--use-files should create files for outputs even if the interface is not provided

[fruffy]

Currently, --use-files seems to generate a .pcap file per interface provided via -i. The problem with this approach is that packets that are emitted on ports which are not provided via -i are not written to a file. This makes testing harder since you can not make sure that a particular packet is dropped unless you pass all 511 ports as arguments.

Is there a way to easily fix this and just generate an output file? Since no name is provided we can freely choose a basename.

[jafingerhut]

I can think of a couple of options:

• Introduce a new command line option that, when present, means "for any port number for which -i is not given, still have --use-files create a file for it". You probably want to specify a directory for those files, since there can be so many of them.

• Change the default behavior so that --use-files, with no additional command line options, creates all of those files.

I guess for v1model there is a well-defined maximum port of 511 because of the bit<9> for the egress_spec field size? I ask mainly because I half-recall someone maybe wanting to specify a drop port value larger than that at one point, but maybe that was only a control plane API port number, not a v1model data plane port number.

[antoninbas]

I think it's ok if someone wants to change that behavior: https://github.com/p4lang/behavioral-model/blob/6ee70b5eff7f510b32c074aaa4f00358f594fecb/src/bm_sim/pcap_file.cpp#L381-L392

A command-line option to toggle that behavior (and maybe specify a basename for the files) would be a good idea.

That being said, when it comes to testing simple_switch_grpc, using the gRPC service to read / write packets should pretty much always be a better idea: https://github.com/p4lang/behavioral-model/blob/main/services/p4/bm/dataplane_interface.proto. And you should always be getting the packets, no matter whether the egress port number matches a valid "front panel" port.

[fruffy]

That being said, when it comes to testing simple_switch_grpc, using the gRPC service to read / write packets should pretty much always be a better idea: https://github.com/p4lang/behavioral-model/blob/main/services/p4/bm/dataplane_interface.proto. And you should always be getting the packets, no matter whether the egress port number matches a valid "front panel" port.

This is definitely something I am considering for the PTF tests, but I have not written the PTF extension for it yet sadly.

In this particularly case I am hitting the limitations with the Thrift-based version, so it might make more sense to augment options there. Seems like it should be straightforward, thanks!

[github-actions[bot]]

This issue is stale because it has been open 180 days with no activity. Remove stale label or comment, or this will be closed in 180 days