Cloud & DevOps Workflows

olares-cli cluster: per-user Kubernetes view of an Olares cluster via the ControlHub backend. Read pods / containers / workloads (Deployment/StatefulSet/DaemonSet) / application spaces (KubeSphere-grouped namespaces) / namespaces / nodes / jobs / cronjobs / Olares-managed middleware (databases, queues, object stores). Mutate via scale / restart / stop / start / delete on workloads, delete / restart on pods, suspend / resume on cronjobs, rerun on jobs, and password set on middleware — every mutating verb is wrapped in a confirmation prompt that --yes opts out of. Watch verbs (pod get -w, workload rollout-status -w, application status -w, pod/container logs -f) poll on --interval, never stream. Use whenever the user asks what's running on my cluster?, tail logs of , restart / scale / delete this workload, who am I on this cluster?, suspend this cronjob, rerun this job, or rotate the password on this database. Do NOT use for app-store lifecycle (use olares-cli market) or host-side install / node-join / OS upgrade (use olares-cli node / os / gpu — those go through kubeconfig, not a profile).

olares-cli dashboard command tree — AI-agent-oriented mirror of the dashboard SPA's Overview2 + Applications2 routes. Covers: the strict dual-shape JSON envelope (leaf items{raw|display}+meta vs aggregated sections+meta), the stable dashboard. . Kind constants under cli/pkg/dashboard/schema.go, the 1:1 Go port of @bytetrade/core/src/monitoring.ts (UnitTypes / GetValueByUnit / GetSuitableUnit / GetSuitableValue / WorthValue / FormatFrequency / ConvertTemperature / GetDiskSize / GetThroughput / FormatTime / GetMinuteValue / GetLastMonitoringData / GetResult) under cli/pkg/dashboard/format/, the shared fetch helpers (FetchClusterMetrics / FetchNodeMetrics / FetchUserMetric / FetchWorkloadsMetrics / BuildRankingEnvelope / FetchSystemIFS / FetchSystemFan / FetchGraphicsList / FetchTaskList / FetchGraphicsDetail / FetchTaskDetail) at the cli/pkg/dashboard/ root, the two-layer architecture with MIRRORED subpackage trees on both sides — cmd shell under cli/cmd/ctl/dashboard/ (directory tree mirrors command tree, one thin .go file per leaf whose RunE is just pkg .RunXxx(...), slim ≤70-line common.go per area carrying var common+prepareClient+unknownSubcommandRunE only) vs pkg layer split into cli/pkg/dashboard/ (infra-only: Envelope/Item/Meta/CommonFlags/Client/Runner/all fetchers/all aggregators/the schema bundle) PLUS cli/pkg/dashboard/ / subpackages (cli/pkg/dashboard/applications/, cli/pkg/dashboard/overview/, cli/pkg/dashboard/overview/disk/, cli/pkg/dashboard/overview/fan/, cli/pkg/dashboard/overview/gpu/) owning every leaf's Run/BuildEnvelope/WriteTable + tests next to the code, with strict no-upward (subpackage→root only, root→nothing) and no-sibling (overview/disk ↛ overview/fan) imports inside pkg — the parent-command sections-envelope defaults (overview = physical+user+ranking, overview disk = main+per-disk partitions, overview fan = live+curve) emit through pkg /default.go's RunDefault, the hardcoded FanCurveTable / FanSpeedMaxCPU / FanSpeedMaxGPU constants kept 1:1 with SPA Fan/config.ts, the --watch HTTP-polling semantics (interval / iterations / timeout / 3-consecutive-failure exit / NDJSON-per-iteration / SIGINT-graceful / ErrTokenInvalidated short-circuit), the --since (sliding window in --watch only) vs --start --end (fixed window) mutual-exclusion rule, the three-state empty-data semantics (HTTP 404 → no integration / HTTP 200 empty → no detected), the capability gates (overview fan = HARD-gated to Olares One via /user-service/api/system/status; overview gpu = SOFT advisory mirroring the SPA sidebar — admin role + CUDA-node check populates meta.note and a stderr advisory but lets the HAMI fetch proceed) producing emptyreason ∈ {notolaresone, nofanintegration, novgpuintegration, vgpuunavailable (HAMI 5xx — meta.error / meta.httpstatus), nogpudetected} plus stderr hints, the Client.EnsureUser sync.Once globalrole cache + Client.EnsureSystemStatus + IsOlaresOne + HasCUDANode capability cache, plus RequireAdmin guard for --user and admin-only commands, and the iteration red-lines that forbid horizontal cmd-area imports (any cross-area share must be hoisted to cli/pkg/dashboard/ — the BuildRankingEnvelope precedent), upward pkg-root←pkg-subpackage imports, sibling pkg subpackage imports, putting leaves directly under cmd/ctl/dashboard/ root instead of an area subpackage, putting envelope/table/watch/fan-out logic in cmd, reintroducing trampoline-alias common.go files, or altering envelope/Kind/watch/sections-default/capability-gate semantics. Use whenever the user mentions olares-cli dashboard, asks 'show overview / cpu / memory / disk / pods / network / fan / gpu / ranking / applications / pods', wants JSON for an AI agent, mentions --watch / --since / --start / --end / --user / --head / --limit / --mode (memory) / --temp-unit / --timezone / --test-connectivity (network) on the dashboard tree, hits errors like 'mutually exclusive', 'watch requires', 'platform-admin', or sees meta.empty=true / emptyreason ∈ {novgpuintegration, vgpuunavailable, nogpudetected, notolaresone, nofanintegration} / stderr lines like 'fan is only available on Olares One devices', '(advisory) GPU sidebar entry is hidden ...', or 'gpu data temporarily unavailable: HAMI returned HTTP 5xx ...'.