Public bug reported:
In the OVN Agent (e.g., metadata agent) within the Neutron project, event
handling for RowEvent instances assumes parallel execution of their run
methods. However, due to the design of the ovsdbapp library's RowEventHandler
class, these methods are executed sequentially [0]. This sequential execution
creates a performance bottleneck, especially in high-traffic environments.
For example, consider the PortBindingEvent class in the OVN Agent, where the
run method uses locks to safely handle concurrent execution [1]:
def run(self, event, row, old):
resync = False
with _SYNC_STATE_LOCK.read_lock():
self.log_row(row)
try:
self.agent.provision_datapath(row)
except ConfigException:
# We're now in the reader lock mode, we need to exit the context
and then use writer lock
resync = True
if resync:
self.agent.resync()
This method relies on _SYNC_STATE_LOCK.read_lock() for thread-safe execution,
which is only meaningful if run methods can execute concurrently. However,
since RowEventHandler in ovsdbapp processes these events sequentially, the
expected concurrency does not occur, resulting in underutilized resources and
increased latency.
Suggested Solution:
Parallelize run Method Execution: Modify RowEventHandler to use concurrency
mechanisms, such as a thread pool, to execute the run methods for different
events in parallel instead of sequentially.
[0]
https://github.com/openstack/ovsdbapp/blob/stable/2024.2/ovsdbapp/event.py#L177
[1]
https://github.com/openstack/neutron/blob/stable/2024.2/neutron/agent/ovn/metadata/agent.py#L135
** Affects: neutron
Importance: Undecided
Assignee: Amir Nikpour (amniik)
Status: New
** Changed in: neutron
Assignee: (unassigned) => Amir Nikpour (amniik)
--
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/2089726
Title:
Sequential Execution of run Methods in OVN Agent
Status in neutron:
New
Bug description:
In the OVN Agent (e.g., metadata agent) within the Neutron project, event
handling for RowEvent instances assumes parallel execution of their run
methods. However, due to the design of the ovsdbapp library's RowEventHandler
class, these methods are executed sequentially [0]. This sequential execution
creates a performance bottleneck, especially in high-traffic environments.
For example, consider the PortBindingEvent class in the OVN Agent, where the
run method uses locks to safely handle concurrent execution [1]:
def run(self, event, row, old):
resync = False
with _SYNC_STATE_LOCK.read_lock():
self.log_row(row)
try:
self.agent.provision_datapath(row)
except ConfigException:
# We're now in the reader lock mode, we need to exit the context
and then use writer lock
resync = True
if resync:
self.agent.resync()
This method relies on _SYNC_STATE_LOCK.read_lock() for thread-safe execution,
which is only meaningful if run methods can execute concurrently. However,
since RowEventHandler in ovsdbapp processes these events sequentially, the
expected concurrency does not occur, resulting in underutilized resources and
increased latency.
Suggested Solution:
Parallelize run Method Execution: Modify RowEventHandler to use concurrency
mechanisms, such as a thread pool, to execute the run methods for different
events in parallel instead of sequentially.
[0]
https://github.com/openstack/ovsdbapp/blob/stable/2024.2/ovsdbapp/event.py#L177
[1]
https://github.com/openstack/neutron/blob/stable/2024.2/neutron/agent/ovn/metadata/agent.py#L135
To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/2089726/+subscriptions
--
Mailing list: https://launchpad.net/~yahoo-eng-team
Post to : yahoo-eng-team@lists.launchpad.net
Unsubscribe : https://launchpad.net/~yahoo-eng-team
More help : https://help.launchpad.net/ListHelp
