Issue: tcp_input_lower

[zeroecco]

Hello! I finally got geth running in nanos but it looks like tcp queuing is not working? I see this error spamming:

tcp_input_lower error: incoming queue full

is there a config or a change I can try to or maybe something else I can do?

This happens on the nightly build and the normal build

[francescolavra]

You can try increasing the size of the socket receive buffer, by setting a "so_rcvbuf" value in the config.json file used by Ops when creating an image. The default buffer size is 208 KB; to change the size to, say, 512 KB, you would insert the following in your config.json file:

  "ManifestPassthrough": {
    "so_rcvbuf": "524288"
  }

[Jackson-Greene-Curtin]

You can try increasing the size of the socket receive buffer, by setting a "so_rcvbuf" value in the config.json file used by Ops when creating an image. The default buffer size is 208 KB; to change the size to, say, 512 KB, you would insert the following in your config.json file:

  "ManifestPassthrough": {
    "so_rcvbuf": "524288"
  }

Hello,

I tried this but it seems like it doesn't solve the issue. I am still getting

tcp_input_lower error: incoming queue full

error messages when at the same point as before changing this. Is this solution outdated or is there another option that is needed?

Thanks

[francescolavra]

The above solution still applies, and there are no other tunables to increase the buffer size.
If you are still getting "incoming queue full" errors after increasing so_rcvbuf, you are likely running into another limit, i.e. the number of incoming TCP packets being buffered: currently the kernel can queue up to 256 packets in a given network socket, so if you hit that limit any additional incoming packets will be dropped even if the total buffer size has not been reached. We could look into removing this limit in the kernel.