Conversation
…ork recreate Diagnosed on a Fedora VM after lerd update --beta migrated the network to dual-stack. Symptom: every phpmyadmin page took 5–10 seconds; DNS lookups from inside containers returned EMSGSIZE "Message too large" on UDP; TCP DNS worked fine; containers that joined the recreated network weren't registered in aardvark-dns (NXDOMAIN for lerd-mysql and friends even though the container was clearly running). Two separate podman/netavark quirks on Fedora's rootless stack: 1. Container eth0 MTU was 65520 (jumbo). glibc's resolver uses IP_MTU_DISCOVER=DO on its UDP socket and kernel returns EMSGSIZE on write() under that MTU on some paths. TCP DNS fallback eventually resolved but cost 5+ seconds per page. cachyos defaults to a sane MTU so the bug was invisible there. 2. RemoveNetwork only unlinked aardvark-dns's config file but the running aardvark-dns process kept its old inode. When containers joined the recreated network netavark wrote a new config file at the same path, but aardvark-dns was reading from the stale inode and never saw the new container entries — half of them ended up with NXDOMAIN. Changes: - New LerdNetworkMTU = "1500" constant. EnsureNetwork and MigrateNetworkToIPv6 both pass --opt mtu=1500 to `podman network create`. 1500 is the universal bridge MTU; hosts that worked before (cachyos) are unaffected. - RemoveNetwork now also runs `pkill -f aardvark-dns` after unlinking the config file. Podman/netavark auto-respawns aardvark-dns on the next container join, against the fresh config. Verified on Fedora 40 rootless podman 5.x: after the manual equivalent of these changes, curl to phpmyadmin dropped from 10 seconds to 220ms.
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
1 participant
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.
Diagnosed on a Fedora 40 VM after
lerd update --betamigrated the lerd podman network to dual-stack. Symptom: every phpmyadmin page took 5–10 seconds. Two separate rootless podman/netavark quirks combined:1. Container eth0 MTU was 65520 (jumbo). Fedora's rootless podman defaults the netns interface MTU to pasta's huge value. glibc's resolver uses
IP_MTU_DISCOVER=DOon its UDP socket and some kernel paths returnedEMSGSIZE("Message too large") on thewrite()of a trivially-sized DNS query. Containers fell back to TCP DNS, which worked but cost 5+ seconds per page. cachyos (my other VM) defaults to a saner MTU so the bug was invisible there.2.
RemoveNetworkunlinked aardvark-dns's config file but didn't kill the running aardvark-dns process. The daemon kept its file descriptor on the old inode. When containers joined the recreated network, netavark wrote a new config at the same path, but aardvark kept reading the stale inode. Half the migrated containers ended up missing from DNS —lerd-mysqlreturnedNXDOMAINeven though the container was clearly running, making phpmyadmin time out on its connection attempt.Changes
LerdNetworkMTU = "1500"constant ininternal/podman/network.go. BothEnsureNetworkandMigrateNetworkToIPv6now pass--opt mtu=1500topodman network create. 1500 is the universal bridge MTU; hosts that already worked (cachyos) are unaffected.RemoveNetworknow runspkill -f aardvark-dnsafter unlinking the aardvark config file. Podman/netavark auto-respawns the daemon on the next container join, against the freshly-written config.Verified
On the Fedora 40 VM after applying the manual equivalent of these changes: curl to phpmyadmin dropped from 10 seconds to 220 ms, and aardvark-dns's registry contained all 10 lerd containers after the migration (before the kill, only 3 were registered).