I investigated this on Noble with systemd 255.4-1ubuntu8.17 and reproduced the nginx bind failure during a combined nginx/systemd- networkd restart, including a restart initiated by needrestart after a library upgrade.
The reproduction uses nginx bound to a specific SLAAC IPv6 address. A temporary one-second ExecStartPre delay makes the race reproducible. Address monitoring shows networkd deleting the address during restart; nginx then fails with EADDRNOTAVAIL before IPv6 becomes usable again. This reproduces the reported failure mechanism, although the original report mentions DHCPv6. The source investigation points to systemd-networkd: Noble restores address/route source metadata, but initial configuration subsequently reclassifies dynamic configuration as foreign, allowing cleanup to remove it. Relevant upstream changes are: - dd6d53a8dc58c5e6e310b09ba7f7a22600a87ba9, which removes this foreignization step: https://github.com/systemd/systemd/commit/dd6d53a8dc58c5e6e310b09ba7f7a22600a87ba9 - 82df2e0f0496834064f32e4f09cc32728c1cbddd, which includes cleanup of restored configuration when dynamic protocols are disabled: https://github.com/systemd/systemd/commit/82df2e0f0496834064f32e4f09cc32728c1cbddd The same nginx reproduction did not fail on Resolute with systemd 259.5-0ubuntu3.4, and no address deletion was observed. A targeted experimental Noble adaptation passes the nginx reproduction, while an unpatched control built with identical options fails. Three integration tests also pass, covering DHCPv4/DHCPv6/SLAAC address preservation across restart and cleanup when DHCP/RA are disabled. Noble still needs a reviewed fix. The upstream changes do not apply directly to its older code, so the appropriate minimal backport remains under review; the experimental patch is not yet an SRU-ready package. Adding a systemd task and targeting Noble. Closing the unattended- upgrades task as Invalid because the reproduced defect lies in networkd’s restart handling. ** Also affects: systemd (Ubuntu) Importance: Undecided Status: New ** Changed in: systemd (Ubuntu) Status: New => Triaged ** Changed in: unattended-upgrades (Ubuntu) Status: Triaged => Invalid ** Changed in: unattended-upgrades (Ubuntu) Status: Invalid => Won't Fix ** Changed in: systemd (Ubuntu) Assignee: (unassigned) => Zara Grigoryan (zgrigoryan) ** Also affects: unattended-upgrades (Ubuntu Noble) Importance: Undecided Status: New ** Also affects: systemd (Ubuntu Noble) Importance: Undecided Status: New ** Changed in: unattended-upgrades (Ubuntu) Importance: Medium => Undecided ** Changed in: systemd (Ubuntu) Importance: Undecided => High -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2067057 Title: Unattended-upgrades order of restart caused failure to bind address To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/2067057/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
