exception - queue full in nlsocket - async_recv()

[laplasz]

example code:

with IPRoute() as ipr:
# With IPRoute objects you have to call bind() manually
ipr.bind(async_cache=True)
while True:
try:
for message in ipr.get():
<....>
except OSError as e:
raise e

callstack:
for message in ipr.get():
File "/usr/local/lib/python3.9/site-packages/pr2modules/netlink/nlsocket.py", line 394, in get
return tuple(self._genlm_get(*argv, **kwarg))
File "/usr/local/lib/python3.9/site-packages/pr2modules/netlink/nlsocket.py", line 788, in get
data = self.recv_ft(bufsize)
File "/usr/local/lib/python3.9/site-packages/pr2modules/netlink/nlsocket.py", line 1067, in recv_plugin
raise data_in
File "/usr/local/lib/python3.9/site-packages/pr2modules/netlink/nlsocket.py", line 580, in async_recv
self.buffer_queue.put_nowait(data)
File "/usr/lib64/python3.9/queue.py", line 191, in put_nowait
return self.put(item, block=False)
File "/usr/lib64/python3.9/queue.py", line 137, in put
raise Full

[svinota]

Looks like the parser doesn't manage to handle the packet flow.

Possible solutions:

1. increase the buffer queue

from pyroute2 import config, IPRoute
config.async_qsize = X
ipr = IPRoute()

# or
from pyroute2 import IPRoute
ipr = IPRoute(async_qsize=X)

Where X is the queue max size, by default 4096. The socket uses 64K buffers for each recv_into(), so consider memory usage as well.

2. use a custom parser

IPRoute objects have .marshal to run parsers on incoming messages.

IPRoute.marshal.msg_map = {msg_type: parser}
IPRoute.marshal.seq_map = {sequence_number: parser}

More on parsers and how to run own implementations: https://docs.pyroute2.org/parser.html

[LuiGuest]

Looks like the parser doesn't manage to handle the packet flow.

Possible solutions:

1. increase the buffer queue

from pyroute2 import config, IPRoute
config.async_qsize = X
ipr = IPRoute()

# or
from pyroute2 import IPRoute
ipr = IPRoute(async_qsize=X)

Where X is the queue max size, by default 4096. The socket uses 64K buffers for each recv_into(), so consider memory usage as well.

2. use a custom parser

IPRoute objects have .marshal to run parsers on incoming messages.

IPRoute.marshal.msg_map = {msg_type: parser}
IPRoute.marshal.seq_map = {sequence_number: parser}

More on parsers and how to run own implementations: https://docs.pyroute2.org/parser.html

As I understand this could happen due to temporal event storm, in which case the parser will lose pace against the socket reader, and the queue will get full.
Is there a way to somehow instruct the API to restrict which events it should deliver via the socket?

[svinota]

@LuiGuest there are three ways setting up that:

1. Using groups -- ipr.bind(groups=...), see pyroute2.netlink.rtnl.RTMGRP_* (becomes deprecated in the Linux kernel, but still used by default in the library)
2. Using membership -- ipr.add_membership(...) / ipr.drop_membership(...), see pyroute2.netlink.rtnl.RTNLGRP_*
3. Using custom parsers and assigning to particular message types (ipr.marshal.msg_map) or sequence numbers (ipr.marshal.seq_map)

...

1. and 2. will instruct the kernel not to send some message types, while 3. makes pyroute2 to drop received messages w/o parsing them (by the default parser)

[LuiGuest]

1. RTMGRP_

Thank You a lot for your help!