Hi Manuel, Thanks for hte info. I enabled the vxlan workaround routes and attempted the curl again. I saw 2 packets hit the vxlan tool on the VNFs, but curl failed and the rest of the packets did not make it through. Updated gist with flows:
https://gist.github.com/trozet/d6547cdc416a27ef7cf5b6c6b91db5cd ofctl Port mapping: compute1: port 4) tap43c21e37 - TestVNF4 (first VNF in chain) port 6) tape118a - http client compute0: port 6) tap867682340-80 - TestVNF5 (second VNF in chain) port 7) - HTTP server Main flow here to look at here is the classifier: cookie=0x1110010001610255, duration=78031.957s, table=11, n_packets=27, n_bytes=1998, tcp,reg0=0x1,tp_src=2000,tp_dst=80 actions=move:NXM_NX_TUN_ID[0..31]->NXM_NX_NSH_C2[],push_nsh,load:0x1->NXM_NX_NSH_MDTYPE[],load:0x3->NXM_NX_NSH_NP[],load:0xc000020a->NXM_NX_NSH_C1[],load:0xa1->NXM_NX_NSP[0..23],load:0xff->NXM_NX_NSI[],load:0xc0000203->NXM_NX_TUN_IPV4_DST[],load:0xa1->NXM_NX_TUN_ID[0..31],output:5 It looks like it is skipping the first SFF. I think that should send to the SFF (compute1, same compute, so maybe just resubmit or goto table) and then progress into SFC tables. Sam, can you comment here? Tim Rozet Red Hat SDN Team ----- Original Message ----- From: "Manuel Buil" <[email protected]> To: "Tim Rozet" <[email protected]>, "Yi Y Yang" <[email protected]>, "Sam Hague" <[email protected]> Cc: "Brady Allen Johnson" <[email protected]>, "OPNFV Tech" <[email protected]> Sent: Tuesday, August 16, 2016 3:40:49 AM Subject: RE: [SFC] multiple compute nodes with Tacker Hi Tim, You still need to do the vxlan workaround. OVS 2.5.90 supports two alternatives: eth+nsh or vxlan-gpe+eth+nsh between SF <----> SFF. However, ODL is currently not capable to render the first one, so we need to do the vxlan-gpe+eth+nsh and thus, the workaround. Brady started to implement the support for eth+nsh but he did not manage to do it before his vacations. Hopefully, we'll get that in September. Regards, Manuel -----Original Message----- From: Tim Rozet [mailto:[email protected]] Sent: Monday, August 15, 2016 11:39 PM To: Manuel Buil; Yi Y Yang; Sam Hague Cc: Brady Allen Johnson; OPNFV Tech Subject: [SFC] multiple compute nodes with Tacker Expanding to larger audience... I tested out Apex today with the ODL build Brady provided. I found there were a few updates needed in Tacker and pushed a patch to the colorado branch: https://github.com/trozet/tacker/commit/d876584242781726d2e4403e69b9d15c08765625 I then created the following nova instances/tacker VNFs: https://goo.gl/photos/F59h6V31fcQSzVFt6 The chain I created was this: testVNF4->testVNF5 The curl to the http server failed (typical classifier matching on source port 2000, dest port 80), and I don't see any packets hitting the vxlan tool on either VNF. However, I do see packets hitting the classifier. Flow gist: https://gist.github.com/trozet/3161092c62fdc6582508e7779e445bf1 I did not use the VXLAN workaround (aka the kernel host route). I thought this was not required anymore with OVS 2.5.90. Yi can you comment here? I haven't had a chance to dig into the flows yet, but if anyone sees the root cause of the problem, let me know. Thanks, Tim Rozet Red Hat SDN Team ----- Original Message ----- From: "Manuel Buil" <[email protected]> To: "Tim Rozet" <[email protected]> Cc: "Brady Allen Johnson" <[email protected]> Sent: Monday, August 15, 2016 12:27:23 PM Subject: RE: Creating symmetric chains with tacker Ok! I did not have time to try that as we are encountering some bugs in Netvirt and I am preparing the demo for Seattle. Please update us if you make it work with several compute nodes :). Regards, Manuel -----Original Message----- From: Tim Rozet [mailto:[email protected]] Sent: Monday, August 15, 2016 3:00 PM To: Manuel Buil Cc: Brady Allen Johnson Subject: Re: Creating symmetric chains with tacker Hi Manuel, Yeah I don't think I ever accounted for that. We would need to look at the driver call when ODL creates the reverse chain, and instantiate 2 chains in tacker to do that. I just got my setup to work with the ODL Brady provided and have a clean deployment. I was going to start looking at using 2 compute nodes today and coming up with a fix for that one. Will respond to this email when I have more info/fixes. Tim Rozet Red Hat SDN Team ----- Original Message ----- From: "Manuel Buil" <[email protected]> To: "Tim Rozet" <[email protected]>, "Brady Allen Johnson" <[email protected]> Sent: Monday, August 15, 2016 8:22:55 AM Subject: RE: Creating symmetric chains with tacker Hey Tim, I tried today and it works :). However, it is impossible to create a classification rule on the second chain which is created (the symmetric chain), right? Thanks, Manuel -----Original Message----- From: Tim Rozet [mailto:[email protected]] Sent: Tuesday, August 09, 2016 5:25 PM To: Brady Allen Johnson Cc: Manuel Buil Subject: Re: Creating symmetric chains with tacker Hi Guys, Simply invoke --symmetrical True when creating a chain: https://github.com/trozet/python-tackerclient/blob/SFC_refactor/tackerclient/tacker/v1_0/sfc/sfc.py#L41 Tim Rozet Red Hat SDN Team ----- Original Message ----- From: "Brady Allen Johnson" <[email protected]> To: "Manuel Buil" <[email protected]>, "Tim Rozet" <[email protected]> Sent: Tuesday, August 9, 2016 7:58:42 AM Subject: Re: Creating symmetric chains with tacker Tim, Just a bit of context: to create symmetric service chains, there is a boolean field in the SFC, SFP, and RSP called "is_symmetric" (or something very similar) that would need to be set to true. Im pretty sure now you either dont set it, or explicitly set it to false. When the field is true, then ODL SFC will create a second, symmetric service chain. Thanks, Brady On 09/08/16 11:21, Manuel Buil wrote: > > Hi Tim, > > Is it possible to create symmetric chains in ODL with a tacker > command? I cannot find the command to do so L. > > Thanks, > > Manuel > _______________________________________________ opnfv-tech-discuss mailing list [email protected] https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss
