Harthelm Docs

Harthelm 1.0.3

What changed in Harthelm 1.0.3: stopped services and the night drain, hart ws wake --wait, Restart in a running stage, the helmet icon.

Last updated 2026-10-19

Released 2026-10-11. API version 0.10.1, unchanged. A patch release: upgrade from any 1.0 release without any step of your own.

Terminal
helm upgrade harthelm oci://<your registry>/charts/harthelm --version 1.0.3 -n harthelm-system -f my-values.yaml

For developers

hart ws wake --wait waits for everything it woke. Without --target, hart ws wake --wait used to wait for the editor only. It now waits for every target the wake started, says on standard error which ones, and fails when one of them cannot wake. Services whose run switch is off are neither woken nor waited for. Against a server before 1.0.3 it keeps waiting for the editor only.

Restart keeps the run going. Restarting one service while the run stage was running could end the stage as stopped, and the Run panel and hart then showed the service as stopped although it ran. A Restart no longer ends the stage; a Stop still does.

The helmet everywhere. The editor's browser tab and app icon, and the pages a workspace address answers with, show the Harthelm helmet instead of the VS Code logo or none. The console's icon is the lime helmet on a transparent background.

Console. With the workspace's tool rail expanded, the labels are short enough to fit on one line ("Env vars", "Landscape"); the full names stay in the tooltips and for screen readers.

For operators

The night drain also puts stopped services to sleep. Services whose run switch is off (never run, or stopped by someone) kept a pod that was not ready on a node the drain took out of service, and the pod stayed Pending all night. The drain now puts them to sleep like the other services. They stay stopped: opening one of their addresses brings its pods back and answers "This service is stopped", and Run or Start wakes and starts them. The idle timeout and a manual sleep still leave stopped services alone. See Sleep and wake.

A workspace's wake answers "stopped" for stopped services. The answer of Wake a workspace says stopped for each stopped or torn-down service instead of its last wake, and nothing is woken for it.

Doctor names allow-all egress rules. A networkPolicy.egress rule with neither to nor ports allows all egress. The network-policy check of hart admin doctor now lists it first, with "(all ports)", and says that it does not restrict egress at all. It stays a warning.

Agent version. A workspace agent of one version running in a service image built FROM the base image of another release reports its own version, not the base image's.

Changed values

None. The comment of networkPolicy.egress now explains the allow-all case.

On this page