Darwin network metrics 32-bit?

[pronoiac]

Thanks for your work!

Heads up, the netdev collector for Darwin is gathering data from ifi_ibytes and ifi_obytes, and those are 32-bit integers, so they're wrapping around after 4 GB of traffic. Graphing node_network_receive_bytes and node_network_transmit_bytes results in a sawtooth pattern, and this introduces noise to irate for those metrics.

There are other resources available - netstat -I en0 -b can report larger numbers. I looked around the netstat code a little but didn't find an obvious fix for the node collector.

Thanks again, and sorry I can't offer a PR for this!

[SuperQ]

Thanks for the report, I guess we need to find a different system call to get the 64-bit data.

In the meantime, you can expand your range query to have more than 2 samples. If your scrape interval is 15s, you need at least [1m] in order to avoid counter reset anomalies.

[SuperQ]

It looks like this may be as simple as fetching data from C.struct_if_data64.

The only thing I don't know is if it will break Darwin 32-bit.

[SuperQ]

If you can, either build from the patch in the PR, or you can grab the build from our CI pipeline to test.

[pronoiac]

I grabbed the build from the CI pipeline, and compared it to an older version. It's not right, but it is different:

• node_network_receive_bytes_total -> 0 for all devices.
• node_network_receive_multicast_total is in the 4e+15-and-16 range across almost all devices, including unused ones.
• node_network_transmit_bytes_total is in the 1e+09 range for all devices, including unused ones.

[SuperQ]

Confirmed, I wasn't looking closely at the output on my laptop. 😦 I have no idea why it fails. I'll keep looking to see if I'm missing something obvious.