I am debugging a strange issue, and I'm not quite sure whether it's a client or server problem:
- On a specific machine (armv5)
git fetch https://codeberg.org/... sometimes gets the server certificate for *.codeberg.page, and then the TLS client library (GnuTLS) complains about it and the connection isn't established (verbose log here)
- DNS doesn't seem to be the issue here, the IP address is always resolved the same (either 217.197.91.145 or 2001:67c:1401:20f0::1)
- The error seems like it would be SNI-related but SNI looks fine in the packet capture, as far as I can tell
- The likelihood of the error is close to 100% inside a podman container, and closer to
(削除) 10% (削除ここまで) 25% outside
(削除) I don't know whether the error is triggered by GnuTLS's randomization of the order of TLS extensions (削除ここまで) Data analysis shows that this seems to be independent of the order of TLS extensions added by GnuTLS
I'd be happy to test suggestions that might bring us closer to the root cause, because it's quite a strange issue.
### Comment
I am debugging a strange issue, and I'm not quite sure whether it's a client or server problem:
- On a specific machine (armv5) `git fetch https://codeberg.org/...` sometimes gets the server certificate for `*.codeberg.page`, and then the TLS client library (GnuTLS) complains about it and the connection isn't established ([verbose log here](https://codeberg.org/neuschaefer/git-host-bug/src/branch/main/fail-verbose.txt))
- DNS doesn't seem to be the issue here, the IP address is always resolved the same (either 217.197.91.145 or 2001:67c:1401:20f0::1)
- The error seems like it would be SNI-related but SNI looks fine in the [packet capture](https://codeberg.org/neuschaefer/git-host-bug/src/branch/main/codeberg.pcap), as far as I can tell
- The likelihood of the error is close to 100% inside a podman container, and closer to ~~10%~~ 25% outside
- ~~I don't know whether the error is triggered by GnuTLS's randomization of the order of TLS extensions~~ [Data analysis](https://codeberg.org/neuschaefer/git-host-bug#data-analysis) shows that this seems to be independent of the order of TLS extensions added by GnuTLS
I'd be happy to test suggestions that might bring us closer to the root cause, because it's quite a strange issue.