NodeScanner fails with "No buffer space available" on slow CAN networks

[mcba]

I have a slow (20000 bits/sec) CAN bus. When using NodeScanner it tries to send packets out too quickly for the bus master (a CANable USB interface), and a buffer somewhere fills up after 15 packets. The error message is:

can.exceptions.CanOperationError: Failed to transmit: No buffer space available [Error Code 105]

A simple, ugly, kludge to work around this is to add a time.sleep(0.1) in the search function in network.py.

Regards,
Michael

[acolomb]

What CAN interface type are you using (setting for python-can), and which operating system?

[mcba]

I'm using socketcan on Linux Debian 14 testing.

[acolomb]

Try this: ip link set can0 txqueuelen 1000

It can be configured to stick on the system using a systemd unit of type "link". I have this configuration snippet on my devices for example in /etc/systemd/network/70-can-usb.link:

[Match]
Kind=can

[Link]
TransmitQueueLength=1000

You need to be using systemd-networkd though for this to take effect. It works fine in parallel to the default NetworkManager (at least on my Ubuntu and Raspberry Pi OS machines): sudo systemctl enable --now systemd-networkd.

[mcba]

I tried the "ip link" command, and it stopped the "No buffer space error". It now scans from 0x601 up to 0x67c where it dies with:

can.exceptions.CanOperationError: Transmit buffer full

in send at line 881 of socketcan.py

[acolomb]

That still looks like the some buffer is filling up. Tried increasing the queue length even further?

You may have better luck asking about this in the python-can project, where the interface code is maintained. Try to get a debug log of what messages are being sent, with timing info. This seems not to be a problem on the CANopen layer, the LSS code just happens to send a burst of messages too quickly for your interface's possible throughput.

[mcba]

I tried 2000 as opposed to 1000, with no change. It looks like I'm hitting another buffer limit. Thanks for your advice.

After a steep learning curve, I'm enjoying using canopen-python.