Skip to Content
Release NotesFuzzball v4.4.1 release notes

Fuzzball v4.4.1 release notes

Fuzzball v4.4.1 is a patch release. It removes a known HTTP/2 vulnerability from the operator image, fixes Web UI logins when the UI hostname is not under the ingress domain, and fixes job logs, inter-job networking and teardown on cloud deployments created with the fuzzball cluster commands.

Enhancements

Deployment

  • Web UI cookie scope. FuzzballOrchestrate and FuzzballFederate accept a new spec.fuzzball.ui.cookieScope setting that controls which hostnames receive the Web UI login cookie. auto (the default) shares the cookie across the ingress domain when the UI hostname is that domain or under it, and keeps it on the UI hostname otherwise. domain always shares it across the ingress domain, and host always keeps it on the UI hostname, in which case workflow service endpoints no longer accept browser logins. A UI under the ingress domain gets the same cookie as in v4.4.0.

Fixes

Security

  • Fixed an issue where the Pulumi provider plugins bundled in the operator image were vulnerable to the HTTP/2 Rapid Reset denial-of-service attack (CVE-2023-44487). The operator now ships plugin versions built with the fix. Existing deployments keep their resources as they are; nothing is re-created on upgrade.

Web UI

  • Fixed an issue where logging in to the Web UI returned to the login page with an expired-session message every time when the UI hostname was not under the ingress domain (for example, ingress domain sys.example.com with the UI at fuzzball-sys.example.com). The browser discarded the login cookie because it was scoped to the ingress domain. With the default auto cookie scope, the cookie is now scoped to the UI hostname in this case.

Deployment

AWS

  • Fixed an issue where the fuzzball cluster aws preflight check required quotas of 120 standard vCPUs and 10 Elastic IPs, far more than a deployment uses. The minimums now match one deployment: 40 standard vCPUs and 3 Elastic IPs.

OCI

  • Fixed an issue where reading any job log on an OCI deployment failed with a 404. The substrate bridge was never granted access to its log bucket, because OCI does not support workload identities in dynamic groups. The deployment now creates a policy scoped to the bridge's service account and the log bucket. Run fuzzball cluster oci update with the v4.4.1 CLI to fix an existing deployment; the policy and dynamic group that older CLIs created in the tenancy root are removed.
  • Fixed an issue where workflow jobs on different OCI compute nodes could not connect to each other by name, because the nodes' network security group admitted only a fixed list of ports. Compute nodes now accept traffic from each other on every port. Existing deployments get the rule on their next fuzzball cluster oci update.

GCP and OCI

  • Fixed an issue where fuzzball cluster gcp destroy and fuzzball cluster oci destroy left every compute node the provisioner had launched running, and on OCI then failed to delete the deployment's network. Both commands now delete those nodes and wait for them to go before tearing down the deployment. If OCI cannot delete them, the teardown still runs and the command ends with an error naming the tag to find the remaining nodes by.

Known issues

  • The macOS CLI in v4.4.1 is signed but not notarized. Installs through Homebrew work normally. A CLI downloaded through a browser, from the Web UI or the CIQ portal, is blocked by macOS Gatekeeper on its first run until you remove the quarantine attribute: xattr -d com.apple.quarantine ~/bin/fuzzball