Hi, For the past few months, I've noticed that my HTTP uploads from containers are really slow. Reproduction: $ cd "$(mktemp -d)" $ head --bytes=100M /dev/urandom > test.bin $ tar cf ./test.tar.gz ./test.bin $ chmod -R a+rx . $ curl --output /dev/null --progress-meter --form file=@./test.tar.gz "https://www.ctan.org/submit/validate" % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 100M 0 345 100 100M 69 20.0M 0:00:04 0:00:04 --:--:-- 23.9M $ podman run --rm --pull=newer --volume="$(realpath .):/srv/:Z" --workdir=/srv/ --network=pasta quay.io/fedora/fedora-minimal curl --output /dev/null --progress-meter --form file=@./test.tar.gz "https://www.ctan.org/submit/validate" % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 6 100M 0 0 6 6912k 0 9517 3:03:39 0:12:23 2:51:16 0 (Stopped after 10 minutes since I got sick of waiting) $ podman run --rm --pull=newer --volume="$(realpath .):/srv/:Z" --workdir=/srv/ --network=host quay.io/fedora/fedora-minimal curl --output /dev/null --progress-meter --form file=@./test.tar.gz "https://www.ctan.org/submit/validate" % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 100M 0 345 100 100M 64 18.7M 0:00:05 0:00:05 --:--:-- 23.8M Some comments about the above: - "www.ctan.org" is a web server running Apache; it isn't my server, but it's where I first noticed this issue, and I have permission from its owners to use it for testing. - The upload speed is constant on the host and with --network=host, but with --network=pasta, it starts high-ish, but quickly and continuously decreases. But the "bytes transferred" number keeps increasing, so the upload hasn't stalled out completely. - I've also had problems with uploads to GitHub from a custom Go binary---it takes ~3 minutes to upload ~600MB, split across 5 files, which is much slower than I would expect, but still faster than the curl/CTAN upload speeds. So I don't think that this is a bug with curl or this specific web server. - The host that I ran this from is a VM in a datacenter, supposedly with a symmetrical 1Gb/s connection. Pinging github.com shows an RTT of 2ms, and pinging www.ctan.org shows an RTT of 100ms. - This problem seems to have started a few months ago. I don't have an exact date (sorry), but it seems to have been fine before then. - The host is running Fedora 43 IoT on x86_64, and I'm using the following software versions: $ rpm --query --queryformat='%{NAME}\t%{VERSION}\n' passt podman containers-common netavark | sort | column --table containers-common 0.64.2 netavark 1.16.1 passt 0^20250919.g623dbf6 podman 5.6.2 I had the same issue when the host/containers were using Fedora 42. - The host has fairly little CPU/network load when I ran these tests, and pasta's CPU usage was low enough that it didn't show up on the first page of htop. Does anybody have any advice on troubleshooting this? I'm willing to test with other servers/commands, compile/test any patches, or provide any other information that might be helpful. Thanks, -- Max