container update no longer fails on Podman when CPU limits are set

src/lib/server/hostconfig-recreate.ts

@@ -0,0 +1,34 @@
+/**
+ * Pure HostConfig sanitizers for container recreation (auto-update).
+ *
+ * recreateContainerFromInspect passes the OLD container's whole HostConfig back to
+ * create. Some fields the daemon REPORTS on inspect it REJECTS on create - notably on
+ * Podman, whose inspect is less normalized than Docker's. These fixups run in-place
+ * just before create.
+ */
+
+interface CpuHostConfig {
+	NanoCpus?: number;
+	CpuPeriod?: number;
+	CpuQuota?: number;
+}
+
+/**
+ * NanoCpus and CpuPeriod/CpuQuota are two encodings of the same CPU limit and are
+ * mutually exclusive at create time. Podman's inspect reports BOTH the absolute
+ * NanoCpus and the derived period/quota, so replaying the HostConfig trips its
+ * "NanoCpus conflicts with CpuPeriod and CpuQuota" error (#1381). Keep NanoCpus (the
+ * direct `cpus:` equivalent) and drop the derived pair. Docker reports period/quota as
+ * 0 for a `--cpus` container, so the condition never fires there - zero change for
+ * Docker, and no effect when a container uses only one of the two encodings.
+ *
+ * Mutates `hostConfig` in place. Returns true when it dropped the pair.
+ */
+export function resolveNanoCpusConflict(hostConfig: CpuHostConfig | null | undefined): boolean {
+	if (!hostConfig) return false;
+	if (!hostConfig.NanoCpus) return false;
+	if (!hostConfig.CpuPeriod && !hostConfig.CpuQuota) return false;
+	delete hostConfig.CpuPeriod;
+	delete hostConfig.CpuQuota;
+	return true;
+}