Hi all, I was trying to clone packets for some experiment. I used the clone3 feature in ingress block and used the I2E clone type. When I debugged the s1.log file, I found that packets are not getting cloned. I sent ICMP packets for testing. *Some Evidence*: 1. standard_metadata.instance_type should be of value 5 but instead value is 0. There is no instance_type with value 5. 2. Wireshark and tcpdump on mininet host network interfaces show no duplicate packets. 3. log_msg function printing content only once in egress block per packet. Expected behavior is to get content printed twice per packet. First time for original and next time for the cloned packet. How can I debug this issue? Attaching the log file for reference. Thanks Sahil Gupta

It seems possible that you have not configured an egress port for the clone/mirror session that you are using in the P4 program? When simple_switch starts, all clone sessions are initialized to a state where they clone packets to _nowhere_, so the clones are effectively dropped. You have a choice in simple_switch of configuring a clone session to send a packet to a single egress port, or to a multicast group id. If you do so, then in the log you would see not only the line: [02:39:41.189] [bmv2] [D] [thread 12629] [97.0] [cxt 0] Cloning packet at ingress which is printed from this line of C++ code: https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L539 you would also see either a line like this, if the clone session is configured for multicast: https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L565 or if the clone session is configured for a single egress port, this line would be executed: https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L569 How does one configure a clone/mirror session? It depends upon the control plane API you are using. If you are using simple_switch_CLI to do it manually, there are commands to do so documented here: https://github.com/p4lang/behavioral-model/blob/master/docs/runtime_CLI.md#mirroring_add-mirroring_add_mc-mirroring_delete-mirroring_get If you are using the P4Runtime API or Thrift API, they have their own message formats for doing so. Andy On Mon, Jan 4, 2021 at 11:56 PM Sahil Gupta <sg5414@rit.edu> wrote: > Hi all, > > I was trying to clone packets for some experiment. > I used the clone3 feature in ingress block and used the I2E clone type. > When I debugged the s1.log file, I found that packets are not getting > cloned. > I sent ICMP packets for testing. > > *Some Evidence*: > 1. standard_metadata.instance_type should be of value 5 but instead value > is 0. > There is no instance_type with value 5. > 2. Wireshark and tcpdump on mininet host network interfaces show no > duplicate packets. > 3. log_msg function printing content only once in egress block per packet. > Expected behavior is to get content printed twice per packet. First time > for original and next time for the cloned packet. > > How can I debug this issue? > Attaching the log file for reference. > > Thanks > Sahil Gupta > > _______________________________________________ > P4-dev mailing list > P4-dev@lists.p4.org > http://lists.p4.org/mailman/listinfo/p4-dev_lists.p4.org >

Thanks Andy for detailed insight!! I will check the configuration of the clone session in my experimental setup. On Tue, Jan 5, 2021 at 10:01 AM Andy Fingerhut <andy.fingerhut@gmail.com> wrote: > It seems possible that you have not configured an egress port for the > clone/mirror session that you are using in the P4 program? When > simple_switch starts, all clone sessions are initialized to a state where > they clone packets to _nowhere_, so the clones are effectively dropped. > You have a choice in simple_switch of configuring a clone session to send a > packet to a single egress port, or to a multicast group id. If you do so, > then in the log you would see not only the line: > > [02:39:41.189] [bmv2] [D] [thread 12629] [97.0] [cxt 0] Cloning packet at > ingress > > which is printed from this line of C++ code: > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L539 > > you would also see either a line like this, if the clone session is > configured for multicast: > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L565 > > or if the clone session is configured for a single egress port, this line > would be executed: > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L569 > > How does one configure a clone/mirror session? It depends upon the > control plane API you are using. If you are using simple_switch_CLI to do > it manually, there are commands to do so documented here: > https://github.com/p4lang/behavioral-model/blob/master/docs/runtime_CLI.md#mirroring_add-mirroring_add_mc-mirroring_delete-mirroring_get > > If you are using the P4Runtime API or Thrift API, they have their own > message formats for doing so. > > Andy > > On Mon, Jan 4, 2021 at 11:56 PM Sahil Gupta <sg5414@rit.edu> wrote: > >> Hi all, >> >> I was trying to clone packets for some experiment. >> I used the clone3 feature in ingress block and used the I2E clone type. >> When I debugged the s1.log file, I found that packets are not getting >> cloned. >> I sent ICMP packets for testing. >> >> *Some Evidence*: >> 1. standard_metadata.instance_type should be of value 5 but instead value >> is 0. >> There is no instance_type with value 5. >> 2. Wireshark and tcpdump on mininet host network interfaces show no >> duplicate packets. >> 3. log_msg function printing content only once in egress block per packet. >> Expected behavior is to get content printed twice per packet. First time >> for original and next time for the cloned packet. >> >> How can I debug this issue? >> Attaching the log file for reference. >> >> Thanks >> Sahil Gupta >> >> _______________________________________________ >> P4-dev mailing list >> P4-dev@lists.p4.org >> http://lists.p4.org/mailman/listinfo/p4-dev_lists.p4.org >> >

Hi Sahil, in case you'd like a BMv2-based example of the feature in use, here's some code that uses cloning as part of a feedback loop between dataplanes: https://github.com/eniac/Flightplan/blob/master/Wharf/Sources/FPRuntime.p4#L172 And this is where resources for cloning are set up via the control plane -- Andy's earlier reply has key background details on this angle: https://github.com/eniac/Flightplan/blob/b9d558d23359629e87807e4e868ef4673b6937f0/Wharf/fpctl.py#L188 The above code might require some effort at your end to understand or run it, but we've deployed on several research systems and it works. Hope that can be helpful to you, alongside the info you've gathered so far. Nik -- www.seas.upenn.edu/~nsultana On Tue, 05 Jan 2021, Sahil Gupta wrote: > Thanks Andy for detailed insight!! > I will check the configuration of the clone session in my experimental > setup. > > On Tue, Jan 5, 2021 at 10:01 AM Andy Fingerhut <andy.fingerhut@gmail.com> > wrote: > > > It seems possible that you have not configured an egress port for the > > clone/mirror session that you are using in the P4 program? When > > simple_switch starts, all clone sessions are initialized to a state where > > they clone packets to _nowhere_, so the clones are effectively dropped. > > You have a choice in simple_switch of configuring a clone session to send a > > packet to a single egress port, or to a multicast group id. If you do so, > > then in the log you would see not only the line: > > > > [02:39:41.189] [bmv2] [D] [thread 12629] [97.0] [cxt 0] Cloning packet at > > ingress > > > > which is printed from this line of C++ code: > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L539 > > > > you would also see either a line like this, if the clone session is > > configured for multicast: > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L565 > > > > or if the clone session is configured for a single egress port, this line > > would be executed: > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L569 > > > > How does one configure a clone/mirror session? It depends upon the > > control plane API you are using. If you are using simple_switch_CLI to do > > it manually, there are commands to do so documented here: > > https://github.com/p4lang/behavioral-model/blob/master/docs/runtime_CLI.md#mirroring_add-mirroring_add_mc-mirroring_delete-mirroring_get > > > > If you are using the P4Runtime API or Thrift API, they have their own > > message formats for doing so. > > > > Andy > > > > On Mon, Jan 4, 2021 at 11:56 PM Sahil Gupta <sg5414@rit.edu> wrote: > > > >> Hi all, > >> > >> I was trying to clone packets for some experiment. > >> I used the clone3 feature in ingress block and used the I2E clone type. > >> When I debugged the s1.log file, I found that packets are not getting > >> cloned. > >> I sent ICMP packets for testing. > >> > >> *Some Evidence*: > >> 1. standard_metadata.instance_type should be of value 5 but instead value > >> is 0. > >> There is no instance_type with value 5. > >> 2. Wireshark and tcpdump on mininet host network interfaces show no > >> duplicate packets. > >> 3. log_msg function printing content only once in egress block per packet. > >> Expected behavior is to get content printed twice per packet. First time > >> for original and next time for the cloned packet. > >> > >> How can I debug this issue? > >> Attaching the log file for reference. > >> > >> Thanks > >> Sahil Gupta

Thanks Nik. I will look into this as well. ☺ On Tue, Jan 5, 2021 at 11:47 AM Nik Sultana <nsultana@seas.upenn.edu> wrote: > Hi Sahil, in case you'd like a BMv2-based example of the feature in use, > here's > some code that uses cloning as part of a feedback loop between dataplanes: > > https://github.com/eniac/Flightplan/blob/master/Wharf/Sources/FPRuntime.p4#L172 > > And this is where resources for cloning are set up via the control plane > -- Andy's earlier reply has key background details on this angle: > > https://github.com/eniac/Flightplan/blob/b9d558d23359629e87807e4e868ef4673b6937f0/Wharf/fpctl.py#L188 > > The above code might require some effort at your end to understand or > run it, but we've deployed on several research systems and it works. > > Hope that can be helpful to you, alongside the info you've gathered so far. > > Nik > > -- > www.seas.upenn.edu/~nsultana > > On Tue, 05 Jan 2021, Sahil Gupta wrote: > > > Thanks Andy for detailed insight!! > > I will check the configuration of the clone session in my experimental > > setup. > > > > On Tue, Jan 5, 2021 at 10:01 AM Andy Fingerhut <andy.fingerhut@gmail.com > > > > wrote: > > > > > It seems possible that you have not configured an egress port for the > > > clone/mirror session that you are using in the P4 program? When > > > simple_switch starts, all clone sessions are initialized to a state > where > > > they clone packets to _nowhere_, so the clones are effectively dropped. > > > You have a choice in simple_switch of configuring a clone session to > send a > > > packet to a single egress port, or to a multicast group id. If you do > so, > > > then in the log you would see not only the line: > > > > > > [02:39:41.189] [bmv2] [D] [thread 12629] [97.0] [cxt 0] Cloning packet > at > > > ingress > > > > > > which is printed from this line of C++ code: > > > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L539 > > > > > > you would also see either a line like this, if the clone session is > > > configured for multicast: > > > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L565 > > > > > > or if the clone session is configured for a single egress port, this > line > > > would be executed: > > > > https://github.com/p4lang/behavioral-model/blob/master/targets/simple_switch/simple_switch.cpp#L569 > > > > > > How does one configure a clone/mirror session? It depends upon the > > > control plane API you are using. If you are using simple_switch_CLI > to do > > > it manually, there are commands to do so documented here: > > > > https://github.com/p4lang/behavioral-model/blob/master/docs/runtime_CLI.md#mirroring_add-mirroring_add_mc-mirroring_delete-mirroring_get > > > > > > If you are using the P4Runtime API or Thrift API, they have their own > > > message formats for doing so. > > > > > > Andy > > > > > > On Mon, Jan 4, 2021 at 11:56 PM Sahil Gupta <sg5414@rit.edu> wrote: > > > > > >> Hi all, > > >> > > >> I was trying to clone packets for some experiment. > > >> I used the clone3 feature in ingress block and used the I2E clone > type. > > >> When I debugged the s1.log file, I found that packets are not getting > > >> cloned. > > >> I sent ICMP packets for testing. > > >> > > >> *Some Evidence*: > > >> 1. standard_metadata.instance_type should be of value 5 but instead > value > > >> is 0. > > >> There is no instance_type with value 5. > > >> 2. Wireshark and tcpdump on mininet host network interfaces show no > > >> duplicate packets. > > >> 3. log_msg function printing content only once in egress block per > packet. > > >> Expected behavior is to get content printed twice per packet. First > time > > >> for original and next time for the cloned packet. > > >> > > >> How can I debug this issue? > > >> Attaching the log file for reference. > > >> > > >> Thanks > > >> Sahil Gupta >

I built my program on top of the Miss Carolina Fernández program. It worked! I will look into the original problem with my code as suggested by Andy latter and report back to the community if I found something new. Thanks for the help guys! 😊 Reference: [1] https://github.com/CarolinaFernandez/p4-tutorials/tree/master/exercises/clone