fallback_version_name = \
utils.config_value(config,
barrele_constant.BRL_LUSTRE_FALLBACK_VERSION)
if lustre_fallback_version_name is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [%s]",
barrele_constant.BRL_LUSTRE_FALLBACK_VERSION, config_fpath,
BARRELE_LUSTRE_FALLBACK_VERSION)
lustre_fallback_version_name = BARRELE_LUSTRE_FALLBACK_VERSION
if lustre_fallback_version_name not in lustre_version.LUSTRE_VERSION_DICT:
log.cl_error(
"unsupported Lustre version [%s] is configured in the "
"config file [%s]", lustre_fallback_version_name, config_fpath)
return None
lustre_fallback_version = \
lustre_version.LUSTRE_VERSION_DICT[lustre_fallback_version_name]
enable_lustre_exp_mdt = utils.config_value(
config, barrele_constant.BRL_ENABLE_LUSTRE_EXP_MDT)
if enable_lustre_exp_mdt is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [False]",
barrele_constant.BRL_ENABLE_LUSTRE_EXP_MDT, config_fpath)
enable_lustre_exp_mdt = False
enable_lustre_exp_ost = utils.config_value(
config, barrele_constant.BRL_ENABLE_LUSTRE_EXP_OST)
if enable_lustre_exp_ost is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [False]",
barrele_constant.BRL_ENABLE_LUSTRE_EXP_OST, config_fpath)
enable_lustre_exp_ost = False
agent_configs = utils.config_value(config, barrele_constant.BRL_AGENTS)
if agent_configs is None:
log.cl_error(
"can NOT find [%s] in the config file, "
"please correct file [%s]", barrele_constant.BRL_AGENTS,
config_fpath)
return None
host_dict = {}
barreleye_server = parse_server_config(log, config, config_fpath,
host_dict)
if barreleye_server is None:
log.cl_error("failed to parse server config")
return None
agent_dict = {}
for agent_config in agent_configs:
hostname_config = utils.config_value(agent_config,
barrele_constant.BRL_HOSTNAME)
if hostname_config is None:
log.cl_error(
"can NOT find [%s] in the config of SSH host "
"[%s], please correct file [%s]",
barrele_constant.BRL_HOSTNAME, hostname_config, config_fpath)
return None
hostnames = cmd_general.parse_list_string(log, hostname_config)
if hostnames is None:
log.cl_error("[%s] as [%s] is invalid in the config file [%s]",
hostname_config, barrele_constant.BRL_HOSTNAME,
config_fpath)
return None
ssh_identity_file = utils.config_value(
agent_config, barrele_constant.BRL_SSH_IDENTITY_FILE)
enable_disk = utils.config_value(agent_config,
barrele_constant.BRL_ENABLE_DISK)
if enable_disk is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [False]",
barrele_constant.BRL_ENABLE_DISK, config_fpath)
enable_disk = False
enable_infiniband = utils.config_value(
agent_config, barrele_constant.BRL_ENABLE_INFINIBAND)
if enable_infiniband is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [False]",
barrele_constant.BRL_ENABLE_INFINIBAND, config_fpath)
enable_infiniband = False
enable_lustre_client = utils.config_value(
agent_config, barrele_constant.BRL_ENABLE_LUSTRE_CLIENT)
if enable_lustre_client is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [False]",
barrele_constant.BRL_ENABLE_LUSTRE_CLIENT, config_fpath)
enable_lustre_client = False
enable_lustre_mds = utils.config_value(
agent_config, barrele_constant.BRL_ENABLE_LUSTRE_MDS)
if enable_lustre_mds is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [True]",
barrele_constant.BRL_ENABLE_LUSTRE_MDS, config_fpath)
enable_lustre_mds = True
enable_lustre_oss = utils.config_value(
agent_config, barrele_constant.BRL_ENABLE_LUSTRE_OSS)
if enable_lustre_oss is None:
log.cl_debug(
"no [%s] is configured in the config file [%s], "
"using default value [True]",
barrele_constant.BRL_ENABLE_LUSTRE_OSS, config_fpath)
enable_lustre_oss = True
for hostname in hostnames:
if hostname in agent_dict:
log.cl_error(
"agent of host [%s] is configured for multiple times",
hostname)
return None
host = ssh_host.get_or_add_host_to_dict(log, host_dict, hostname,
ssh_identity_file)
if host is None:
return None
agent = barrele_agent.BarreleAgent(
host,
barreleye_server,
enable_disk=enable_disk,
enable_lustre_oss=enable_lustre_oss,
enable_lustre_mds=enable_lustre_mds,
enable_lustre_client=enable_lustre_client,
enable_infiniband=enable_infiniband)
agent_dict[hostname] = agent
local_host = ssh_host.get_local_host()
instance = BarreleInstance(workspace, config, config_fpath, log_to_file,
logdir_is_default, local_host, collect_interval,
continuous_query_periods, jobstat_pattern,
lustre_fallback_version, enable_lustre_exp_mdt,
enable_lustre_exp_ost, host_dict, agent_dict,
barreleye_server)
return instance