Hi Rain, Inline
> On Aug 6, 2025, at 1:04 AM, rain_alex via lists.fd.io > <wangdaobei=gmail....@lists.fd.io> wrote: > > Dear community developers: > > I use vnet_connect function to initiate a large number of connections to the > nginx server. > I have calculated the latency and found that there are a large number of syn > retransmissions. > After capturing the packet, I discovered an issue: when the VPP accesses it > again using the same source port after 10 seconds, nginx mistakenly considers > it to be a packet from the previous connection and sends back an incorrect > ACK message. This doesn't always happen, the probability is about one in ten. Is the source port pinned by your app or allocated by the transport layer? > I switched to using Linux to write the client and also made a large number of > connections. Sometimes, it would use the same source port every 2 seconds, > but there were no problems. > The nginx server has been configured 2msl = 1. Both sides are Ubuntu. Interesting. You expect nginx to have closed the connections and for linux stack to be now in time-wait state? If yes, maybe linux rejects the connection because of the initial sequence number. > I don't even know if this is a VPP issue. > The captured package was included with the problem. IP "10.1.1.181" is nginx > server. > If any experts can answer my questions, I would be extremely grateful. Yes, it might be the above. That is the syn that’s rejected has a seq number lower than the seq number of the connection in time-wait. At this point we don’t keep track of the seq number used in the previous instances of the 5-tuple and apparently the random generation that we’re using currently might not be good enough for this use case. Definitely needs improvement. Regards, Florin > > Best Regards, > Rain > > > > > <ser_1.pcap> > >
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#26243): https://lists.fd.io/g/vpp-dev/message/26243 Mute This Topic: https://lists.fd.io/mt/114562621/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
