package main import ( "connectrpc.com/connect" "context" "fmt" "log" "os" "os/exec" homelabv1 "pwa-homelab-mon/gen/homelab/v1" "regexp" "sort" "strings" "time" ) // dockerC is one container as seen by `docker inspect`. type dockerC struct { name, project, status, policy string } // --- services: systemd user units + docker containers, one flat list --- // Service names double as file/unit/container selectors — no separators, no traversal. var svcNameRe = regexp.MustCompile(`^[a-z0-9_.-]+$`) const svcTimeout = 8 * time.Second func (s *server) services(ctx context.Context) []*homelabv1.Service { if s.mockSvcs != nil { return s.mockSvcs } svcs := append(s.systemdServices(ctx), s.dockerServices(ctx)...) // живые (active/partial) наверх, остальное (inactive, not-found, failed, error) вниз; внутри групп — по имени rank := func(svc *homelabv1.Service) int { if svc.Error == "" && (svc.State == "active" || svc.State == "partial") { return 0 } return 1 } sort.SliceStable(svcs, func(i, j int) bool { if ri, rj := rank(svcs[i]), rank(svcs[j]); ri != rj { return ri < rj } return svcs[i].Name < svcs[j].Name }) return svcs } // logProbeChange logs poll-path probe failures once per change: the poll repeats every 3s per client, // and the same error every tick is noise. msg "" = probe healthy, logs a single recovery line. func (s *server) logProbeChange(key, msg string) { prev, loaded := s.probeErr.LoadOrStore(key, msg) if !loaded { if msg != "" { log.Printf("probe %s: %s", key, msg) } return } if prev.(string) == msg { return } s.probeErr.Store(key, msg) if msg == "" { log.Printf("probe %s: recovered", key) } else { log.Printf("probe %s: %s", key, msg) } } func dockerOut(ctx context.Context, args ...string) (string, error) { ctx, cancel := context.WithTimeout(ctx, svcTimeout) defer cancel() out, err := exec.CommandContext(ctx, "docker", args...).Output() return string(out), err } func (s *server) dockerContainers(ctx context.Context) ([]dockerC, error) { ids, err := dockerOut(ctx, "ps", "-aq") if err != nil { return nil, err } if strings.TrimSpace(ids) == "" { return nil, nil } out, err := dockerOut(ctx, append([]string{"inspect", "-f", `{{index .Config.Labels "com.docker.compose.project"}}|{{.Name}}|{{.State.Status}}|{{.HostConfig.RestartPolicy.Name}}`}, strings.Fields(ids)...)...) if err != nil { return nil, err } var cs []dockerC for _, line := range strings.Split(strings.TrimSpace(out), "\n") { f := strings.Split(line, "|") if len(f) != 4 { continue } cs = append(cs, dockerC{project: f[0], name: strings.TrimPrefix(f[1], "/"), status: f[2], policy: f[3]}) } sort.Slice(cs, func(i, j int) bool { return cs[i].name < cs[j].name }) return cs, nil } // working drops one-shot helpers (restart policy no + already exited: *-init, migrations). They belong to // their project's card but must not drag its aggregate down, and must not be re-run by the start button. func working(cs []dockerC) []dockerC { out := make([]dockerC, 0, len(cs)) for _, c := range cs { if c.policy == "no" && c.status != "running" { continue } out = append(out, c) } return out } func (s *server) dockerServices(ctx context.Context) []*homelabv1.Service { cs, err := s.dockerContainers(ctx) if err != nil { s.logProbeChange("docker", err.Error()) return []*homelabv1.Service{{Kind: homelabv1.ServiceKind_DOCKER, Name: "docker", Error: err.Error()}} } s.logProbeChange("docker", "") groups := map[string][]dockerC{} var keys []string for _, c := range cs { key := c.project if key == "" { key = c.name // docker run without a compose label -> its own card } if _, ok := groups[key]; !ok { keys = append(keys, key) } groups[key] = append(groups[key], c) } sort.Strings(keys) svcs := make([]*homelabv1.Service, 0, len(keys)) for _, k := range keys { if svc := dockerCard(k, working(groups[k])); svc != nil { svcs = append(svcs, svc) } } return svcs } // dockerCard aggregates a project: all running -> active, none -> inactive, else partial (up/total). func dockerCard(name string, cs []dockerC) *homelabv1.Service { if len(cs) == 0 { return nil } up := 0 names := make([]string, 0, len(cs)) for _, c := range cs { if c.status == "running" { up++ } names = append(names, c.name) } state := "partial" switch { case up == len(cs): state = "active" case up == 0: state = "inactive" } return &homelabv1.Service{Kind: homelabv1.ServiceKind_DOCKER, Name: name, State: state, Substate: fmt.Sprintf("%d/%d", up, len(cs)), Containers: names} } func (s *server) units() ([]string, error) { entries, err := os.ReadDir(s.srvDir) if err != nil { return nil, err } units := make([]string, 0, len(entries)) for _, e := range entries { if n := strings.TrimSuffix(e.Name(), ".service"); n != e.Name() && svcNameRe.MatchString(n) { units = append(units, n) } } sort.Strings(units) return units, nil } func (s *server) systemdServices(ctx context.Context) []*homelabv1.Service { units, err := s.units() if err != nil { s.logProbeChange("units", err.Error()) return []*homelabv1.Service{{Kind: homelabv1.ServiceKind_SYSTEMD, Name: s.srvDir, Error: err.Error()}} } s.logProbeChange("units", "") svcs := make([]*homelabv1.Service, 0, len(units)) for _, u := range units { svcs = append(svcs, s.systemdStatus(ctx, u)) } return svcs } // ponytail: one systemctl per unit (~26 per 3s poll, ~10ms each). Batch into a single // `systemctl show u1 u2 ... -p ...` if the poll ever shows up in a slow trace. func (s *server) systemdStatus(ctx context.Context, unit string) *homelabv1.Service { svc := &homelabv1.Service{Kind: homelabv1.ServiceKind_SYSTEMD, Name: unit} ctx, cancel := context.WithTimeout(ctx, svcTimeout) defer cancel() // show is the machine-readable probe (always exit 0, unlike status). LoadState=not-found means the user // disabled the unit on purpose — a state, not an error. out, err := exec.CommandContext(ctx, "systemctl", "--user", "show", unit+".service", "-p", "ActiveState,SubState,LoadState,UnitFileState").Output() // systemctl prints properties in its own fixed table order (not the -p order), so parse by key props := map[string]string{} for _, line := range strings.Split(string(out), "\n") { if k, v, ok := strings.Cut(strings.TrimSpace(line), "="); ok { props[k] = v } } svc.State, svc.Substate = props["ActiveState"], props["SubState"] svc.LoadState, svc.UnitFileState = props["LoadState"], props["UnitFileState"] if svc.LoadState == "" { svc.LoadState = "not-found" } if err != nil { svc.Error = err.Error() s.logProbeChange("systemctl/"+unit, svc.Error) } else { s.logProbeChange("systemctl/"+unit, "") } return svc } // ServiceAction runs a fixed systemctl/docker command; only `name` is user input and it is regex-checked, // never interpolated into a shell string. func (s *server) ServiceAction(ctx context.Context, req *connect.Request[homelabv1.ServiceActionRequest]) (*connect.Response[homelabv1.ServiceActionResponse], error) { name, kind := req.Msg.GetName(), req.Msg.GetKind() if !svcNameRe.MatchString(name) { return nil, connect.NewError(connect.CodeInvalidArgument, fmt.Errorf("bad service name %q", name)) } var args []string switch req.Msg.GetOp() { case homelabv1.ServiceOp_START: args = []string{"start"} case homelabv1.ServiceOp_STOP: args = []string{"stop"} case homelabv1.ServiceOp_RESTART: args = []string{"restart"} default: return nil, connect.NewError(connect.CodeInvalidArgument, fmt.Errorf("bad op for %q", name)) } if s.mockSvcs != nil { return connect.NewResponse(&homelabv1.ServiceActionResponse{Output: s.mockAct(name, kind, args[0])}), nil } ctx, cancel := context.WithTimeout(ctx, svcTimeout) defer cancel() var cmd *exec.Cmd if kind == homelabv1.ServiceKind_DOCKER { targets, err := s.dockerTargets(ctx, name, args[0]) if err != nil { return nil, connect.NewError(connect.CodeInternal, err) } if len(targets) == 0 { return connect.NewResponse(&homelabv1.ServiceActionResponse{Output: "no containers to " + args[0]}), nil } cmd = exec.CommandContext(ctx, "docker", append([]string{args[0]}, targets...)...) } else { cmd = exec.CommandContext(ctx, "systemctl", "--user", args[0], name+".service") } out, err := cmd.CombinedOutput() if err != nil { log.Printf("service %s/%s %s: ERROR %v: %s", kind, name, args[0], err, out) return nil, connect.NewError(connect.CodeInternal, fmt.Errorf("%v: %s", err, out)) } log.Printf("service %s/%s %s -> %s", kind, name, args[0], strings.TrimSpace(string(out))) return connect.NewResponse(&homelabv1.ServiceActionResponse{Output: strings.TrimSpace(string(out))}), nil } // dockerTargets resolves a card name (compose project, or bare container name) to the containers an op applies to. func (s *server) dockerTargets(ctx context.Context, name, op string) ([]string, error) { cs, err := s.dockerContainers(ctx) if err != nil { return nil, err } var targets []string for _, c := range working(cs) { if c.project != name && c.name != name { continue } if running := c.status == "running"; (op == "start" && !running) || (op != "start" && running) { targets = append(targets, c.name) } } return targets, nil } // ServiceInfo returns raw status output for the modal: `systemctl status` / `docker ps -a` for the card. func (s *server) ServiceInfo(ctx context.Context, req *connect.Request[homelabv1.ServiceInfoRequest]) (*connect.Response[homelabv1.ServiceInfoResponse], error) { name, kind := req.Msg.GetName(), req.Msg.GetKind() if !svcNameRe.MatchString(name) { return nil, connect.NewError(connect.CodeInvalidArgument, fmt.Errorf("bad service name %q", name)) } ctx, cancel := context.WithTimeout(ctx, svcTimeout) defer cancel() if s.mockSvcs != nil { return connect.NewResponse(&homelabv1.ServiceInfoResponse{Output: s.mockInfo(name, kind)}), nil } var out string var err error if kind == homelabv1.ServiceKind_DOCKER { // compose project first, bare container (no label) as fallback out, err = dockerOut(ctx, "ps", "-a", "--filter", "label=com.docker.compose.project="+name) if strings.TrimSpace(out) == "" { out, err = dockerOut(ctx, "ps", "-a", "--filter", "name="+name) } } else { // status exits non-zero for inactive units (code 3) — that is valid output, not an error c := exec.CommandContext(ctx, "systemctl", "--user", "status", name+".service", "--no-pager", "-n", "0") b, cerr := c.CombinedOutput() out, err = string(b), cerr } if err != nil && strings.TrimSpace(out) == "" { log.Printf("service info %s/%s: ERROR %v", kind, name, err) return nil, connect.NewError(connect.CodeInternal, err) } log.Printf("service info %s/%s ok", kind, name) return connect.NewResponse(&homelabv1.ServiceInfoResponse{Output: strings.TrimRight(out, "\n")}), nil }