Control: reassign -1 src:udptunnel Hi Marcos!
On Mon, 10 Mar 2025 16:14:02 -0300 Marcos Talau <ta...@debian.org> wrote: > Control: reassign 1094138 nmap > > Hi there! > > I ran the udptunnel test in a s390x testbed without udptunnel, I only used > ncat, and sometimes, it fails. > > I write some scripts [1] to help with the ncat tests. I ran for 100 times > the script "test-ncat-udp-series.sh", resulting in 10k tests. > [1] https://people.debian.org/~talau/tmp/test-ncat-udp > > 10k tests (using "test-ncat-udp-series.sh") results in a porterbox s390x: > - Fail: Avg: 3.01%, StdDev: 2.15636 > > 10k tests (using "test-ncat-udp-series.sh") results in a porterbox amd64: > - Fail: Avg: 0.12%, StdDev: 0.381576 > > After the tests, I decided to make tests changing ncat client to socat client, > scripts (test-ncat-udp_socat-client-series.sh, test-ncat-udp_socat-client.sh). > Below the results: > > 10k tests (using "test-ncat-udp_socat-client-series.sh") results in a > porterbox s390x: > - Fail: Avg: 0.0%, StdDev: 0 > > 10k tests (using "test-ncat-udp_socat-client-series.sh") results in a > porterbox amd64: > - Fail: Avg: 0.0%, StdDev: 0 > > Analyzing the results, it can be seen that communication using ncat-client and > ncat-server presents failures. By changing ncat-client to socat, > communications > no longer present failures. This indicates to me that ncat (in client mode) > has > some bug. I cannot confirm you findings. In a VM on my local amd64 machine I see or don't see failures depending on the number of virtual CPUs when running either of your scripts. I also noticed that one cycle of test- ncat-udp_socat-client-series.sh takes more time, about 150%, than a cycle of test-ncat-udp-series.sh. That makes me assume that socat mitigates the problem in certain environments by spending more time internally. The measure which solves the issue for me is to insert a short delay after 'transmit_upd_data'. This seems reasonable as one should not expect data can be sent, received and written to a file in almost no time. Here's what I changed in both of your scripts: $ diff -u3 orig/test-ncat-udp_socat-client.sh test-ncat-udp_socat-client.sh --- orig/test-ncat-udp_socat-client.sh 2025-03-10 14:06:55.000000000 +0000 +++ test-ncat-udp_socat-client.sh 2025-03-23 22:07:14.223161539 +0000 @@ -63,6 +63,7 @@ test_communication() { transmit_udp_data + sleep 0.05 if ! received_udp_data; then echo "ncat UDP communication failed" $ diff -u3 orig/test-ncat-udp.sh test-ncat-udp.sh --- orig/test-ncat-udp.sh 2025-03-10 14:06:56.000000000 +0000 +++ test-ncat-udp.sh 2025-03-23 22:07:22.827289754 +0000 @@ -63,6 +63,7 @@ test_communication() { transmit_udp_data + sleep 0.05 if ! received_udp_data; then echo "ncat UDP communication failed" With these changes applied transmission failures vanish in my setup. I suggest you try this in your environment(s), potentially with an increased delay. Cheers, Sven -- GPG Fingerprint 3DF5 E8AA 43FC 9FDF D086 F195 ADF5 0EDA F8AD D585
signature.asc
Description: This is a digitally signed message part
