class SlurmdbdCharm(CharmBase):
"""Slurmdbd Charm."""
_stored = StoredState()
def __init__(self, *args):
"""Set the default class attributes."""
super().__init__(*args)
self._stored.set_default(
db_info=dict(),
slurmdbd_config=dict(),
munge_key_available=False,
slurm_installed=False,
)
self._db = MySQLClient(self, "db")
self._nrpe = Nrpe(self, "nrpe-external-master")
self._slurm_manager = SlurmManager(self, "slurmdbd")
self._slurmdbd = Slurmdbd(self, "slurmdbd")
self._slurmdbd_peer = SlurmdbdPeer(self, "slurmdbd-peer")
event_handler_bindings = {
self.on.install: self._on_install,
self.on.config_changed: self._write_config_and_restart_slurmdbd,
self._db.on.database_available: self._write_config_and_restart_slurmdbd,
self._slurmdbd_peer.on.slurmdbd_peer_available: self._write_config_and_restart_slurmdbd,
self._slurmdbd.on.slurmdbd_available: self._write_config_and_restart_slurmdbd,
self._slurmdbd.on.slurmdbd_unavailable: self._on_slurm_configurator_unavailable,
self._slurmdbd.on.munge_key_available: self._on_munge_key_available,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
self._slurm_manager.install(self.config["snapstore-channel"])
self._stored.slurm_installed = True
self.unit.status = ActiveStatus("slurm snap successfully installed")
def _on_upgrade(self, event):
"""Handle upgrade charm event."""
self._slurm_manager.upgrade()
def _on_munge_key_available(self, event):
if not self._stored.slurm_installed:
event.defer()
return
munge_key = self._slurmdbd.get_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
self._slurm_manager.restart_munged()
self._stored.munge_key_available = True
def _on_slurm_configurator_unavailable(self, event):
self._stored.munge_key_available = False
self._check_status()
def _check_status(self) -> bool:
"""Check that we have the things we need."""
db_info = self._stored.db_info
munge_key_available = self._stored.munge_key_available
slurm_installed = self._stored.slurm_installed
slurmdbd_info = self._slurmdbd_peer.get_slurmdbd_info()
deps = [
slurmdbd_info,
db_info,
slurm_installed,
munge_key_available,
]
if not all(deps):
if not db_info:
self.unit.status = BlockedStatus(
"Need relation to MySQL."
)
elif not munge_key_available:
self.unit.status = BlockedStatus(
"Need relation to slurm-configurator."
)
return False
return True
def _write_config_and_restart_slurmdbd(self, event):
"""Check for prereqs before writing config/restart of slurmdbd."""
# Ensure all pre-conditions are met with _check_status(), if not
# defer the event.
if not self._check_status():
event.defer()
return
db_info = self._stored.db_info
slurmdbd_info = self._slurmdbd_peer.get_slurmdbd_info()
slurmdbd_stored_config = dict(self._stored.slurmdbd_config)
slurmdbd_config = {
**self.config,
**slurmdbd_info,
**db_info,
}
if slurmdbd_config != slurmdbd_stored_config:
self._stored.slurmdbd_config = slurmdbd_config
self._slurm_manager.render_slurm_configs(slurmdbd_config)
self._slurm_manager.restart_slurm_component()
# Only the leader can set relation data on the application.
# Enforce that no one other then the leader trys to set
# application relation data.
if self.model.unit.is_leader():
self._slurmdbd.set_slurmdbd_info_on_app_relation_data(
slurmdbd_config,
)
self.unit.status = ActiveStatus("slurmdbd available")
def get_port(self):
"""Return the port from slurm-ops-manager."""
return self._slurm_manager.port
def get_hostname(self):
"""Return the hostname from slurm-ops-manager."""
return self._slurm_manager.hostname
def set_db_info(self, db_info):
"""Set the db_info in the stored state."""
self._stored.db_info = db_info
class SlurmrestdCharm(CharmBase):
"""Operator charm responsible for lifecycle operations for slurmrestd."""
_stored = StoredState()
def __init__(self, *args):
"""Initialize charm and configure states and events to observe."""
super().__init__(*args)
self._stored.set_default(
slurm_installed=False,
slurmrestd_restarted=False,
)
self._slurm_manager = SlurmManager(self, "slurmrestd")
self._slurmrestd = SlurmrestdRequires(self, 'slurmrestd')
event_handler_bindings = {
self.on.install: self._on_install,
self.on.upgrade_charm: self._on_upgrade,
self._slurmrestd.on.config_available:
self._on_check_status_and_write_config,
self._slurmrestd.on.munge_key_available:
self._on_configure_munge_key,
self._slurmrestd.on.restart_slurmrestd:
self._on_restart_slurmrestd,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
self._slurm_manager.install(self.config["snapstore-channel"])
self.unit.status = ActiveStatus("slurm installed")
self._stored.slurm_installed = True
def _on_upgrade(self, event):
"""Upgrade charm event handler."""
slurm_config = self._check_status()
snapstore_channel = self.config["snapstore-channel"]
self._slurm_manager.upgrade(slurm_config, snapstore_channel)
def _on_restart_slurmrestd(self, event):
"""Resart the slurmrestd component."""
self._slurm_manager.restart_slurm_component()
def _on_configure_munge_key(self, event):
"""Configure the munge key.
1) Get the munge key from the stored state of the slurmrestd relation
2) Write the munge key to the munge key path and chmod
3) Restart munged
4) Set munge_key_available in charm stored state
"""
if not self._stored.slurm_installed:
event.defer()
return
munge_key = self._slurmrestd.get_stored_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
self._slurm_manager.restart_munged()
self._stored.munge_key_available = True
def _check_status(self):
slurm_config = self._slurmrestd.get_stored_slurm_config()
munge_key_available = self._stored.munge_key_available
slurm_configurator_joined = self._slurmrestd.is_joined
# Check and see if we have what we need for operation.
if not slurm_configurator_joined:
self.unit.status = BlockedStatus(
"Needed relations: slurm-configurator")
return None
elif not (munge_key_available and slurm_config):
self.unit.status = WaitingStatus("Waiting on: configuration")
return None
return dict(slurm_config)
def _on_check_status_and_write_config(self, event):
slurm_config = self._check_status()
if not slurm_config:
event.defer()
return
self._slurm_manager.render_slurm_configs(slurm_config)
# Only restart slurmrestd the first time the node is brought up.
if not self._stored.slurmrestd_restarted:
self._slurm_manager.restart_slurm_component()
self._stored.slurmrestd_restarted = True
self.unit.status = ActiveStatus("slurmrestd available")
class SlurmdCharm(CharmBase):
"""Slurmd lifecycle events."""
_stored = StoredState()
def __init__(self, *args):
"""Init _stored attributes and interfaces, observe events."""
super().__init__(*args)
self._stored.set_default(
munge_key_available=False,
slurmd_restarted=False,
user_node_state=str(),
partition_name=str(),
)
self._nrpe = Nrpe(self, "nrpe-external-master")
self._slurm_manager = SlurmManager(self, "slurmd")
self._slurmd = Slurmd(self, "slurmd")
self._slurmd_peer = SlurmdPeer(self, "slurmd-peer")
event_handler_bindings = {
self.on.install: self._on_install,
self.on.upgrade_charm: self._on_upgrade,
self.on.start: self._on_check_status_and_write_config,
self.on.config_changed: self._on_config_changed,
self._slurmd_peer.on.slurmd_peer_available:
self._on_set_partition_info_on_app_relation_data,
self._slurmd_peer.on.slurmd_peer_departed:
self._on_set_partition_info_on_app_relation_data,
self._slurmd.on.slurm_config_available:
self._on_check_status_and_write_config,
self._slurmd.on.slurm_config_unavailable:
self._on_check_status_and_write_config,
self._slurmd.on.restart_slurmd:
self._on_restart_slurmd,
self._slurmd.on.munge_key_available: self._on_write_munge_key,
self.on.set_node_state_action: self._on_set_node_state_action,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
self._slurm_manager.install(self.config["snapstore-channel"])
if self.model.unit.is_leader():
self._get_set_partition_name()
logger.debug(f"PARTITION_NAME: {self._stored.partition_name}")
self._stored.slurm_installed = True
self.unit.status = ActiveStatus("Slurm installed")
def _on_upgrade(self, event):
slurm_config = self._check_status()
if not slurm_config:
event.defer()
return
self._slurm_manager.upgrade(
slurm_config,
self.config["snapstore-channel"]
)
def _on_config_changed(self, event):
if self.model.unit.is_leader():
self._get_set_partition_name()
if self._check_status():
self._on_set_partition_info_on_app_relation_data(
event
)
def _on_write_munge_key(self, event):
if not self._stored.slurm_installed:
event.defer()
return
munge_key = self._slurmd.get_stored_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
self._slurm_manager.restart_munged()
self._stored.munge_key_available = True
def _on_check_status_and_write_config(self, event):
slurm_config = self._check_status()
if not slurm_config:
event.defer()
return
# if slurm_config['configless']:
# slurmctld_hostname = slurm_config['active_controller_hostname']
# self._slurm_manager.configure_slurmctld_hostname(
# slurmctld_hostname
# )
# self._slurm_manager.restart_slurm_component()
# else:
# Ensure we aren't dealing with a StoredDict before trying
# to render the slurm.conf.
slurm_config = dict(slurm_config)
self._slurm_manager.render_slurm_configs(slurm_config)
# Only restart slurmd the first time the node is brought up.
if not self._stored.slurmd_restarted:
self._slurm_manager.restart_slurm_component()
self._stored.slurmd_restarted = True
self.unit.status = ActiveStatus("slurmd available")
def _on_restart_slurmd(self, event):
self._slurm_manager.restart_slurm_component()
def _check_status(self):
munge_key_available = self._stored.munge_key_available
slurm_installed = self._stored.slurm_installed
slurm_config = self._slurmd.get_stored_slurm_config()
slurmd_joined = self._slurmd.is_joined
if not slurmd_joined:
self.unit.status = BlockedStatus(
"Needed relations: slurm-configurator"
)
return None
elif not (munge_key_available and slurm_config and slurm_installed):
self.unit.status = WaitingStatus(
"Waiting on: configuration"
)
return None
return dict(slurm_config)
def _on_set_node_state_action(self, event):
"""Set the node state."""
self._stored.user_node_state = event.params["node-state"]
self._on_set_partition_info_on_app_relation_data(event)
def _on_set_partition_info_on_app_relation_data(self, event):
"""Set the slurm partition info on the application relation data."""
# Only the leader can set data on the relation.
if self.framework.model.unit.is_leader():
# If the relation with slurm-configurator exists then set our
# partition info on the application relation data.
# This handler shouldn't fire if the relation isn't made,
# but add this extra check here just incase.
if self._slurmd.is_joined:
partition = self._assemble_partition()
if partition:
self._slurmd.set_partition_info_on_app_relation_data(
partition
)
return
event.defer()
return
def _assemble_partition(self):
"""Assemble the partition info."""
partition_name = self._stored.partition_name
partition_config = self.config.get("partition-config")
partition_state = self.config.get("partition-state")
slurmd_inventory = self._assemble_slurmd_inventory()
return {
"inventory": slurmd_inventory,
"partition_name": partition_name,
"partition_state": partition_state,
"partition_config": partition_config,
}
def _assemble_slurmd_inventory(self):
"""Apply mutations to nodes in the partition, return slurmd nodes."""
slurmd_inventory = self._slurmd_peer.get_slurmd_inventory()
if not slurmd_inventory:
return None
# If the user has set custom state for nodes
# ensure we update the state for the targeted nodes.
user_node_state = self._stored.user_node_state
if user_node_state:
node_states = {
item.split("=")[0]: item.split("=")[1]
for item in user_node_state.split(",")
}
# Copy the slurmd_inventory returned from the the slurmd-peer
# relation to a temporary variable that we will use to
# iterate over while we conditionally make modifications to the
# original inventory.
slurmd_inventory_tmp = copy.deepcopy(slurmd_inventory)
# Iterate over the slurmd nodes in the partition and check
# for nodes that need their state modified.
for partition in slurmd_inventory_tmp:
partition_tmp = copy.deepcopy(partition)
for slurmd_node in partition["inventory"]:
if slurmd_node["hostname"] in node_states.keys():
slurmd_node_tmp = copy.deepcopy(slurmd_node)
slurmd_node_tmp["state"] = \
node_states[slurmd_node["hostname"]]
partition_tmp["inventory"].remove(slurmd_node)
partition_tmp["inventory"].append(slurmd_node_tmp)
slurmd_inventory.remove(partition)
slurmd_inventory.append(partition_tmp)
return slurmd_inventory
def _get_set_partition_name(self):
"""Set the partition name."""
# Determine if a partition-name config exists, if so
# ensure the self._stored.partition_name is consistent with the
# supplied config.
# If no partition name has been specified then generate one.
partition_name = self.config.get("partition-name")
if partition_name:
if partition_name != self._stored.partition_name:
self._stored.partition_name = partition_name
elif not self._stored.partition_name:
self._stored.partition_name = f"juju-compute-{random_string()}"
return
def get_partition_name(self):
"""Return the partition_name."""
return self._stored.partition_name
def get_slurm_component(self):
"""Return the slurm component."""
return self._slurm_manager.slurm_component
def get_hostname(self):
"""Return the hostname."""
return self._slurm_manager.hostname
def get_port(self):
"""Return the port."""
return self._slurm_manager.port
class SlurmctldCharm(CharmBase):
"""Slurmctld lifecycle events."""
_stored = StoredState()
def __init__(self, *args):
"""Init _stored attributes and interfaces, observe events."""
super().__init__(*args)
self._stored.set_default(
munge_key_available=False,
slurmctld_controller_type=str(),
)
self._nrpe = Nrpe(self, "nrpe-external-master")
self._slurm_manager = SlurmManager(self, "slurmctld")
self._slurmctld = Slurmctld(self, "slurmctld")
self._slurmctld_peer = SlurmctldPeer(self, "slurmctld-peer")
event_handler_bindings = {
self.on.install: self._on_install,
self._slurmctld.on.slurm_config_available: self._on_check_status_and_write_config,
self._slurmctld.on.scontrol_reconfigure: self._on_scontrol_reconfigure,
self._slurmctld.on.restart_slurmctld: self._on_restart_slurmctld,
self._slurmctld.on.munge_key_available: self._on_write_munge_key,
self._slurmctld_peer.on.slurmctld_peer_available: self._on_slurmctld_peer_available,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
self._slurm_manager.install(self.config["snapstore-channel"])
self._stored.slurm_installed = True
self.unit.status = ActiveStatus("slurm snap successfully installed")
def _on_upgrade(self, event):
slurm_config = dict(self._check_status())
snapstore_channel = self.config["snapstore-channel"]
self._slurm_manager.upgrade(slurm_config, snapstore_channel)
def _on_write_munge_key(self, event):
if not self._stored.slurm_installed:
event.defer()
return
munge_key = self._slurmctld.get_stored_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
self._slurm_manager.restart_munged()
self._stored.munge_key_available = True
def _on_slurmctld_peer_available(self, event):
if self.framework.model.unit.is_leader():
if self._slurmctld.is_joined:
slurmctld_info = self._slurmctld_peer.get_slurmctld_info()
if slurmctld_info:
self._slurmctld.set_slurmctld_info_on_app_relation_data(
slurmctld_info
)
return
event.defer()
return
def _on_check_status_and_write_config(self, event):
slurm_config = self._check_status()
if not slurm_config:
event.defer()
return
self._slurm_manager.render_slurm_configs(dict(slurm_config))
self.unit.status = ActiveStatus("slurmctld available")
def _on_restart_slurmctld(self, event):
self._slurm_manager.restart_slurm_component()
def _on_scontrol_reconfigure(self, event):
self._slurm_manager.slurm_cmd("scontrol", "reconfigure")
def _check_status(self):
munge_key_available = self._stored.munge_key_available
slurm_installed = self._stored.slurm_installed
slurm_config = self._slurmctld.get_stored_slurm_config()
slurmctld_joined = self._slurmctld.is_joined
if not slurmctld_joined:
self.unit.status = BlockedStatus(
"Relations needed: slurm-configurator"
)
return None
elif not (munge_key_available and slurm_installed and slurm_config):
self.unit.status = WaitingStatus(
"Waiting on: configuration"
)
return None
return slurm_config
def get_slurm_component(self):
"""Return the slurm component."""
return self._slurm_manager.slurm_component
def get_hostname(self):
"""Return the hostname."""
return self._slurm_manager.hostname
def get_port(self):
"""Return the port."""
return self._slurm_manager.port
class SlurmdCharm(CharmBase):
"""Slurmd lifecycle events."""
_stored = StoredState()
on = SlurmdCharmEvents()
def __init__(self, *args):
"""Init _stored attributes and interfaces, observe events."""
super().__init__(*args)
self._stored.set_default(nhc_conf=str(),
slurm_installed=False,
slurmctld_available=False,
slurmctld_started=False,
cluster_name=str())
self._slurm_manager = SlurmManager(self, "slurmd")
self._fluentbit = FluentbitClient(self, "fluentbit")
# interface to slurmctld, should only have one slurmctld per slurmd app
self._slurmd = Slurmd(self, "slurmd")
self._slurmd_peer = SlurmdPeer(self, "slurmd-peer")
event_handler_bindings = {
self.on.install: self._on_install,
self.on.upgrade_charm: self._on_upgrade,
self.on.update_status: self._on_update_status,
self.on.config_changed: self._on_config_changed,
self.on.slurmctld_started: self._on_slurmctld_started,
self.on.slurmd_start: self._on_slurmd_start,
self.on.check_etcd: self._on_check_etcd,
self._slurmd.on.slurmctld_available: self._on_slurmctld_available,
self._slurmd.on.slurmctld_unavailable:
self._on_slurmctld_unavailable,
# fluentbit
self.on["fluentbit"].relation_created:
self._on_configure_fluentbit,
# actions
self.on.version_action: self._on_version_action,
self.on.node_configured_action: self._on_node_configured_action,
self.on.get_node_inventory_action:
self._on_get_node_inventory_action,
self.on.show_nhc_config_action: self._on_show_nhc_config,
# infiniband actions
self.on.get_infiniband_repo_action: self.get_infiniband_repo,
self.on.set_infiniband_repo_action: self.set_infiniband_repo,
self.on.install_infiniband_action: self.install_infiniband,
self.on.uninstall_infiniband_action: self.uninstall_infiniband,
self.on.start_infiniband_action: self.start_infiniband,
self.on.enable_infiniband_action: self.enable_infiniband,
self.on.stop_infiniband_action: self.stop_infiniband,
self.on.is_active_infiniband_action: self.is_active_infiniband,
# nvdia actions
self.on.nvidia_repo_action: self.nvidia_repo,
self.on.nvidia_package_action: self.nvidia_package,
self.on.nvidia_install_action: self.nvidia_install,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
"""Perform installation operations for slurmd."""
self.unit.set_workload_version(Path("version").read_text().strip())
self.unit.status = WaitingStatus("Installing slurmd")
custom_repo = self.config.get("custom-slurm-repo")
successful_installation = self._slurm_manager.install(custom_repo)
logger.debug(f"### slurmd installed: {successful_installation}")
if successful_installation:
self._stored.slurm_installed = True
else:
self.unit.status = BlockedStatus("Error installing slurmd")
event.defer()
self._check_status()
def _on_configure_fluentbit(self, event):
"""Set up Fluentbit log forwarding."""
self._configure_fluentbit()
def _configure_fluentbit(self):
logger.debug("## Configuring fluentbit")
cfg = list()
cfg.extend(self._slurm_manager.fluentbit_config_nhc)
cfg.extend(self._slurm_manager.fluentbit_config_slurm)
self._fluentbit.configure(cfg)
def _on_upgrade(self, event):
"""Perform upgrade operations."""
self.unit.set_workload_version(Path("version").read_text().strip())
def _on_update_status(self, event):
"""Handle update status."""
self._check_status()
def _check_status(self) -> bool:
"""Check if we heve all needed components.
- partition name
- slurm installed
- slurmctld available and working
- munge key configured and working
"""
if self._slurm_manager.needs_reboot:
self.unit.status = BlockedStatus("Machine needs reboot")
return False
if not self.get_partition_name():
self.unit.status = WaitingStatus("Waiting on charm configuration")
return False
if not self._stored.slurm_installed:
self.unit.status = BlockedStatus("Error installing slurmd")
return False
if not self._slurmd.is_joined:
self.unit.status = BlockedStatus("Need relations: slurmctld")
return False
if not self._stored.slurmctld_available:
self.unit.status = WaitingStatus("Waiting on: slurmctld")
return False
if not self._slurm_manager.check_munged():
self.unit.status = BlockedStatus("Error configuring munge key")
return False
if not self._stored.slurmctld_started:
self.unit.status = WaitingStatus("Waiting slurmctld to start")
return False
self.unit.status = ActiveStatus("slurmd available")
return True
def ensure_slurmd_starts(self, max_attemps=10) -> bool:
"""Ensure slurmd is up and running."""
logger.debug("## Stoping slurmd")
self._slurm_manager.slurm_systemctl('stop')
for i in range(max_attemps):
if self._slurm_manager.slurm_is_active():
logger.debug("## Slurmd running")
break
else:
logger.warning("## Slurmd not running, trying to start it")
self.unit.status = WaitingStatus("Starting slurmd")
self._slurm_manager.restart_slurm_component()
sleep(2 + i)
if self._slurm_manager.slurm_is_active():
return True
else:
self.unit.status = BlockedStatus("Cannot start slurmd")
return False
def _set_slurmctld_available(self, flag: bool):
"""Change stored value for slurmctld availability."""
self._stored.slurmctld_available = flag
def _set_slurmctld_started(self, flag: bool):
"""Change stored value for slurmctld started."""
self._stored.slurmctld_started = flag
def _on_slurmctld_available(self, event):
"""Get data from slurmctld and send inventory."""
if not self._stored.slurm_installed:
event.defer()
return
logger.debug(
'#### Slurmctld available - setting overrides for configless')
# get slurmctld host:port from relation and override systemd services
host = self._slurmd.slurmctld_hostname
port = self._slurmd.slurmctld_port
self._slurm_manager.create_configless_systemd_override(host, port)
self._slurm_manager.daemon_reload()
self._write_munge_key_and_restart_munge()
self._set_slurmctld_available(True)
self._on_set_partition_info_on_app_relation_data(event)
self._check_status()
# check etcd for hostnames
self.on.check_etcd.emit()
def _on_check_etcd(self, event):
"""Check if node is accounted for.
Check if slurmctld accounted for this node's inventory for the first
time, if so, emit slurmctld_started event, so the node can start the
daemon.
"""
host = self._slurmd.slurmctld_address
port = self._slurmd.etcd_port
logger.debug(f"## Connecting to etcd3 in {host}:{port}")
client = Etcd3Client(host=host, port=port, api_path="/v3/")
logger.debug("## Querying etcd3 for node list")
try:
v = client.get(key="all_nodes")
logger.debug(f"## Got: {v}")
except Exception as e:
logger.error(
f"## Unable to connect to {host} to get list of nodes: {e}")
event.defer()
return
node_accounted = False
if v:
hostnames = json.loads(v[0])
logger.debug(f"### etcd3 node list: {hostnames}")
if self.hostname in hostnames:
self.on.slurmctld_started.emit()
node_accounted = True
if not node_accounted:
logger.debug("## Node not accounted for. Deferring.")
event.defer()
def _on_slurmctld_unavailable(self, event):
logger.debug("## Slurmctld unavailable")
self._set_slurmctld_available(False)
self._set_slurmctld_started(False)
self._slurm_manager.slurm_systemctl('stop')
self._check_status()
def _on_slurmctld_started(self, event):
"""Set flag to True and emit slurmd_start event."""
self._set_slurmctld_started(True)
self.on.slurmd_start.emit()
def _on_slurmd_start(self, event):
if not self._check_status():
event.defer()
return
# only set up fluentbit if we have a relation to it
if self._fluentbit._relation is not None:
self._configure_fluentbit()
# at this point, we have slurm installed, munge configured, and we know
# slurmctld accounted for this node. It should be safe to start slurmd
if self.ensure_slurmd_starts():
logger.debug("## slurmctld started and slurmd is running")
else:
event.defer()
self._check_status()
def _on_config_changed(self, event):
"""Handle charm configuration changes."""
if self.model.unit.is_leader():
logger.debug("## slurmd config changed - leader")
self._on_set_partition_info_on_app_relation_data(event)
nhc_conf = self.model.config.get('nhc-conf')
if nhc_conf:
if nhc_conf != self._stored.nhc_conf:
self._stored.nhc_conf = nhc_conf
self._slurm_manager.render_nhc_config(nhc_conf)
def get_partition_name(self) -> str:
"""Return the partition_name in the slurmd relation."""
# Determine if a user-supplied partition-name config exists, if so
# ensure the partition_name is consistent with the supplied config.
# If no partition name has been specified then generate one.
partition_name = self._slurmd_peer.partition_name
partition_name_from_config = self.config.get("partition-name")
if partition_name:
if partition_name_from_config:
partition_name_from_config = partition_name_from_config.replace(
' ', '-')
if partition_name != partition_name_from_config:
self._set_partition_name(partition_name_from_config)
partition_name = partition_name_from_config
else:
logger.debug("Partition name unchanged.")
else:
logger.debug("Partition name unchanged.")
else:
partition_name = f"osd-{self.app.name}"
logger.debug(f"Partition name: {partition_name}")
self._set_partition_name(partition_name)
return partition_name
def _set_partition_name(self, name: str):
"""Set the partition_name in the slurmd relation."""
if self.model.unit.is_leader():
self._slurmd_peer.partition_name = name
def _write_munge_key_and_restart_munge(self):
logger.debug('#### slurmd charm - writting munge key')
self._slurm_manager.configure_munge_key(
self._slurmd.get_stored_munge_key())
if self._slurm_manager.restart_munged():
logger.debug("## Munge restarted succesfully")
else:
logger.error("## Unable to restart munge")
def _on_version_action(self, event):
"""Return version of installed components.
- Slurm
- munge
- NHC
- infiniband
"""
version = {}
version['slurm'] = self._slurm_manager.slurm_version()
version['munge'] = self._slurm_manager.munge_version()
version['nhc'] = self._slurm_manager.nhc_version()
version['infiniband'] = self._slurm_manager.infiniband_version()
event.set_results(version)
def _on_node_configured_action(self, event):
"""Remove node from DownNodes."""
# trigger reconfig
self._slurmd.configure_new_node()
logger.debug('### This node is not new anymore')
def _on_get_node_inventory_action(self, event):
"""Return node inventory."""
inventory = self._slurmd.node_inventory
event.set_results({'inventory': inventory})
def get_infiniband_repo(self, event):
"""Return the currently used infiniband repository."""
repo = self._slurm_manager.infiniband.repository
event.set_results({'infiniband-repo': repo})
def set_infiniband_repo(self, event):
"""Set the infiniband repository."""
repo = event.params["repo"]
logger.debug(f"#### setting custom infiniband repo: {repo}")
repo = base64.b64decode(repo).decode()
self._slurm_manager.infiniband.repository = repo
def install_infiniband(self, event):
"""Install infiniband."""
logger.debug("#### Installing Infiniband")
self._slurm_manager.infiniband.install()
event.set_results({'installation': 'Successfull. Please reboot node.'})
self.unit.status = BlockedStatus("Need reboot for Infiniband")
def uninstall_infiniband(self, event):
"""Install infiniband."""
logger.debug("#### Uninstalling Infiniband")
self._slurm_manager.infiniband.uninstall()
def start_infiniband(self, event):
"""Start Infiniband systemd service."""
logger.debug("#### Starting Infiniband service")
self._slurm_manager.infiniband.start()
def enable_infiniband(self, event):
"""Enable Infiniband systemd service."""
logger.debug("#### Enabling Infiniband service")
self._slurm_manager.infiniband.enable()
def stop_infiniband(self, event):
"""Stop Infiniband systemd service."""
logger.debug("#### Stoping Infiniband service")
self._slurm_manager.infiniband.stop()
def is_active_infiniband(self, event):
"""Check if Infiniband systemd service is arctive."""
status = self._slurm_manager.infiniband.is_active()
logger.debug(f"#### Infiniband service is-active: {status}")
event.set_results({'infiniband-is-active': status})
def nvidia_repo(self, event):
"""Set or get the used nvidia repository."""
repo = event.params.get("repo", None)
if repo:
self._slurm_manager.nvidia.repository = base64.b64decode(
repo).decode()
event.set_results(
{'nvidia-repo': self._slurm_manager.nvidia.repository})
def nvidia_package(self, event):
"""Set or get the used nvidia package."""
package = event.params.get("package", None)
if package or package == "":
# user supplied a package name -> store it
self._slurm_manager.nvidia.package = package
event.set_results(
{'nvidia-package': self._slurm_manager.nvidia.package})
def nvidia_install(self, event):
"""Install nvidia drivers."""
logger.debug("#### Installing nvidia drivers: %s",
self._slurm_manager.nvidia.package)
self._slurm_manager.nvidia.install()
event.set_results({'installation': 'Successfull. Please reboot node.'})
self.unit.status = BlockedStatus("Need reboot for nvidia")
def _on_show_nhc_config(self, event):
"""Show current nhc.conf."""
nhc_conf = self._slurm_manager.get_nhc_config()
event.set_results({"nhc.conf": nhc_conf})
def _on_set_partition_info_on_app_relation_data(self, event):
"""Set the slurm partition info on the application relation data."""
# Only the leader can set data on the relation.
if self.model.unit.is_leader():
# If the relation with slurmctld exists then set our
# partition info on the application relation data.
# This handler shouldn't fire if the relation isn't made,
# but add this extra check here just incase.
if self._slurmd.is_joined:
partition = self._assemble_partition()
if partition:
self._slurmd.set_partition_info_on_app_relation_data(
partition)
else:
event.defer()
else:
event.defer()
def _assemble_partition(self):
"""Assemble the partition info."""
partition_name = self.get_partition_name()
partition_config = self.config.get("partition-config")
partition_state = self.config.get("partition-state")
logger.debug(f"## partition_name: {partition_name}")
return {
"partition_name": partition_name,
"partition_state": partition_state,
"partition_config": partition_config,
}
@property
def hostname(self) -> str:
"""Return the hostname."""
return self._slurm_manager.hostname
@property
def cluster_name(self) -> str:
"""Return the cluster-name."""
return self._stored.cluster_name
@cluster_name.setter
def cluster_name(self, name: str):
"""Set the cluster-name."""
self._stored.cluster_name = name
class SlurmdbdCharm(CharmBase):
"""Slurmdbd Charm."""
_stored = StoredState()
on = SlurmdbdCharmEvents()
def __init__(self, *args):
"""Set the default class attributes."""
super().__init__(*args)
self._stored.set_default(db_info=dict(),
jwt_available=False,
munge_available=False,
slurm_installed=False,
cluster_name=str())
self._db = MySQLClient(self, "db")
# self._nrpe = Nrpe(self, "nrpe-external-master")
self._slurm_manager = SlurmManager(self, "slurmdbd")
self._slurmdbd = Slurmdbd(self, "slurmdbd")
self._slurmdbd_peer = SlurmdbdPeer(self, "slurmdbd-peer")
self._fluentbit = FluentbitClient(self, "fluentbit")
event_handler_bindings = {
self.on.install:
self._on_install,
self.on.upgrade_charm:
self._on_upgrade,
self.on.update_status:
self._on_update_status,
self.on.config_changed:
self._write_config_and_restart_slurmdbd,
self.on.jwt_available:
self._on_jwt_available,
self.on.munge_available:
self._on_munge_available,
self.on.write_config:
self._write_config_and_restart_slurmdbd,
self._db.on.database_available:
self._write_config_and_restart_slurmdbd,
self._db.on.database_unavailable:
self._on_db_unavailable,
self._slurmdbd_peer.on.slurmdbd_peer_available:
self._write_config_and_restart_slurmdbd,
self._slurmdbd.on.slurmctld_available:
self._on_slurmctld_available,
self._slurmdbd.on.slurmctld_unavailable:
self._on_slurmctld_unavailable,
# fluentbit
self.on["fluentbit"].relation_created:
self._on_fluentbit_relation_created,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
"""Perform installation operations for slurmdbd."""
self.unit.set_workload_version(Path("version").read_text().strip())
self.unit.status = WaitingStatus("Installing slurmdbd")
custom_repo = self.config.get("custom-slurm-repo")
successful_installation = self._slurm_manager.install(custom_repo)
if successful_installation:
self._stored.slurm_installed = True
self.unit.status = ActiveStatus("slurmdbd successfully installed")
else:
self.unit.status = BlockedStatus("Error installing slurmdbd")
event.defer()
return
self._check_status()
def _on_fluentbit_relation_created(self, event):
"""Set up Fluentbit log forwarding."""
self._configure_fluentbit()
def _configure_fluentbit(self):
logger.debug("## Configuring fluentbit")
cfg = list()
cfg.extend(self._slurm_manager.fluentbit_config_nhc)
cfg.extend(self._slurm_manager.fluentbit_config_slurm)
self._fluentbit.configure(cfg)
def _on_upgrade(self, event):
"""Perform upgrade operations."""
self.unit.set_workload_version(Path("version").read_text().strip())
def _on_update_status(self, event):
"""Handle update status."""
self._check_status()
def _on_jwt_available(self, event):
"""Retrieve and configure the jwt_rsa key."""
# jwt rsa lives in slurm spool dir, it is created when slurm is installed
if not self._stored.slurm_installed:
event.defer()
return
jwt_rsa = self._slurmdbd.get_jwt_rsa()
self._slurm_manager.configure_jwt_rsa(jwt_rsa)
self._stored.jwt_available = True
def _on_munge_available(self, event):
"""Retrieve munge key and start munged."""
# munge is installed together with slurm
if not self._stored.slurm_installed:
event.defer()
return
munge_key = self._slurmdbd.get_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
if self._slurm_manager.restart_munged():
logger.debug("## Munge restarted succesfully")
self._stored.munge_available = True
else:
logger.error("## Unable to restart munge")
self.unit.status = BlockedStatus("Error restarting munge")
event.defer()
def _on_db_unavailable(self, event):
self._stored.db_info = dict()
# TODO tell slurmctld that slurmdbd left?
self._check_status()
def _on_slurmctld_available(self, event):
self.on.jwt_available.emit()
self.on.munge_available.emit()
self.on.write_config.emit()
if self._fluentbit._relation is not None:
self._configure_fluentbit()
def _on_slurmctld_unavailable(self, event):
"""Reset state and charm status when slurmctld broken."""
self._stored.jwt_available = False
self._stored.munge_available = False
self._check_status()
def _is_leader(self):
return self.model.unit.is_leader()
def _write_config_and_restart_slurmdbd(self, event):
"""Check for prereqs before writing config/restart of slurmdbd."""
# Ensure all pre-conditions are met with _check_status(), if not
# defer the event.
if not self._check_status():
event.defer()
return
db_info = self._stored.db_info
slurmdbd_info = self._slurmdbd_peer.get_slurmdbd_info()
# settings from the config.yaml
config = {"slurmdbd_debug": self.config.get("slurmdbd-debug")}
slurmdbd_config = {
**config,
**slurmdbd_info,
**db_info,
}
self._slurm_manager.slurm_systemctl("stop")
self._slurm_manager.render_slurm_configs(slurmdbd_config)
# At this point, we must guarantee that slurmdbd is correctly
# initialized. Its startup might take a while, so we have to wait
# for it.
self._check_slurmdbd()
# Only the leader can set relation data on the application.
# Enforce that no one other then the leader trys to set
# application relation data.
if self.model.unit.is_leader():
self._slurmdbd.set_slurmdbd_info_on_app_relation_data(
slurmdbd_config, )
self._check_status()
def _check_slurmdbd(self, max_attemps=3) -> None:
"""Ensure slurmdbd is up and running."""
logger.debug("## Checking if slurmdbd is active")
for i in range(max_attemps):
if self._slurm_manager.slurm_is_active():
logger.debug("## Slurmdbd running")
break
else:
logger.warning("## Slurmdbd not running, trying to start it")
self.unit.status = WaitingStatus("Starting slurmdbd")
self._slurm_manager.restart_slurm_component()
sleep(1 + i)
if self._slurm_manager.slurm_is_active():
self._check_status()
else:
self.unit.status = BlockedStatus("Cannot start slurmdbd")
def _check_status(self) -> bool:
"""Check that we have the things we need."""
if self._slurm_manager.needs_reboot:
self.unit.status = BlockedStatus("Machine needs reboot")
return False
slurm_installed = self._stored.slurm_installed
if not slurm_installed:
self.unit.status = BlockedStatus("Error installing slurm")
return False
# we must be sure to initialize the charms correctly. Slurmdbd must
# first connect to the db to be able to connect to slurmctld correctly
slurmctld_available = (self._stored.jwt_available
and self._stored.munge_available)
statuses = {
"MySQL": {
"available": self._stored.db_info != dict(),
"joined": self._db.is_joined
},
"slurcmtld": {
"available": slurmctld_available,
"joined": self._slurmdbd.is_joined
}
}
relations_needed = list()
waiting_on = list()
for component in statuses.keys():
if not statuses[component]["joined"]:
relations_needed.append(component)
if not statuses[component]["available"]:
waiting_on.append(component)
if len(relations_needed):
msg = f"Need relations: {','.join(relations_needed)}"
self.unit.status = BlockedStatus(msg)
return False
if len(waiting_on):
msg = f"Wating on: {','.join(waiting_on)}"
self.unit.status = WaitingStatus(msg)
return False
slurmdbd_info = self._slurmdbd_peer.get_slurmdbd_info()
if not slurmdbd_info:
self.unit.status = WaitingStatus("slurmdbd starting")
return False
if not self._slurm_manager.check_munged():
self.unit.status = WaitingStatus("munged starting")
return False
self.unit.status = ActiveStatus("slurmdbd available")
return True
def get_port(self):
"""Return the port from slurm-ops-manager."""
return self._slurm_manager.port
def get_hostname(self):
"""Return the hostname from slurm-ops-manager."""
return self._slurm_manager.hostname
def set_db_info(self, db_info):
"""Set the db_info in the stored state."""
self._stored.db_info = db_info
@property
def cluster_name(self) -> str:
"""Return the cluster-name."""
return self._stored.cluster_name
@cluster_name.setter
def cluster_name(self, name: str):
"""Set the cluster-name."""
self._stored.cluster_name = name
class SlurmrestdCharm(CharmBase):
"""Operator charm responsible for lifecycle operations for slurmrestd."""
_stored = StoredState()
def __init__(self, *args):
"""Initialize charm and configure states and events to observe."""
super().__init__(*args)
self._stored.set_default(slurm_installed=False,
slurmrestd_restarted=False,
cluster_name=str())
self._slurm_manager = SlurmManager(self, "slurmrestd")
self._slurmrestd = SlurmrestdRequires(self, 'slurmrestd')
self._fluentbit = FluentbitClient(self, "fluentbit")
event_handler_bindings = {
self.on.install:
self._on_install,
self.on.upgrade_charm:
self._on_upgrade,
self.on.update_status:
self._on_update_status,
self._slurmrestd.on.config_available:
self._on_check_status_and_write_config,
self._slurmrestd.on.config_unavailable:
self._on_config_unavailable,
self._slurmrestd.on.munge_key_available:
self._on_configure_munge_key,
self._slurmrestd.on.jwt_rsa_available:
self._on_configure_jwt_rsa,
self._slurmrestd.on.restart_slurmrestd:
self._on_restart_slurmrestd,
# fluentbit
self.on["fluentbit"].relation_created:
self._on_fluentbit_relation_created,
}
for event, handler in event_handler_bindings.items():
self.framework.observe(event, handler)
def _on_install(self, event):
"""Perform installation operations for slurmrestd."""
self.unit.set_workload_version(Path("version").read_text().strip())
self.unit.status = WaitingStatus("Installing slurmrestd")
custom_repo = self.config.get("custom-slurm-repo")
successful_installation = self._slurm_manager.install(custom_repo)
if successful_installation:
self.unit.status = ActiveStatus("slurmrestd installed")
self._stored.slurm_installed = True
self._slurm_manager.start_munged()
else:
self.unit.status = BlockedStatus("Error installing slurmrestd")
event.defer()
self._check_status()
def _on_fluentbit_relation_created(self, event):
"""Set up Fluentbit log forwarding."""
self._configure_fluentbit()
def _configure_fluentbit(self):
logger.debug("## Configuring fluentbit")
cfg = list()
cfg.extend(self._slurm_manager.fluentbit_config_nhc)
cfg.extend(self._slurm_manager.fluentbit_config_slurm)
self._fluentbit.configure(cfg)
def _on_upgrade(self, event):
"""Perform upgrade operations."""
self.unit.set_workload_version(Path("version").read_text().strip())
self._check_status()
def _on_update_status(self, event):
"""Handle update status."""
self._check_status()
def _on_config_unavailable(self, event):
"""Handle the config unavailable due to relation broken."""
# when the config becomes unavailable, we have to set this flag to False,
# so the next time the config becoms avaiable, the daemon restarts
self._stored.slurmrestd_restarted = False
self._check_status()
def _on_restart_slurmrestd(self, event):
"""Resart the slurmrestd component."""
logger.debug("## _on_restart_slurmrestd")
if not self._check_status():
event.defer()
return
self._slurm_manager.restart_slurm_component()
self._stored.slurmrestd_restarted = True
def _on_configure_munge_key(self, event):
"""Configure the munge key.
1) Get the munge key from the stored state of the slurmrestd relation
2) Write the munge key to the munge key path and chmod
3) Restart munged
"""
if not self._stored.slurm_installed:
event.defer()
return
logger.debug("## configuring new munge key")
munge_key = self._slurmrestd.get_stored_munge_key()
self._slurm_manager.configure_munge_key(munge_key)
self._slurm_manager.restart_munged()
def _on_configure_jwt_rsa(self, event):
if not self._stored.slurm_installed:
event.defer()
return
logger.debug("## configuring new jwt rsa")
jwt_rsa = self._slurmrestd.get_stored_jwt_rsa()
self._slurm_manager.configure_jwt_rsa(jwt_rsa)
def _check_status(self) -> bool:
if self._slurm_manager.needs_reboot:
self.unit.status = BlockedStatus("Machine needs reboot")
return False
if not self._stored.slurm_installed:
self.unit.status = BlockedStatus("Error installing slurmrestd")
return False
# Check and see if we have what we need for operation.
if not self._slurmrestd.is_joined:
self.unit.status = BlockedStatus("Need relations: slurmctld")
return False
slurmctld_available = (self._slurmrestd.get_stored_munge_key()
and self._slurmrestd.get_stored_jwt_rsa()
and self._slurmrestd.get_stored_slurm_config())
if not slurmctld_available:
self.unit.status = WaitingStatus("Waiting on: slurmctld")
return True
self.unit.status = ActiveStatus("slurmrestd available")
return True
def _on_check_status_and_write_config(self, event):
if not self._check_status():
event.defer()
return
slurm_config = self._slurmrestd.get_stored_slurm_config()
if slurm_config:
self._slurm_manager.render_slurm_configs(slurm_config)
self.cluster_name = slurm_config.get("cluster_name")
else:
logger.error(f"## weird slurmconfig: {slurm_config}")
# Only restart slurmrestd the first time the node is brought up.
if not self._stored.slurmrestd_restarted:
self._on_restart_slurmrestd(event)
if self._fluentbit._relation is not None:
self._configure_fluentbit()
@property
def cluster_name(self) -> str:
"""Return the cluster-name."""
return self._stored.cluster_name
@cluster_name.setter
def cluster_name(self, name: str):
"""Set the cluster-name."""
self._stored.cluster_name = name