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 _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 test_delete_empty_parents_other_exception(self):
m_client = Mock()
m_client.delete = Mock()
m_client.delete.side_effect = etcd.EtcdValueError()
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(m_client.delete.mock_calls, [
call("foo/bar/baz/biff", dir=True, timeout=5),
])
def test_delete_empty_parents_mainline(self):
m_client = Mock()
m_client.delete = Mock()
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(m_client.delete.mock_calls, [
call("foo/bar/baz/biff", dir=True, timeout=5),
call("foo/bar/baz", dir=True, timeout=5),
call("foo/bar", dir=True, timeout=5),
])
def test_delete_empty_parents_not_found(self):
m_client = Mock()
m_client.delete = Mock()
m_client.delete.side_effect = [None, etcd.EtcdKeyNotFound(), None]
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(m_client.delete.mock_calls, [
call("foo/bar/baz/biff", dir=True, timeout=5),
call("foo/bar/baz", dir=True, timeout=5),
call("foo/bar", dir=True, timeout=5),
])
def test_delete_empty_parents_other_exception(self):
m_client = Mock()
m_client.delete = Mock()
m_client.delete.side_effect = etcd.EtcdValueError()
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(
m_client.delete.mock_calls,
[
call("foo/bar/baz/biff", dir=True, timeout=5),
]
)
def test_delete_empty_parents_mainline(self):
m_client = Mock()
m_client.delete = Mock()
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(
m_client.delete.mock_calls,
[
call("foo/bar/baz/biff", dir=True, timeout=5),
call("foo/bar/baz", dir=True, timeout=5),
call("foo/bar", dir=True, timeout=5),
]
)
def test_delete_empty_parents_not_empty(self):
m_client = Mock()
m_client.delete = Mock()
m_client.delete.side_effect = [
None,
etcd.EtcdDirNotEmpty(),
]
delete_empty_parents(m_client, "/foo/bar/baz/biff", "/foo")
self.assertEqual(
m_client.delete.mock_calls,
[
call("foo/bar/baz/biff", dir=True, timeout=5),
call("foo/bar/baz", dir=True, timeout=5),
]
)
def _write_endpoint_status_to_etcd(self, ep_id, status):
"""
Try to actually write the status dict into etcd or delete the key
if it is no longer needed.
"""
status_key = ep_id.path_for_status
if status:
_log.debug("Writing endpoint status %s = %s", ep_id, status)
self.client.set(status_key, json.dumps(status))
else:
_log.debug("Removing endpoint status %s", ep_id)
try:
self.client.delete(status_key)
except EtcdKeyNotFound:
_log.debug("Tried to delete %s but it was already gone", status_key)
# Clean up any now-empty parent directories.
delete_empty_parents(
self.client,
status_key.rsplit("/", 1)[0], # Snip off final path segment.
dir_for_felix_status(self._config.HOSTNAME),
)
def _write_endpoint_status_to_etcd(self, ep_id, status):
"""
Try to actually write the status dict into etcd or delete the key
if it is no longer needed.
"""
status_key = ep_id.path_for_status
if status:
_log.debug("Writing endpoint status %s = %s", ep_id, status)
self.client.set(status_key, json.dumps(status))
else:
_log.debug("Removing endpoint status %s", ep_id)
try:
self.client.delete(status_key)
except EtcdKeyNotFound:
_log.debug("Tried to delete %s but it was already gone",
status_key)
# Clean up any now-empty parent directories.
delete_empty_parents(
self.client,
status_key.rsplit("/", 1)[0], # Snip off final path segment.
dir_for_felix_status(self._config.HOSTNAME))
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)
