restart
Restart existing regular replicas of one service. See the service-set overload.
Restart existing containers selected by exact project and service labels. Null services selects the whole project; an empty set makes no Docker requests. By default only confirmed regular replicas are included; includeOneOff also includes one-off containers and containers with unknown one-off metadata. Discovery completes before mutations. Requests run sequentially in snapshot order, continuing after HTTP errors and preserving each SDK result in the returned report. Start/stop accept HTTP 304 (already in the requested state) as success. An outer success means the batch completed, not that every container succeeded. Missing projects/services produce an empty report. No resources are created or removed. Transport/decoding failures and cancellation propagate, stopping subsequent requests; already applied changes remain and a failed request may have reached Docker. No dependency ordering, readiness checks, retries or rollback are provided. timeoutSeconds is the per-container stop grace period: null uses Docker defaults, -1 waits indefinitely, 0 kills immediately, positive values wait that many seconds.