def _attempt_cleanup(self):
our_host_dir = "/".join(
[FELIX_STATUS_DIR, self._config.HOSTNAME, "workload"])
try:
# Grab all the existing status reports.
response = self.client.read(our_host_dir, recursive=True)
except EtcdKeyNotFound:
_log.info("No endpoint statuses found, nothing to clean up")
else:
# Mark all statuses we find as dirty. This will result in any
# unknown endpoints being cleaned up.
for node in response.leaves:
combined_id = get_endpoint_id_from_key(node.key)
if combined_id:
_log.debug(
"Endpoint %s removed by resync, marking "
"status key for cleanup", combined_id)
self._mark_endpoint_dirty(combined_id)
elif node.dir:
# This leaf is an empty directory, try to clean it up.
# This is safe even if another thread is adding keys back
# into the directory.
_log.debug("Found empty directory %s, cleaning up",
node.key)
delete_empty_parents(self.client, node.key, our_host_dir)
def clean_up_endpoint_statuses(self, our_endpoints_ids):
"""
Mark any endpoint status reports for non-existent endpoints
for cleanup.
:param set our_endpoints_ids: Set of endpoint IDs for endpoints on
this host.
"""
if not self._config.REPORT_ENDPOINT_STATUS:
_log.debug("Endpoint status reporting disabled, ignoring.")
return
our_host_dir = "/".join([FELIX_STATUS_DIR, self._config.HOSTNAME, "workload"])
try:
# Grab all the existing status reports.
response = self.client.read(our_host_dir, recursive=True)
except EtcdKeyNotFound:
_log.info("No endpoint statuses found, nothing to clean up")
else:
for node in response.leaves:
combined_id = get_endpoint_id_from_key(node.key)
if combined_id and combined_id not in our_endpoints_ids:
# We found an endpoint in our status reporting tree that
# wasn't in the main tree. Mark it as dirty so the status
# reporting thread will clean it up.
_log.debug("Endpoint %s removed by resync, marking " "status key for cleanup", combined_id)
self._status_reporter.mark_endpoint_dirty(combined_id, async=True)
elif node.dir:
# This leaf is an empty directory, try to clean it up.
# This is safe even if another thread is adding keys back
# into the directory.
_log.debug("Found empty directory %s, cleaning up", node.key)
delete_empty_parents(self.client, node.key, our_host_dir)
def _attempt_cleanup(self):
our_host_dir = "/".join([FELIX_STATUS_DIR, self._config.HOSTNAME,
"workload"])
try:
# Grab all the existing status reports.
response = self.client.read(our_host_dir,
recursive=True)
except EtcdKeyNotFound:
_log.info("No endpoint statuses found, nothing to clean up")
else:
# Mark all statuses we find as dirty. This will result in any
# unknown endpoints being cleaned up.
for node in response.leaves:
combined_id = get_endpoint_id_from_key(node.key)
if combined_id:
_log.debug("Endpoint %s removed by resync, marking "
"status key for cleanup",
combined_id)
self._mark_endpoint_dirty(combined_id)
elif node.dir:
# This leaf is an empty directory, try to clean it up.
# This is safe even if another thread is adding keys back
# into the directory.
_log.debug("Found empty directory %s, cleaning up",
node.key)
delete_empty_parents(self.client, node.key, our_host_dir)
def _on_snapshot_loaded(self, etcd_snapshot_response):
"""Called whenever a snapshot is loaded from etcd.
Updates the driver with the current state.
"""
LOG.info("Started processing status-reporting snapshot from etcd")
endpoints_by_host = collections.defaultdict(set)
hosts_with_live_felix = set()
# First pass: find all the Felixes that are alive.
for etcd_node in etcd_snapshot_response.leaves:
key = etcd_node.key
felix_hostname = datamodel_v1.hostname_from_status_key(key)
if felix_hostname:
# Defer to the code for handling an event.
hosts_with_live_felix.add(felix_hostname)
self._on_status_set(etcd_node, felix_hostname)
continue
# Second pass: find all the endpoints associated with a live Felix.
for etcd_node in etcd_snapshot_response.leaves:
key = etcd_node.key
endpoint_id = datamodel_v1.get_endpoint_id_from_key(key)
if endpoint_id:
if endpoint_id.host in hosts_with_live_felix:
LOG.debug("Endpoint %s is on a host with a live Felix.",
endpoint_id)
self._report_status(
endpoints_by_host,
endpoint_id,
etcd_node.value
)
else:
LOG.debug("Endpoint %s is not on a host with live Felix;"
"marking it down.",
endpoint_id)
self.calico_driver.on_port_status_changed(
endpoint_id.host,
endpoint_id.endpoint,
None,
)
continue
# Find any removed endpoints.
for host, endpoints in self._endpoints_by_host.iteritems():
current_endpoints = endpoints_by_host.get(host, set())
removed_endpoints = endpoints - current_endpoints
for endpoint_id in removed_endpoints:
LOG.debug("Endpoint %s removed by resync.")
self.calico_driver.on_port_status_changed(
host,
endpoint_id.endpoint,
None,
)
# Swap in the newly-loaded state.
self._endpoints_by_host = endpoints_by_host
LOG.info("Finished processing status-reporting snapshot from etcd")
def _on_snapshot_loaded(self, etcd_snapshot_response):
"""Called whenever a snapshot is loaded from etcd.
Updates the driver with the current state.
"""
LOG.info("Started processing status-reporting snapshot from etcd")
endpoints_by_host = collections.defaultdict(set)
hosts_with_live_felix = set()
# First pass: find all the Felixes that are alive.
for etcd_node in etcd_snapshot_response.leaves:
key = etcd_node.key
felix_hostname = datamodel_v1.hostname_from_status_key(key)
if felix_hostname:
# Defer to the code for handling an event.
hosts_with_live_felix.add(felix_hostname)
self._on_status_set(etcd_node, felix_hostname)
continue
# Second pass: find all the endpoints associated with a live Felix.
for etcd_node in etcd_snapshot_response.leaves:
key = etcd_node.key
endpoint_id = datamodel_v1.get_endpoint_id_from_key(key)
if endpoint_id:
if endpoint_id.host in hosts_with_live_felix:
LOG.debug("Endpoint %s is on a host with a live Felix.",
endpoint_id)
self._report_status(
endpoints_by_host,
endpoint_id,
etcd_node.value
)
else:
LOG.debug("Endpoint %s is not on a host with live Felix;"
"marking it down.",
endpoint_id)
self.calico_driver.on_port_status_changed(
endpoint_id.host,
endpoint_id.endpoint,
None,
)
continue
# Find any removed endpoints.
for host, endpoints in self._endpoints_by_host.iteritems():
current_endpoints = endpoints_by_host.get(host, set())
removed_endpoints = endpoints - current_endpoints
for endpoint_id in removed_endpoints:
LOG.debug("Endpoint %s removed by resync.")
self.calico_driver.on_port_status_changed(
host,
endpoint_id.endpoint,
None,
)
# Swap in the newly-loaded state.
self._endpoints_by_host = endpoints_by_host
LOG.info("Finished processing status-reporting snapshot from etcd")
def parse_if_endpoint(config, etcd_node):
combined_id = get_endpoint_id_from_key(etcd_node.key)
if combined_id:
# Got an endpoint.
if etcd_node.action == "delete":
_log.debug("Found deleted endpoint %s", combined_id)
endpoint = None
else:
endpoint = parse_endpoint(config, combined_id, etcd_node.value)
# EndpointId does the interning for us.
return combined_id, endpoint
return None, None
def parse_if_endpoint(config, etcd_node):
combined_id = get_endpoint_id_from_key(etcd_node.key)
if combined_id:
# Got an endpoint.
if etcd_node.action == "delete":
_log.debug("Found deleted endpoint %s", combined_id)
endpoint = None
else:
endpoint = parse_endpoint(config, combined_id, etcd_node.value)
# EndpointId does the interning for us.
return combined_id, endpoint
return None, None
def _on_ep_set(self, response, hostname, workload, endpoint):
"""Called when the status key for a particular endpoint is updated.
Reports the status to the driver and caches the existence of the
endpoint.
"""
ep_id = datamodel_v1.get_endpoint_id_from_key(response.key)
if not ep_id:
LOG.error(
"Failed to extract endpoint ID from: %s. Ignoring "
"update!", response.key)
return
self._report_status(self._endpoints_by_host, ep_id, response.value)
def _on_ep_set(self, response, hostname, workload, endpoint):
"""Called when the status key for a particular endpoint is updated.
Reports the status to the driver and caches the existence of the
endpoint.
"""
ep_id = datamodel_v1.get_endpoint_id_from_key(response.key)
if not ep_id:
LOG.error("Failed to extract endpoint ID from: %s. Ignoring "
"update!", response.key)
return
self._report_status(self._endpoints_by_host,
ep_id,
response.value)
def _on_ep_delete(self, response, hostname, workload, endpoint):
"""Called when the status key for an endpoint is deleted.
This typically means the endpoint has been deleted. Reports
the deletion to the driver.
"""
LOG.debug("Port %s/%s/%s deleted", hostname, workload, endpoint)
endpoint_id = datamodel_v1.get_endpoint_id_from_key(response.key)
self._endpoints_by_host[hostname].discard(endpoint_id)
if not self._endpoints_by_host[hostname]:
del self._endpoints_by_host[hostname]
self.calico_driver.on_port_status_changed(
hostname,
endpoint,
None,
)
def _on_ep_delete(self, response, hostname, workload, endpoint):
"""Called when the status key for an endpoint is deleted.
This typically means the endpoint has been deleted. Reports
the deletion to the driver.
"""
LOG.debug("Port %s/%s/%s deleted", hostname, workload, endpoint)
endpoint_id = datamodel_v1.get_endpoint_id_from_key(response.key)
self._endpoints_by_host[hostname].discard(endpoint_id)
if not self._endpoints_by_host[hostname]:
del self._endpoints_by_host[hostname]
self.calico_driver.on_port_status_changed(
hostname,
endpoint,
None,
)
def clean_up_endpoint_statuses(self, our_endpoints_ids):
"""
Mark any endpoint status reports for non-existent endpoints
for cleanup.
:param set our_endpoints_ids: Set of endpoint IDs for endpoints on
this host.
"""
if not self._config.REPORT_ENDPOINT_STATUS:
_log.debug("Endpoint status reporting disabled, ignoring.")
return
our_host_dir = "/".join(
[FELIX_STATUS_DIR, self._config.HOSTNAME, "workload"])
try:
# Grab all the existing status reports.
response = self.client.read(our_host_dir, recursive=True)
except EtcdKeyNotFound:
_log.info("No endpoint statuses found, nothing to clean up")
else:
for node in response.leaves:
combined_id = get_endpoint_id_from_key(node.key)
if combined_id and combined_id not in our_endpoints_ids:
# We found an endpoint in our status reporting tree that
# wasn't in the main tree. Mark it as dirty so the status
# reporting thread will clean it up.
_log.debug(
"Endpoint %s removed by resync, marking "
"status key for cleanup", combined_id)
self._status_reporter.mark_endpoint_dirty(combined_id,
async=True)
elif node.dir:
# This leaf is an empty directory, try to clean it up.
# This is safe even if another thread is adding keys back
# into the directory.
_log.debug("Found empty directory %s, cleaning up",
node.key)
delete_empty_parents(self.client, node.key, our_host_dir)
