Skip to content
SHC Docs

Live Service Tracking

Layer 6. Services. Services

shc install follows the operation created by shcd and displays per-service progress. shc update and shc uninstall also use shcd; an unreachable daemon refuses these verbs with X503738DMNREQ before deployment work.

flowchart TB
CLI["shc install"] --> API["shcd install handler"]
API --> OP["Operation id"]
API --> RUN["RunInstallWithObserver"]
RUN --> UP["Runtime up with progress polling"]
UP --> OBS["StepObserver"]
OBS --> SSE["Operation SSE stream"]
OP --> CLIENT["CLI follows operation"]
SSE --> CLIENT

The handler in ce modules/app/router.go creates the operation, then RunInstallWithObserver in install_async.go passes its observer into ExecuteCompose. The client follows /api/method/shc.operation.stream/{operation_id} through commands/operation_sse.go.

ComposeUpWithProgress in modules/app/compose_progress.go polls runtime project state while the blocking up operation runs. Its default interval is 500 ms. core/runtimecmd selects the runtime command and parses its output; runtimes without that driver use the project executor.

RollupServices converts the observations into RawContainer values and calls DeriveDeploymentServices. Each emitted operations.ServiceInfo keeps status and health separate, with current and desired replica counts. See status states.

A created container observed during this active converge reads starting. A service absent from a poll keeps its prior progress state. The completed count includes services whose status is running or complete and whose health has settled; starting and unhealthy health keep them unfinished.

Polling is best effort. A command or parse error skips that observation; the blocking up result decides whether the operation succeeded. The poller stops before the final observation is emitted, so it cannot overwrite the terminal frame.

If the final observation is unavailable, FallbackFinal emits running with health none on success, or failed on error. That frame reports the operation result without claiming an observed health check.

commands/operation_live.go renders progress from the operation frames. commands/operation_sse.go remembers the last sequence and reconnects with Last-Event-Id after an interrupted stream, within its resume budget. A terminal complete or error frame ends the follow.

This progress stream is separate from the tenant event stream described in event lifecycle.