Table of Contents
Open Table of Contents
- The Week in One Paragraph
- 🚨 Top Stories
- 1. containerd security patch day — 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 (October 13)
- 2. Grafana patch wave #2 — 13.2.4 / 13.1.8 / 13.0.11 / 12.4.13 (October 13)
- 3. HashiCorp Vault 2.1.2 — security release with an ACME breaking change (October 7)
- 4. Terraform v1.16.6 — HTTP/2 resource exhaustion fix (CVE-2026-78669) (October 14)
- 5. Rook v1.21.0 — new minor with three breaking changes (October 6)
- ☸️ Kubernetes & Cloud Native
- ☁️ Cloud Platforms
- 🏗️ Infrastructure as Code & Platform Engineering
- 📊 Observability & SRE
- 🔐 DevSecOps & Infrastructure Security
- 🛠️ Tools & Releases Worth Watching
- 💡 SRE Takeaways of the Week
- Sources
The Week in One Paragraph
Security-patch week. containerd shipped a coordinated security patch day (October 13) across every supported line — 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 — fixing CVE-2026-95839 plus hardening for CRI log symlinks, .. path components in image references, and the default seccomp profile. Grafana ran its second patch wave in two weeks (13.2.4 / 13.1.8 / 13.0.11 / 12.4.13) with four more CVEs. HashiCorp Vault 2.1.2 fixed plugin-catalog escape and ACL canonicalization bypasses and hardened ACME sign-verbatim. Terraform v1.16.6 landed with an HTTP/2 resource-exhaustion fix (CVE-2026-78669). On the release side, Rook v1.21.0 shipped with breaking changes (msgrv2 required by default, Helm OCI tags lose the v prefix) and OpenTelemetry Collector v0.163.0 closed out the week’s observability news.
All items below were published between October 7 and October 14, 2026, and every claim links to a primary source.
🚨 Top Stories
1. containerd security patch day — 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 (October 13)
What happened: containerd shipped coordinated security releases across all supported lines (2.4.2, 2.3.7, plus 2.2.10 and 2.0.14 per the release notes). Security content:
- CVE-2026-95839 plus two additional GHSAs (GHSA-8647-2jxj-p2rc, GHSA-5p7c-hggf-c748, GHSA-hc94-p972-ch66 — the latter two not yet published as public advisory pages at writing time).
- CRI: no longer follows symlinks for container log files.
- Image distribution: rejects image references containing
..path components; 2.4.2 additionally fixes excessive memory usage during image pulls when registry downloads outpace content writes (bounded fetch readers / throttled chunk buffers). - Runtime: default seccomp profile now allowlists socket domains.
- Snapshotters (2.3.7): filters ID-mapping labels from image annotations during unpack.
- Rebuilt with Go 1.26.9 / 1.27.2 (Go security releases).
Why it matters: Second containerd security event in three weeks (CVE-2026-53493 was the September 24 one). The pull-path memory-exhaustion fix in 2.4.2 is the operationally interesting one — a slow-registry / fast-pull divergence could balloon node memory before this patch.
Who is affected: Every Kubernetes node running containerd — which is effectively everyone. Managed-Kubernetes users depend on provider node-image refreshes.
Recommended action: Same playbook as the September containerd CVE: confirm node containerd versions fleet-wide (not just kubelet versions), patch in the next window, and track your provider’s node-image release notes for ≥ 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14.
2. Grafana patch wave #2 — 13.2.4 / 13.1.8 / 13.0.11 / 12.4.13 (October 13)
What happened: Two weeks after the September 29 security patch day, Grafana shipped another coordinated set across the same four release lines (13.2.4, 13.1.8, 13.0.11, 12.4.13):
- Security fixes for CVE-2026-15577, CVE-2026-33374, CVE-2026-93294, CVE-2026-95502 (details pending published advisories at writing time).
- CVE-2026-92539: IAM fix for SCIM (Enterprise).
- Alerting now strips sensitive headers from OAuth2 cross-host redirects — a data-leak-class fix worth having even aside from the CVEs.
Why it matters: Two security patch days in three weeks suggests active vulnerability research against Grafana right now. If you defer Grafana upgrades because “it’s just dashboards,” the last three weeks are the counter-argument — the September wave was all authorization flaws, and this one includes credential-header stripping.
Recommended action: Apply your line’s patch (12.4.13 / 13.0.11 / 13.1.8 / 13.2.4). If you skipped the September 29 round, both rounds apply. Watch for the public GHSA pages to assess exact exposure.
3. HashiCorp Vault 2.1.2 — security release with an ACME breaking change (October 7)
What happened: Vault v2.1.2 shipped with security fixes and one deliberate breaking change:
- core/plugin: plugin catalog entries restored from a raft snapshot, written through
sys/raw, or replicated could no longer run binaries outsideplugin_directory— a containment fix for the plugin execution path. - core (Enterprise): prevents policy-name canonicalization bypasses in ACL parameter restrictions.
- core/policy: request-time ACL construction now ignores malformed legacy policy references containing path-navigation sequences.
- Breaking (PKI/ACME): ACME finalization with the default
sign-verbatimdirectory policy now rejects CSRs containing URI, email, or Other SANs — ACME challenges only validate DNS names and IPs. Old behavior available viadefault_directory_policy = "sign-verbatim-unsafe", which accepts unverified identity claims in certs.
Why it matters: The plugin-catalog fix closes a path where snapshot/replication data could influence which binary Vault executes. The ACME change can break certificate issuance automation at upgrade time if your CSRs carry non-DNS/IP SANs.
Recommended action: Upgrade Vault clusters in the next maintenance window. Before upgrading ACME consumers, inventory CSR SAN types; if anything relies on URI/email SANs under sign-verbatim, decide between fixing the CSRs and explicitly accepting -unsafe.
4. Terraform v1.16.6 — HTTP/2 resource exhaustion fix (CVE-2026-78669) (October 14)
What happened: Terraform v1.16.6 shipped with two bug fixes and a security note:
- Prevents HTTP/2 resource exhaustion via CVE-2026-78669 — exploitation would require a compromised (untrusted) provider/module registry, which HashiCorp considers outside Terraform core’s security perimeter, but the fix landed anyway.
terraform initcan now find workspaces selected by key/value tags whose keys contain uppercase letters.terraform testallows the module under test to use deprecated outputs.
Also this week: v1.17.0-rc1 (October 7) and v1.17.0-rc2 (October 14) — the 1.17 line with GA Terraform Policy and -minimal-refresh is approaching release.
Why it matters: The HTTP/2 fix matters if you run a private registry or mirror — that’s exactly the “trusted by you, compromised by someone else” scenario. The workspace-tag bug fix unblocks teams using uppercase tag keys for workspace selection.
Recommended action: Patch v1.16.6 into CI images at your normal cadence. If you’ve been waiting on 1.17 for policy GA or -minimal-refresh, rc2 is the build to run against your large states now.
5. Rook v1.21.0 — new minor with three breaking changes (October 6)
What happened: Rook v1.21.0 shipped at the edge of the window. Breaking changes:
- Helm OCI chart tags drop the
vprefix (1.21.0, notv1.21.0) — update any tooling that references the chart by tag. - Ceph msgrv2 required by default (needs kernel ≥ 5.11). Escape hatch:
network.connections.requireMsgr2: falseon the CephCluster CR, or the equivalentcephClusterSpecvalue in Helm. - New NFS installs must define a CephBlockPool for the
.nfspool — Rook no longer auto-creates it. Upgraded clusters unaffected.
Features: RBD QoS via VolumeAttributesClass (krbd mounter, cgroup v2 io.max), automated OSD replacement via annotation, and readiness for Ceph v21 (Umbrella).
Why it matters: Storage operators: msgrv2-by-default is a connectivity-level change — mixed-version or old-kernel node pools can fail mounts after upgrade if you don’t audit kernels first.
Recommended action: Before upgrading, check node kernel versions against the 5.11 msgrv2 requirement and set requireMsgr2: false where needed. Fix chart-tag references in your GitOps tooling before the OCI pull breaks.
☸️ Kubernetes & Cloud Native
- Kubernetes v1.38.0-alpha.2 — October 7. Second alpha of the 1.38 cycle; no patch releases this week — the September 23 patches (v1.37.1 etc.) remain current.
- containerd 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 — see Top Story #1.
- Rook v1.21.0 — see Top Story #5.
- Argo CD v3.5.5 / v3.4.11 / v3.3.16 — October 12. Bug-fix patches across three lines: hydrator retries notes pushes on “cannot lock ref” errors, repo-server no longer leaks goroutines or serves stale manifests after a CMP (Config Management Plugin) failure, and events.k8s.io events get proper counts/timestamps. The CMP-failure leak is the one that wedges repo-servers over time — patch if you run CMPs.
- Argo Rollouts v1.10.1 / v1.9.2 / v1.8.5 — October 12. Security fix for GHSA-366v-5xmx-36vh / CVE-2026-82277 (see Security section): the Rollouts dashboard now binds localhost only by default. Breaking for anyone exposing the dashboard via a custom Helm chart — you now need
--address 0.0.0.0explicitly. Also allows Bidirectional mountPropagation on privileged containers. - External Secrets Operator v2.12.0 — October 6. Reverts the v2.11 breaking change to
imageRegistryin the Helm chart, adds prefix/suffix support to the Password generator, and fixes AWS Certificate Manager and Akeyless missing-secret error handling to return properNoSecretErr. - OpenTelemetry Operator v0.161.0 — October 13. Two breaking changes: standalone OpAMP Bridge agents get new instance IDs (now computed from the Kubernetes cluster ID, requiring
geton thekube-systemnamespace), and the long-deprecated--create-rbac-permissionsflag is removed — RBAC creation is always autodetected at startup.
☁️ Cloud Platforms
Cloud provider official update channels (AWS What’s New, Azure Updates, Google Cloud, Cloudflare blog) were not reachable from this run’s research environment, so this section stays short rather than padded with unverified items. Adjacent signals worth tracking:
- containerd rollout on managed Kubernetes: Top Story #1’s fixes land on EKS/GKE/AKS via provider node-image refreshes, not cluster upgrades. Watch node-image release notes for containerd ≥ 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 — same motion as the September CVE-2026-53493 wave.
- GitHub Changelog was also unreachable this run — verify any Actions/runner changes against your own pipelines before relying on this digest for CI/CD news.
🏗️ Infrastructure as Code & Platform Engineering
- Terraform v1.16.6 / v1.17.0-rc1/rc2 — see Top Story #4.
- Ansible v2.22.0rc1 — October 13. First release candidate of ansible-core 2.22, following the September betas. GA is close — last call to run your collections against the pre-release before it lands.
- Backstage v1.56.0-next.2 / next.3 — October 6/13. Next-line builds only this week; no new stable patches after the September 29 v1.55.3 TechDocs fix.
- OpenTofu: no new releases this window — v1.13.0 (September 30) remains current.
📊 Observability & SRE
- Grafana 13.2.4 / 13.1.8 / 13.0.11 / 12.4.13 — see Top Story #2.
- OpenTelemetry Collector v0.163.0 / v1.69.0 — October 14. No end-user breaking renames this time (the
queuebatch→queue_batchrename was last release). Notable: sending-queue deadlock fix when blocked offers are cancelled withblock_on_overflowenabled; an O(n²) → O(1) fix in mdatagen-generated code for scrapes with many data points per resource; oversized-profile splitting now drops only the offending profile instead of the whole queued batch; distributions can opt into a smallerresourcedetectionprocessordetector set via builder config. - Grafana Mimir 3.2.2 / 3.1.7 — October 7. Security patch: the Alertmanager component could load local files referenced by configuration — CVE-2026-81938. If you run Mimir’s Alertmanager with tenant-supplied configs, treat this as a local-file-inclusion fix and patch promptly.
- Loki Operator v0.12.0 — October 7. Breaking: the
schemasfield is now required in the storage spec. Also adds retention-based removal of expired schemas, enablesflush_on_shutdownwith a longer ingester termination grace period (less data loss on rollout), and bumps the Loki operand to 3.7.8. - VictoriaMetrics v1.154.0 — October 9/12. Two security fixes: vmauth now rate-limits requests with invalid auth tokens even when
unauthorized_userURLs are configured (previously proxied without delay, enabling token brute-force — GHSA-rpfx-hrcp-f73x), and vmui XSS fixes on Metric Relabeling Debug / Query Analyzer / Dashboards pages (GHSA-m2x2-9xpq-mmq5). Built with Go 1.27.2. - Grafana Alloy v1.21.0-rc.0 — October 14. First RC of the next line — check the v1.20 breaking changes from two weeks ago are absorbed before testing.
🔐 DevSecOps & Infrastructure Security
- containerd — CVE-2026-95839 + GHSA-8647-2jxj-p2rc / 5p7c-hggf-c748 / hc94-p972-ch66 — see Top Story #1. Patched in 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14. Note: three of the four GHSA pages were not yet public at writing time; treat the release notes as authoritative and re-check advisory pages for severity scoring.
- Grafana — CVE-2026-15577 / 33374 / 93294 / 95502 + CVE-2026-92539 (SCIM, Enterprise) — see Top Story #2. Advisory details pending publication; patch first, assess after.
- Vault v2.1.2 — see Top Story #3. Plugin-catalog escape, ACL canonicalization bypass (Enterprise), and malformed legacy policy reference handling. Plus the deliberate ACME sign-verbatim tightening.
- Terraform — CVE-2026-78669 — HTTP/2 resource exhaustion via compromised registry, fixed in v1.16.6. Low practical risk for public-registry users; relevant if you operate a private registry or mirror.
- Argo Rollouts — GHSA-366v-5xmx-36vh / CVE-2026-82277 — dashboard bound a non-local interface by default; v1.10.1 / v1.9.2 / v1.8.5 bind localhost. If you expose the Rollouts dashboard, upgrade and re-expose deliberately with
--address 0.0.0.0behind proper auth. - Grafana Mimir — CVE-2026-81938 — Alertmanager local-file load via configuration, fixed in 3.2.2 / 3.1.7.
- VictoriaMetrics — GHSA-rpfx-hrcp-f73x (vmauth token brute-force) / GHSA-m2x2-9xpq-mmq5 (vmui XSS) — fixed in v1.154.0.
Operational guidance: this week is “patch the edge of your observability stack” — Grafana, Mimir Alertmanager, Rollouts dashboard, vmauth all had exposure-relevant fixes in seven days. Container runtimes need a second fleet patch in three weeks. None of this waits comfortably for a monthly cycle.
🛠️ Tools & Releases Worth Watching
- Rook v1.21.0 — RBD QoS via VolumeAttributesClass and annotation-driven automated OSD replacement are genuinely useful for Ceph-on-Kubernetes operators; the msgrv2 default forces a long-overdue protocol modernization. rook/rook
- OpenTelemetry Collector v0.163.0 — sending-queue deadlock fix and O(n²)→O(1) datapoint recording fix are the kind of quiet performance/correctness work that matters at scale. open-telemetry/opentelemetry-collector
- Argo CD v3.5.5 — repo-server goroutine leak + stale manifests after CMP failure is a real availability fix for anyone running Config Management Plugins. argoproj/argo-cd
- External Secrets Operator v2.12.0 — un-breaks the v2.11
imageRegistrychart change and hardens missing-secret error semantics for AWS ACM and Akeyless. external-secrets/external-secrets - Terraform v1.17.0-rc2 — GA Terraform Policy and
-minimal-refreshare close; test against your largest states now so the GA upgrade is uneventful. hashicorp/terraform
💡 SRE Takeaways of the Week
- Patch containerd fleet-wide — again. 2.4.2 / 2.3.7 / 2.2.10 / 2.0.14 fix CVE-2026-95839 plus log-symlink, path-traversal, and seccomp hardening. Verify node containerd versions directly; don’t trust the kubelet version as a proxy.
- Apply the second Grafana wave (13.2.4 / 13.1.8 / 13.0.11 / 12.4.13). Four CVEs plus SCIM and OAuth header fixes. Two patch days in three weeks — stop treating Grafana as low-risk internal tooling.
- Audit Rollouts dashboard exposure before upgrading to v1.10.1. CVE-2026-82277 changes the default bind to localhost; if you intentionally expose it, add
--address 0.0.0.0and put it behind auth deliberately. - Check node kernels before Rook v1.21.0. msgrv2 (kernel ≥ 5.11) is now required by default — set
requireMsgr2: falsewhere you can’t meet it, and fix OCI chart-tag references (no morevprefix). - Plan the Loki Operator v0.12.0 schema migration.
schemasis now required in the storage spec — update your LokiStack CRs before the operator upgrade, not after.
Sources
- containerd 2.4.2 — containerd/containerd
- containerd 2.3.7 — containerd/containerd
- containerd GHSA-8647-2jxj-p2rc
- containerd GHSA-5p7c-hggf-c748
- containerd GHSA-hc94-p972-ch66
- Grafana 13.2.4 — grafana/grafana
- Grafana 13.1.8 — grafana/grafana
- Grafana 13.0.11 — grafana/grafana
- Grafana 12.4.13 — grafana/grafana
- Vault v2.1.2 — hashicorp/vault
- Terraform v1.16.6 — hashicorp/terraform
- Terraform v1.17.0-rc1 — hashicorp/terraform
- Terraform v1.17.0-rc2 — hashicorp/terraform
- Rook v1.21.0 — rook/rook
- Kubernetes v1.38.0-alpha.2 — kubernetes/kubernetes
- Argo CD v3.5.5 — argoproj/argo-cd
- Argo Rollouts v1.10.1 — argoproj/argo-rollouts
- Argo Rollouts GHSA-366v-5xmx-36vh (CVE-2026-82277)
- External Secrets Operator v2.12.0 — external-secrets/external-secrets
- OpenTelemetry Operator v0.161.0 — open-telemetry/opentelemetry-operator
- OpenTelemetry Collector v0.163.0 — open-telemetry/opentelemetry-collector
- Grafana Mimir 3.2.2 — grafana/mimir
- Loki Operator v0.12.0 — grafana/loki
- VictoriaMetrics v1.154.0 — VictoriaMetrics/VictoriaMetrics
- Ansible v2.22.0rc1 — ansible/ansible
- Backstage v1.56.0-next.3 — backstage/backstage
- Grafana Alloy v1.21.0-rc.0 — grafana/alloy