Skip to content

chore: containerd v2.3.5 and the nftables masquerade backend - #26

Merged
CMGS merged 1 commit into
masterfrom
chore/containerd-2.3.5-nftables
Sep 7, 2026
Merged

chore: containerd v2.3.5 and the nftables masquerade backend#26
CMGS merged 1 commit into
masterfrom
chore/containerd-2.3.5-nftables

Conversation

@CMGS

@CMGS CMGS commented Sep 6, 2026

Copy link
Copy Markdown
Contributor
  • containerd_version v2.3.4 → v2.3.5 (CVE-2026-53495, GHSA-rp3h-jf77-q9p4, shim load/delete fixes). The lane's containerd node was upgraded the same way (release tarball, checksum verified) and passes the containerd e2e set.
  • The bridge conflist sets ipMasqBackend: nftables and nftables joins the essential packages (the bridge plugin drives nft for outbound NAT). With the iptables backend, 100 concurrent container starts on an 8-vCPU node serialize on the xtables lock (~0.55 s each) and the slowest exceed the oci-hook's 60 s CNI timeout, so only 32–41 of 100 come up; with nftables the same 100 start in 9–11 s with none lost and 50 start in 5.6 s instead of 34–43 s (two interleaved rounds on the lane, core v0.1.4 and perf: bound the creates in flight on one node core#735 both).
  • The configuration doc follows.

containerd v2.3.5 carries CVE-2026-53495 and GHSA-rp3h-jf77-q9p4. The bridge
conflist now sets ipMasqBackend: nftables and nftables joins the essential
packages: with the iptables backend, 100 concurrent container starts on an
8-vCPU node serialize on the xtables lock (~0.55 s each) and the slowest
exceed the oci-hook's 60 s CNI timeout, so only 34-41 of 100 come up; with
nftables the same 100 start in 9-11 s with none lost (50 in 5.6 s instead of
34 s).
@CMGS
CMGS merged commit e6a73b2 into master Sep 7, 2026
1 check passed
@CMGS
CMGS deleted the chore/containerd-2.3.5-nftables branch September 7, 2026 11:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant