diff --git a/readme-vars.yml b/readme-vars.yml
index 27a1076..33fb2a0 100644
--- a/readme-vars.yml
+++ b/readme-vars.yml
@@ -132,6 +132,7 @@ init_diagram: |
   "kasm:latest" <- Base Images
 # changelog
 changelogs:
+  - {date: "23.09.26:", desc: "Remove a stale containerd pid on start, which could keep dockerd from starting after an unclean shutdown."}
   - {date: "07.08.26:", desc: "Fix NVIDIA passthrough to Workspace containers by generating a CDI specification inside the DinD layer."}
   - {date: "04.08.26:", desc: "Give kasm_manager a healthcheck start period so a cold start does not leave kasm_proxy created but never started."}
   - {date: "16.04.26:", desc: "Update for 1.18.1 release. Use rolling service images. Bump docker to v29."}
diff --git a/root/etc/s6-overlay/s6-rc.d/init-config-kasm/run b/root/etc/s6-overlay/s6-rc.d/init-config-kasm/run
index c12e153..c3becc3 100755
--- a/root/etc/s6-overlay/s6-rc.d/init-config-kasm/run
+++ b/root/etc/s6-overlay/s6-rc.d/init-config-kasm/run
@@ -13,6 +13,10 @@ if [[ -S /var/run/docker.sock ]]; then
   rm -f /var/run/docker.sock
 fi
 
+# Clean up old containerd pid, which dockerd takes for a running containerd if
+# another process has been given that pid since an unclean shutdown
+rm -f /var/run/docker/containerd/containerd.pid
+
 # The NVIDIA container toolkit on the host injects the driver into this container,
 # but nothing carries it over to the Workspace containers started inside the DinD
 # layer. Describe the devices and driver we were handed in a CDI specification, so