Daily Engineering Intelligence

Technology Newspaper

← Back to Home

Engineering Intelligence — Sunday

November 15, 2026

🔥 What Changed

A quiet Sunday with two items. OpenTelemetry-Go published two advisories for bugs already fixed in v1.47.0 (released 10-02). If your services are still on v1.46 or earlier, attacker-controlled baggage headers and U+FFFD-padded attribute values can get past the limits you thought you had. A new Kafka bug (KAFKA-21491) describes a KRaft controller that resigns over and over, so a broker can never rejoin. It needs ELR, a min.insync.replicas change and a topic delete. Affects 4.0–4.3; the issue is open with no fix version yet. Nothing in Worth Your Time cleared the bar today.

Also seen: fix PRs are up for two holds: Go #83076 (phiopt loop-carried bool, fixes #83036) and #83078 (HTTP/2 Body.Read on cancel, fixes #83068), plus #83074 for #83069. None are reviewed yet. Go #83097 s390x Move/Zero "bad pointer in Go heap"; #83087 x/text norm Writer inserts U+034F on small writes; #83086 proposal to deprecate the misspelled x509.VerifyOptions.MaxConstraintComparisions; a batch of wasm assembler fixes (#83088–#83099). Kafka: implementation sub-tasks for KIP-1279 native cluster mirroring filed (KAFKA-21495–21504: CLUSTER_MIRROR ACL/config types, mirror metadata records); KIP-1397 share-group sink connectors under discussion; KAFKA-21492 Streams EOS segmented-store wipe. Cloudflare on AI penetration-testing orchestrators.

Holds: Go 1.27.5/1.26.12 (CVE-2026-97047 #82948, CVE-2026-97044 #82374 backports); Go #83036/#83068 fix PRs landing; Go #82945 TLSNextProto h2 regression; Keycloak CVE-2026-120984 fix (still 26.8.2); gopls release for #82821 + #82891; CVE-2026-98432 OpenShift CoreOS; x/oauth2 CL 854305; Kafka 3.9.3 (code freeze 11-20); KAFKA-21491 fix; Postgres 19 GA ~11-19.

OpenTelemetry-Go: attribute length limit fully bypassed with U+FFFD runes (CVE-2026-128085) and W3C baggage allocation amplification (CVE-2026-128086). Upgrade to v1.47.0

GHSA-q7gw-24fr-5p69 / CVE-2026-128085 (medium, 5.3): an incomplete fix for the earlier GHSA-p9f8-wvj8-2fg8. The "optimistic" truncation fast path in attrnorm skips every valid U+FFFD rune without counting it. A value of 1,000 U+FFFD runes (3,000 bytes) is stored verbatim even with AttributeValueLengthLimit set to 5. This affects otel/sdk trace limits (≥ 1.33.0), sdk/log (≥ 0.9.0) and the auto-instrumentation tracer in otel/trace (≥ 1.35.0). GHSA-rrfc-gjm9-cgrj / CVE-2026-128086 (medium, 5.3): baggage.Parse and propagation.Baggage.Extract eagerly turn every baggage property into objects. A valid 8,191-byte header with one member and 4,094 properties causes about 778 KB of allocation per parse (~95× the input) and keeps about 164 KB live. The advisory says this is bounded amplification and GC pressure, not proven OOM. Affected: ≥ 1.0.0-RC1 through 1.46.x. Both are fixed in v1.47.0.

Why you care: Almost every Go microservice links otel-go, and many extract inbound baggage on public routes before auth. The length-limit bypass means a span attribute cap you rely on to protect your collector or log pipeline doesn't hold. Action: go list -m go.opentelemetry.io/otel and bump to v1.47.0 (it needs Go ≥ 1.26), or strip baggage at the edge if you don't use it.
Read GHSA-q7gw-24fr-5p69 → Also: GHSA-rrfc-gjm9-cgrj →

KAFKA-21491: KRaft controller resigns repeatedly and a broker never rejoins because ClearElrRecord replay leaves stale brokersToElrs entries (4.0–4.3)

A ClearElrRecord is written whenever min.insync.replicas changes. When it's replayed, removeTopicElrs() replaces the partition registration directly instead of going through updatePartitionInfo(), so the brokersToElrs index keeps brokers that are no longer in any ELR. Deleting the topic later only clears index entries for the partition's current (now empty) ELR, so the stale entry outlives the topic. When that broker restarts after an unclean shutdown, or is unregistered, generateLeaderAndIsrUpdates() throws RuntimeException: Topic ID … existed in isrMembers, but not in the topics map. That isn't an ApiException, so the controller resigns, and the next leader replays the same state and fails the same way. The reproduction uses default config on trunk: RF 3, min.insync.replicas=2, kill -9 two brokers so one lands in the ELR, lower min ISR to 1, delete the topic, restart the broker. Listed affected versions are 4.0.0 through 4.3.0. Status: Open, unassigned, no fix version yet.

Why you care: Kafka is a top-weighted messaging skill, and this is a sharp lesson in KRaft internals: a secondary index that isn't updated through the same code path as the primary record ends up as a cluster-wide availability bug. Until a fix ships, avoid changing min.insync.replicas on topics with a non-empty ELR shortly before deleting them, and watch for repeated controller resignations after unclean broker restarts.
Read KAFKA-21491 →