Daily Engineering Intelligence

Technology Newspaper

← Back to Home

Engineering Intelligence — Monday

November 9, 2026

🔥 What Changed

A quiet Monday with two items. Keycloak CVE-2026-120984 now has a fix in Red Hat Build of Keycloak 26.4.18 (shipped 11-07), but upstream Keycloak still has no fixed release. For Worth Your Time, a Go write-up on deterministic simulation testing of Raft with a fake clock and seed-replayable chaos.

Holds: Go 1.27.4/1.26.11 releases tomorrow, Tue 11-10 US business hours (12 private CVEs 97033–97043 and 97054; go.dev/dl still 1.27.3/1.26.10); upstream Keycloak fix for CVE-2026-120984; Kafka 3.9.3 code freeze 11-20; gopls release for #82821 + #82891; CL 858982 (#82853/#82861); CVE-2026-98432 OpenShift CoreOS; Postgres 19 RC2 ~11-12 / GA ~11-19.

Keycloak CVE-2026-120984 update: Red Hat Build of Keycloak 26.4.18 ships the fix; upstream still unfixed

Follow-up to yesterday's CVSS 9.1 report. Red Hat shipped Build of Keycloak 26.4.18 on 11-07 (RHSA-2026:96037 packages, RHSA-2026:96038 OpenShift images) listing CVE-2026-120984 as fixed. Upstream Keycloak (through 26.8.2, releases page still shows 26.8.2 from 11-05) has no fixed release, and Red Hat lists no mitigation. Not in CISA KEV; no reported exploitation. A third-party tracker suggests, as an unofficial narrowing step, limiting manage-clients and avoiding client create/import with authorization services and an authorizationSettings payload.

Why you care: Auth/IAM is a rising board category. If you run RHBK, there is now a concrete upgrade target; if you run upstream, the control remains auditing who holds manage-clients until a release lands. Hold stays open for the upstream fix.
Read Red Hat CVE page →

📚 Worth Your Time

Niklas Weber: Testing Raft with a fake clock (raftlab, deterministic chaos in Go)

A from-scratch Go Raft implementation whose core is a pure state machine; transport, clock, storage and state machine are injected, so a five-node cluster runs on a fake clock in a single-threaded test and each run is a pure function of its seed. The chaos test crashes nodes, partitions the network and injects loss/duplication/delay for 40 virtual seconds, checking election safety and committed-prefix agreement every 10 ms, then convergence and acknowledged-write durability. Bugs found: stale timer callbacks deposing a healthy leader, delayed AppendEntries erasing newer entries, reordered replies moving nextIndex backwards. It also states the limit: the one bug the fake clock could not reproduce was in the clock's own concurrency. Personal project (~5,100 lines of Go); read-check is weaker than a real linearizability checker.

Why you care: A small, concrete Go example of deterministic simulation testing, the same idea as the OpenAI Pando and Go runtime/bubble (#82785) items reported earlier this month, with honest limits. Worth skimming for the injected-interfaces pattern.
Read article →