Highlights
HPA auto-tune keeps the original pod request when a later resize or a conflicting write would move the stored base. The published image is built with Go 1.27.2.
Bug fixes
- HPA utilization moved again after a second resize. The first retune stores the containers in the base. A later resize adds only a container missing from that list. A v0.1.33 base with no list is not raised when a counted container grows until its old request equals the stored quantity, including the same match with native sidecars removed. The container list is recorded and the quantity stays. A quantity that was already doubled stays until you delete the target and request annotations (#1037, #1053).
- A resize that left the utilization percent unchanged forgot the original request. The first time that happens, Attune stores the pre-resize request. The next resize scales from that stored request (#1046).
- A conflicting HPA write put the earlier target back. The retry reads the HPA again and decides from that object. A status-only conflict still publishes this resize. Turning auto-tune off, or pointing the HPA at another workload, skips the write (#1052).
- Prometheus info notices are logged and the sample is kept. An info notice is not a failed query (#1043).
Upgrading
No CRD change and no policy YAML change. When this release is published, upgrade the chart to 0.1.34, or set the image tag to 0.1.34 or v0.1.34.
The image is built with Go 1.27.2. golang.org/x/net is v0.61.0. The Prometheus client is v1.25.0, prometheus/common is v0.72.1, and the Prometheus library used to parse queries is v0.315.0. You do not upgrade your Prometheus server for this. AWS SDK packages move to their current patch releases.
If attune.io/original-cpu-request or attune.io/original-memory-request is already larger than the pod total from before the first Attune resize, delete that pair and the containers annotation if it is set. Leave attune.io/auto-tune set. The next resize stores a fresh base and the list. See Upgrading.
Contributors
Thanks to external contributors in this release:
- @DrakeJohnson6545 for matching the security scan schedule in the docs (#1039)
- @konih for reporting a full stored HPA base rewritten when it equals the history old sum (#1036)
- @konih for reporting an HPA retune retry that wrote an earlier target over a concurrent edit (#1049)