Hi Ravikiran,
> On 13 Sep 2021, at 13:00, RaviKiran Veldanda <ravi.jup...@gmail.com> wrote: > > This definition is way too vague to tell anything. > > Hi Andrew, > Please find my answers below > What is the setup ? (Cores and their config) > >>>> Its 1U server, with 72cores available and 5cores for VPP. With 100G NIC. > >>>> VAPI with 512 request and response size. > > What are the exact rules being pushed ? > >>> ACL Permit rules pushed using VPP API. > set acl-plugin acl permit dst 172.172.0.0/24 > abf policy add id 0 acl 0 via 192.168.1.5 memif1/0 > abf attach ip4 policy 0 BondedEthernet0.1100 > Same kind of rules for 100s of subnets > Every few minutes, the rules will be added and deleted. What is sitting on the other end of the memif interfaces? Are those applications experiencing any changes / restarts throughout, or are they permanently run so no transient events in memif ? > > What are the type of traffic that is passing ? > >> No much traffic passing. I am testing only VAPI for ACL rules. Could you see if it is still reproducible if you disconnect the other components altogether ? (I am trying to see how reproducible it is in an isolated case, and if it is then hopefully it is something that could be made into a standalone harness that you might be able to share so i could try reproducing it locally. > > If you see the VPP stuck - what does looking at it with GDB tell about the > state of it ? > >> I missed this point... I can provide next time when it hangs. > > Is this reliably reproducible in the lab ? > >> Yes, after every several hours of addition and deletion its reproducible. Cool, thanks a lot! > > Please let me know if you need anything more. Another question that I forgot - which way of interacting are you using for the API ? Shared memory or the Unix socket ? My line of thinking involves narrowing this issue into the minimal test case, as I described above… primarily I am curious about isolating away all the potential memif transitions and the physical interface traffic, to clarify if this can be purely some acl+abf api interaction artifact or there is more to the story… P.s. The harness (when we narrow down the problem domain) I am planning to build using the current work-in-progress Rust bindings (cf: https://github.com/ayourtch/latest-vpp-api-playground) - but if you have something that you can share that uses eg govpp, will be interesting to use that. —a > //Ravi > >
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#20117): https://lists.fd.io/g/vpp-dev/message/20117 Mute This Topic: https://lists.fd.io/mt/85566777/21656 Mute #vapi:https://lists.fd.io/g/vpp-dev/mutehashtag/vapi Mute #acl:https://lists.fd.io/g/vpp-dev/mutehashtag/acl Mute #abf:https://lists.fd.io/g/vpp-dev/mutehashtag/abf Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
