Daily Engineering Intelligence

Technology Newspaper

← Back to Home

Engineering Intelligence — Thursday

November 12, 2026

🔥 What Changed

A normal Thursday with four items. Apache HTTP Server 2.4.71 fixes a critical mod_http2 race in assign_headers() that can be turned into remote heap corruption and RCE on 2.4.55–2.4.70, so check any httpd in front of your services. NATS Server 2.15.3 / 2.14.10 went GA (the RC hold from last week), built on Go 1.27.4 and carrying a batch of JetStream Raft and filestore fixes. The Go team formally proposed graduating most of simd/simd/archsimd out of GOEXPERIMENT for Go 1.28 (#83009). For Worth Your Time, Cloudflare explains why its global control-plane consensus service Meerkat uses the leaderless-progress protocol QuePaxa instead of Raft.

Also seen: glibc advisories batch (GLIBC-SA-2026-0027, wide scanf %s/%[ 1-byte overflow, 2.2–2.44). HashiCorp Vault v2.2.2 tagged (changelog not yet published). Grafana 12.4.15 / 13.0.13 / 13.1.10 / 13.2.6 / 13.3.1 patch train. Terraform 1.17.2. Kyverno 1.20.0-rc.5. GitHub Agentic Workflows GA (InfoQ). Werner Vogels on the AWS Emulator and the engineer's job when agents write code. Go #82948 (CVE-2026-97047, HTTP/2 trailer DoS) still open, milestone Go1.28. Go #82989/#82990 log/slog AddSource and JSONHandler performance issues. Node.js security release pre-alert.

Holds: Go CVE-2026-97047 (#82948, CL 862201; watch for 1.27.5/1.26.12); Go #82945 TLSNextProto h2 regression; Go #82977/#82978 windows heap-corruption backports; Keycloak CVE-2026-120984 upstream fix; gopls release for #82821 + #82891 (still v0.24.0); Vault 2.2.2 changelog; CVE-2026-98432 OpenShift CoreOS; x/oauth2 CL 854305; Kafka 3.9.3 (code freeze 11-20); Postgres 19 RC2 / GA ~11-19; Cloudflare Meerkat verification follow-ups.

Apache HTTP Server 2.4.71: critical mod_http2 race (CVE-2026-119618) allows remote heap corruption and RCE on 2.4.55–2.4.70

CVE-2026-119618 (critical): improper synchronization in mod_http2's assign_headers() lets concurrent HTTP/2 stream handling corrupt the heap, which Apache rates as leading to RCE. Affected: httpd 2.4.55 through 2.4.70 and standalone mod_h2 2.0.8–2.0.42. It was found independently by Anthropic's OSS scanner (reported 10-26) and Google Cloud Vulnerability Research (11-03). The same release fixes CVE-2026-106023 (important), a mod_md NULL dereference that crashes child processes (2.4.30–2.4.70), and CVE-2026-109163 (low), a 1-byte stack overwrite in mod_vhost_alias (2.4.69–2.4.70). Wiz reports that no public exploit or in-the-wild use is known yet.

Why you care: httpd is still a common reverse proxy and TLS terminator in front of Go/gRPC backends, ingress sidecars and legacy edges, and the bug needs no authentication, only HTTP/2 enabled. Action: upgrade to 2.4.71 (including distro and container base images). If you can't upgrade yet, set Protocols http/1.1 to turn off h2/h2c.
Read Apache httpd advisory → Also: www.wiz.io →

NATS Server 2.15.3 / 2.14.10 GA: built on Go 1.27.4, with JetStream Raft health, snapshot and peer-removal fixes plus filestore scan speedups

This closes last week's RC hold. Both lines ship on Go 1.27.4, so they pick up Tuesday's toolchain security fixes. Raft/JetStream fixes: paused followers can now request newer snapshots and compact their logs while catching up, which prevents unbounded log growth and memory use when replay resumes (#8842). Removed peers in unmanaged Raft groups can no longer rejoin as leader (#8847). Raft health checks now wait for the leader's initial data to be applied and stay stable across heartbeats and leader changes (#8817). Meta-group setup failures no longer leak the WAL (#8722), and JetStream no longer shuts down early in lame-duck mode (#8708). Performance: filestore subject matching and pending counts skip deleted-message ranges and stop scanning earlier (#8784/#8875 and others), and consumer redelivery queues no longer copy the backlog on every dequeue (#8635). New in 2.15.3: verify: optional client TLS, ping_interval for leafnodes and gateways, --test_reload to tell whether a config change needs a restart (exit code 3), and a healthz leaf-remotes-connected=true gate.

Why you care: NATS is on your board (Messaging), and this is a compact real-world list of Raft failure modes in production: snapshot catch-up, removed-peer leadership and health-check flapping. Anyone running JetStream R3/R5 clusters should take it.
Read NATS v2.15.3 release → Also: github.com →

Go proposal #83009: graduate most of simd and simd/archsimd (AMD64 AVX/AVX2/AVX-512, ARM64 NEON, Wasm) out of GOEXPERIMENT, targeting Go 1.28

This is the formal proposal behind the "simd graduation" line in the Go Turns 17 post. It would graduate exactly the operations covered by an executable pure-Go specification (#81819) that defines each op's semantics across architectures and is also used for conformance testing (it turned up QEMU bugs). The simd package is portable and vector-size-agnostic and gets a software fallback on other platforms. ARM64 SVE is prototyped (#79781) but is not proposed for GA yet. Open API questions: whether Vec.Equal should keep returning a Mask (with == added for boolean equality, or a rename to Equals staged with //go:fix), and whether to add typeless Bits128/256/512 types to simplify reshaping between vector shapes.

Why you care: Go is your top-weighted skill. If accepted, this is the first standard SIMD API for hot paths (parsing, hashing, compression, vector search) with no assembly or cgo. Now is the time to comment on the API shape.
Read Go proposal #83009 →

📚 Worth Your Time

Choosing QuePaxa: why Cloudflare's global consensus service Meerkat avoids leader-dependent protocols like Raft

Meerkat is Cloudflare's experimental, globally replicated consensus log (a KV store like etcd, Consul or ZooKeeper) for its internal control plane. The post explains why Raft, Multi-Paxos and VR degrade on volatile WAN links. Destructive interference: competing proposers, and competing Raft candidates during elections, keep blocking each other. Metastable failures: a brief network blip triggers elections that keep themselves going. Fault detection: timeouts mistake a slow leader for a dead one. Cloudflare cites a 2020 switch partial failure that made a consensus-backed dashboard 80× slower for six hours. QuePaxa keeps a leader fast path that takes one round trip (over 95% of Meerkat decisions), then falls back to randomized proposer priorities and proposal adoption, so proposers converge instead of competing and no failure detection is needed to make progress. In Cloudflare's tests, a failed leader dropped throughput to about 65% rather than stalling. Cloudflare is extending QuePaxa from crash-stop to crash-recovery and formally verifying the Rust implementation, with follow-up posts promised.

Why you care: Distributed Systems is your #2 weight. This is a clear, practical explanation of why leader-based consensus has availability limits on a WAN, which is useful for system-design interviews and for reasoning about etcd/KRaft/JetStream behaviour under partial network failure.
Read Choosing QuePaxa →