Repository navigation
More fixes to accomodate new CI runners - #1773
Merged
Merged
Conversation
bruno-fs
commented
Oct 1, 2026
bruno-fs
commented
Contributor
- disable tests failing on new runners due to egress rules
- fix networking issues with https on squid proxy
basic-ftp install from an ftp:// URL. On our new runners outbound TCP/21 to the internet is denied at falls through to an implicit deny. The connection is dropped rather than refused, so the tests hang for the full timeout instead of failing. Tracked as INSTALLER-4942. Only the Fedora scenarios are affected. The RHEL defaults point at download.devel.redhat.com, which is internal and answers on 21 from these hosts.
Peeking at the ClientHello reads the SNI, which lets squid ask the parent proxy for a hostname and keeps the access log readable. But on an intercepted port squid then verifies that the SNI resolves to the address the client actually connected to, and answers a mismatch with an HTTP 409. It has not bumped the connection at that point, so the error page goes out as cleartext on a socket the client is still TLS-handshaking on. Go reports "server gave HTTP response to HTTPS client", OpenSSL reports "wrong version number", and a bootc install dies immediately. The mismatch is routine rather than an attack. quay.io serves blobs from cdn01.quay.io, which is on Akamai and rotates its answer set within seconds, so the address a test VM resolved is frequently absent from the set squid resolves a moment later. Measured on a runner, two of three addresses changed between queries four seconds apart, while quay.io itself returned the same eight every time - and every 409 in a full day of that runner's access log was this one host. There is no way to turn the check off: squid 5 always answers a suspicious intercepted CONNECT with a 409, host_verify_strict notwithstanding. Not peeking means there is no SNI, nothing to verify, and squid asks the parent for the raw address, which it accepts. The cost is that 443 is logged by address rather than by name. The failure was expensive out of proportion to the three seconds it takes: anaconda treats it as a fatal payload error and parks on the error dialog, so the VM stays up until virt-install's --wait 30 expires and the test is billed a full thirty minutes, twice with the retry. On gh-runner-01 a fedora_rawhide bootc run took 69 minutes with three such timeouts before this change and under five minutes with none after it. Assisted-by: Claude Opus 5
bruno-fs
marked this pull request as ready for review
October 2, 2026 09:21
Contributor
|
/test-os-variants |
rvykydal
approved these changes
Oct 2, 2026
rvykydal
left a comment
rvykydal
left a comment
Contributor
There was a problem hiding this comment.
Looks good to me. Thank you!
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.