diff --git a/README.md b/README.md index 410ada6..94914d7 100644 --- a/README.md +++ b/README.md @@ -32,7 +32,7 @@ statuspage │ ├── config.go yaml config loading + defaults │ ├── monitor.go probes, ssh runs, status/history/metrics handlers │ ├── mock.go deterministic mock data for --mock / screenshots -│ └── history.go sqlite history + metrics storage and downsampling +│ └── history.go sqlite history + metrics storage ├── go.mod module definition (statuspage) ├── src/ the frontend (Preact + TS + CSS) ├── example.config.yaml documented, fully commented-out example config @@ -203,8 +203,8 @@ Controls how metric plots share their x (time) axis: - `false` (default): each metric plot scales to its own recorded time span, so a brand new metric isn't drowned out by an older one. -- `true`: all plots are aligned to the same window (the last ~24h) and are - downsampled together, so you can compare shapes directly. +- `true`: all plots are aligned to the same window (the last ~24h), so you can + compare shapes directly. ### `ping` @@ -359,7 +359,7 @@ ssh: echo "uptime: $(uptime -p)" echo "load: $(cat /proc/loadavg)" echo "mem: $(free -h | awk '/^Mem:/ {print $3 "/" $2}')" - echo "top: $(ps -eo %cpu,comm --sort=-%cpu | head -4 | tr '\n' ';')" + echo "top: $(ps --no-headers -eo %cpu,comm --sort=-%cpu | head -4 | tr '\n' ';')" ``` Suggested bounds for the metrics above: @@ -455,10 +455,8 @@ groups: | ------------ | ------------------------------------------------ | | `name` | restrict to one metric name | | `min`, `max` | unix seconds window to fetch | -| `max_points` | max returned points (downsampled, default 100) | -| `shared` | align downsampling to the `min`/`max` window | -The frontend asks for the last ~24h at `max_points=100` per metric. +The frontend asks for the last ~24h per metric. ## Storage diff --git a/example.config.yaml b/example.config.yaml index 740b05f..525dbe2 100644 --- a/example.config.yaml +++ b/example.config.yaml @@ -62,7 +62,7 @@ # echo "load: $(cat /proc/loadavg)" # echo "mem: $(free -h | awk '/^Mem:/ {print $3 "/" $2}')" # echo "disk: $(df -h / | awk 'NR==2 {print $4 " free"}')" -# echo "top: $(ps -eo %cpu,comm --sort=-%cpu | head -4 | tr '\n' ';')" +# echo "top: $(ps --no-headers -eo %cpu,comm --sort=-%cpu | head -4 | tr '\n' ';')" # # Script format, one line per item: # name:status:info -> a check chip (status: on / off / down / anything) diff --git a/index.html b/index.html index 7e8502d..ca374a4 100644 --- a/index.html +++ b/index.html @@ -4,6 +4,7 @@ Status Page + 0.001 { + t.Errorf("uptime %v, want %v", *ms.Uptime, want) + } +} diff --git a/server/history.go b/server/history.go index adee1fa..d321cba 100644 --- a/server/history.go +++ b/server/history.go @@ -108,6 +108,16 @@ func (h *History) RecordMetric(machine, name string, ts int64, value float64, un } } +// Uptime counts the number of 'up' and 'down' history ticks for a machine. +// A machine with no history returns zeroes. +func (h *History) Uptime(machine string) (up, down int) { + h.db.QueryRow(`SELECT + COALESCE(SUM(CASE WHEN status = 'up' THEN 1 ELSE 0 END), 0), + COALESCE(SUM(CASE WHEN status = 'down' THEN 1 ELSE 0 END), 0) + FROM history WHERE machine = ?`, machine).Scan(&up, &down) + return +} + func (h *History) Query(machine string, limit int) ([]HistoryEntry, error) { if limit <= 0 || limit > historyPerMachine { limit = historyPerMachine @@ -132,10 +142,7 @@ func (h *History) Query(machine string, limit int) ([]HistoryEntry, error) { return entries, rows.Err() } -func (h *History) QueryMetrics(machine, name string, minTS, maxTS int64, maxPoints int, shared bool) ([]MetricEntry, error) { - if maxPoints <= 0 || maxPoints > metricsPerName*5 { - maxPoints = 100 - } +func (h *History) QueryMetrics(machine, name string, minTS, maxTS int64) ([]MetricEntry, error) { q := `SELECT ts, name, value, unit FROM metrics WHERE machine = ?` args := []any{machine} if name != "" { @@ -166,90 +173,7 @@ func (h *History) QueryMetrics(machine, name string, minTS, maxTS int64, maxPoin } entries = append(entries, e) } - if err := rows.Err(); err != nil { - return nil, err - } - var lo, hi int64 - if shared { - lo = minTS - hi = maxTS - } - return downsampleMetrics(entries, lo, hi, maxPoints), nil -} - -func downsampleMetrics(entries []MetricEntry, lo, hi int64, maxPoints int) []MetricEntry { - if maxPoints < 1 { - return entries - } - groups := map[string][]MetricEntry{} - order := []string{} - for _, e := range entries { - if _, ok := groups[e.Name]; !ok { - order = append(order, e.Name) - } - groups[e.Name] = append(groups[e.Name], e) - } - out := []MetricEntry{} - for _, name := range order { - out = append(out, downsampleGroup(groups[name], lo, hi, maxPoints)...) - } - sort.Slice(out, func(i, j int) bool { return out[i].TS < out[j].TS }) - return out -} - -func downsampleGroup(g []MetricEntry, lo, hi int64, maxPoints int) []MetricEntry { - if len(g) <= maxPoints { - return g - } - return alignGroup(g, lo, hi, maxPoints) -} - -// alignGroup downsamples g to at most maxPoints buckets whose boundaries are -// aligned to the [lo, hi] window, so all machines share the same bucket centers. -func alignGroup(g []MetricEntry, lo, hi int64, maxPoints int) []MetricEntry { - if len(g) == 0 { - return nil - } - if lo <= 0 { - lo = g[0].TS - } - if hi <= 0 { - hi = g[len(g)-1].TS - } - span := hi - lo - if span <= 0 { - span = 1 - } - bucketSize := float64(span) / float64(maxPoints) - buckets := make([][]MetricEntry, maxPoints) - for _, e := range g { - idx := int(float64(e.TS-lo) / bucketSize) - if idx >= maxPoints { - idx = maxPoints - 1 - } - if idx < 0 { - idx = 0 - } - buckets[idx] = append(buckets[idx], e) - } - res := make([]MetricEntry, 0, maxPoints) - for i, b := range buckets { - if len(b) == 0 { - continue - } - var sum float64 - for _, e := range b { - sum += e.Value - } - last := b[len(b)-1] - res = append(res, MetricEntry{ - TS: lo + int64((float64(i)+0.5)*bucketSize), - Name: last.Name, - Value: sum / float64(len(b)), - Unit: last.Unit, - }) - } - return res + return entries, rows.Err() } func (h *History) QueryMetricsAll(minTS, maxTS int64) ([]MetricAggregate, [2]int64, error) { diff --git a/server/index.go b/server/index.go new file mode 100644 index 0000000..e694518 --- /dev/null +++ b/server/index.go @@ -0,0 +1,103 @@ +package main + +import ( + "fmt" + "html" + "io" + "log" + "net/http" + "os" + "path/filepath" + "strings" +) + +// indexPlaceholder marks the spot in index.html where the server injects the +// title/description meta tags derived from the loaded config. +const indexPlaceholder = "" + +// maxSummaryNames caps how many host patterns a single group line shows before +// collapsing into "+N more". +const maxSummaryNames = 4 + +// machineSummary returns one line per configured group with its machine count +// and the distinct host patterns it was expanded from (falling back to machine +// names for top-level machines), mirroring the per-group recap printed by +// --check. +func machineSummary(cfg *Config) []string { + byGroup := map[string]int{} + patterns := map[string][]string{} + seen := map[string]map[string]bool{} + for _, m := range cfg.Machines { + byGroup[m.Group]++ + } + for _, g := range cfg.Groups { + seen[g.Name] = map[string]bool{} + for _, mc := range g.Machines { + if !seen[g.Name][mc.Host] { + seen[g.Name][mc.Host] = true + patterns[g.Name] = append(patterns[g.Name], mc.Host) + } + } + } + lines := make([]string, 0, len(cfg.Groups)+1) + for _, g := range cfg.Groups { + lines = append(lines, formatGroupSummary(g.Name, byGroup[g.Name], patterns[g.Name])) + } + if n := byGroup[""]; n > 0 { + names := []string{} + s := map[string]bool{} + for _, m := range cfg.Machines { + if m.Group == "" && !s[m.Name] { + s[m.Name] = true + names = append(names, m.Name) + } + } + lines = append(lines, formatGroupSummary("(top-level)", n, names)) + } + return lines +} + +func formatGroupSummary(name string, n int, items []string) string { + line := fmt.Sprintf("%s: %d", name, n) + if len(items) == 0 { + return line + } + show := items + extra := "" + if len(items) > maxSummaryNames { + show = items[:maxSummaryNames] + extra = fmt.Sprintf(" +%d more", len(items)-maxSummaryNames) + } + return fmt.Sprintf("%s (%s%s)", line, strings.Join(show, ", "), extra) +} + +func metaTags(title, description string) string { + if description == "" { + description = "status page" + } + return fmt.Sprintf( + "\n ", + html.EscapeString(title), html.EscapeString(description), + ) +} + +// indexHandler serves the built SPA, replacing indexPlaceholder in index.html +// with config-derived meta tags. All other paths are delegated to the static +// file server. +func indexHandler(dist, title, description string) http.Handler { + indexPath := filepath.Join(dist, "index.html") + index, err := os.ReadFile(indexPath) + if err != nil { + log.Fatalf("reading %s: %v", indexPath, err) + } + body := strings.Replace(string(index), indexPlaceholder, metaTags(title, description), 1) + fs := http.FileServer(http.Dir(dist)) + return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + if r.URL.Path != "/" && r.URL.Path != "/index.html" { + fs.ServeHTTP(w, r) + return + } + w.Header().Set("Content-Type", "text/html; charset=utf-8") + io.WriteString(w, body) + }) +} diff --git a/server/main.go b/server/main.go index af10d3e..3fe55dd 100644 --- a/server/main.go +++ b/server/main.go @@ -7,6 +7,7 @@ import ( "os" "os/exec" "path/filepath" + "strings" "github.com/joho/godotenv" "github.com/spf13/pflag" @@ -20,16 +21,9 @@ func envOr(key, def string) string { } func printConfigSummary(path string, cfg *Config) { - byGroup := map[string]int{} - for _, m := range cfg.Machines { - byGroup[m.Group]++ - } log.Printf("config %s ok: %d machines", path, len(cfg.Machines)) - for _, g := range cfg.Groups { - log.Printf(" %s: %d", g.Name, byGroup[g.Name]) - } - if n := byGroup[""]; n > 0 { - log.Printf(" (top-level): %d", n) + for _, line := range machineSummary(cfg) { + log.Printf(" %s", line) } } @@ -45,10 +39,6 @@ func buildFrontend() error { return c.Run() } -func spaHandler() http.Handler { - return http.FileServer(http.Dir("dist")) -} - func main() { _ = godotenv.Load() @@ -76,6 +66,14 @@ func main() { } } + cfg, err := LoadConfig(*configPath) + if err != nil { + log.Fatalf("loading config: %v", err) + } + log.Printf("loaded %d machines from %s", len(cfg.Machines), *configPath) + title := cfg.Title + desc := strings.Join(machineSummary(cfg), ", ") + mux := http.NewServeMux() if *mock { @@ -87,7 +85,7 @@ func main() { mux.HandleFunc("GET /api/metrics/{machine}", mm.metricsHandler) mux.HandleFunc("POST /api/refresh/{machine}", mm.refreshHandler) if !*devServer { - mux.Handle("GET /", spaHandler()) + mux.Handle("GET /", indexHandler("dist", title, desc)) } log.Printf("listening on %s (mock)", *addr) if err := http.ListenAndServe(*addr, mux); err != nil { @@ -102,12 +100,6 @@ func main() { } } - cfg, err := LoadConfig(*configPath) - if err != nil { - log.Fatalf("loading config: %v", err) - } - log.Printf("loaded %d machines from %s", len(cfg.Machines), *configPath) - ctx, cancel := context.WithCancel(context.Background()) defer cancel() @@ -127,7 +119,7 @@ func main() { mux.HandleFunc("GET /api/metrics/{machine}", mon.metricsHandler) mux.HandleFunc("POST /api/refresh/{machine}", mon.refreshHandler) if !*devServer { - mux.Handle("GET /", spaHandler()) + mux.Handle("GET /", indexHandler("dist", title, desc)) } log.Printf("listening on %s", *addr) diff --git a/server/mock.go b/server/mock.go index 85d797d..4508f57 100644 --- a/server/mock.go +++ b/server/mock.go @@ -136,10 +136,25 @@ func (mm *MockMonitor) sshResult(m mockMachine, now time.Time) *sshResultState { return &sshResultState{Ok: true, Result: sb.String(), ExitCode: nil, LastRun: now} } +func mockUptime(m mockMachine) *float64 { + switch m.Status { + case statusDown: + u := 55.0 + float64(mockSeed(m.ID)%20) + return &u + case statusDegraded: + u := 92.0 + float64(mockSeed(m.ID)%7) + return &u + default: + u := 99.0 + float64(mockSeed(m.ID)%10)/10 + return &u + } +} + func (mm *MockMonitor) machineStatus(m mockMachine, now time.Time) machineStatus { ms := machineStatus{ ID: m.ID, Name: m.Name, Host: m.Host, Status: m.Status, LastPing: &now, IP: m.IP, IPs: []string{m.IP}, + Uptime: mockUptime(m), } switch m.Status { case statusDown: diff --git a/server/monitor.go b/server/monitor.go index 00d439b..b152379 100644 --- a/server/monitor.go +++ b/server/monitor.go @@ -77,6 +77,7 @@ type machineStatus struct { ICMP string `json:"icmp"` TCP string `json:"tcp"` LastPing *time.Time `json:"lastPing"` + Uptime *float64 `json:"uptime,omitempty"` SSHConfigured bool `json:"sshConfigured"` SSH *sshResultState `json:"ssh"` Metrics []metricStatus `json:"metrics"` @@ -648,6 +649,13 @@ func (m *Monitor) StatusPayload() statusPayload { ms.ICMP = checkNA ms.TCP = checkNA } + if m.history != nil { + up, down := m.history.Uptime(mc.ID) + if up+down > 0 { + u := float64(up) / float64(up+down) * 100 + ms.Uptime = &u + } + } if group == "" { p.Machines = append(p.Machines, ms) return @@ -703,16 +711,6 @@ func (m *Monitor) metricsHandler(w http.ResponseWriter, r *http.Request) { return } q := r.URL.Query() - maxPoints := 100 - if v := q.Get("max_points"); v != "" { - if n, err := strconv.Atoi(v); err == nil { - maxPoints = n - } - } else if v := q.Get("limit"); v != "" { - if n, err := strconv.Atoi(v); err == nil { - maxPoints = n - } - } var minTS, maxTS int64 if v := q.Get("min"); v != "" { minTS, _ = strconv.ParseInt(v, 10, 64) @@ -720,7 +718,7 @@ func (m *Monitor) metricsHandler(w http.ResponseWriter, r *http.Request) { if v := q.Get("max"); v != "" { maxTS, _ = strconv.ParseInt(v, 10, 64) } - entries, err := m.history.QueryMetrics(machine, q.Get("name"), minTS, maxTS, maxPoints, q.Get("shared") != "") + entries, err := m.history.QueryMetrics(machine, q.Get("name"), minTS, maxTS) if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) return diff --git a/src/components/Grid.tsx b/src/components/Grid.tsx index e8eccb1..eb74cc6 100644 --- a/src/components/Grid.tsx +++ b/src/components/Grid.tsx @@ -51,6 +51,12 @@ export function MachineCard({ {m.sshConfigured && } + {m.uptime !== undefined && ( + = 99 ? 'ok' : 'fail'} + /> + )}
e.stopPropagation()}> @@ -64,6 +70,12 @@ export function MachineCard({ last ping {formatTime(m.lastPing)} + + uptime + + {m.uptime !== undefined ? `${m.uptime.toFixed(1)}%` : '—'} + +
diff --git a/src/components/StatusModal.tsx b/src/components/StatusModal.tsx index 05889a3..8df1c08 100644 --- a/src/components/StatusModal.tsx +++ b/src/components/StatusModal.tsx @@ -57,11 +57,10 @@ export function StatusModal({ setMetricEntries(null) const nowS = Math.floor(Date.now() / 1000) const min = nowS - 86400 - const shared = sharedMetricWindow ? '&shared=1' : '' Promise.all( metricNames.map((name) => fetch( - `/api/metrics/${encodeURIComponent(m.id)}?name=${encodeURIComponent(name)}&min=${min}&max=${nowS}&max_points=100${shared}` + `/api/metrics/${encodeURIComponent(m.id)}?name=${encodeURIComponent(name)}&min=${min}&max=${nowS}` ).then((r) => (r.ok ? r.json() : Promise.reject(new Error(`HTTP ${r.status}`)))) ) ) @@ -78,7 +77,7 @@ export function StatusModal({ return () => { cancelled = true } - }, [m.id, metricNames.join(','), sharedMetricWindow]) + }, [m.id, metricNames.join(',')]) const metricPlots = (() => { if (!metricEntries) return null @@ -247,24 +246,23 @@ export function StatusModal({ {hist && hist.length > 0 && (
- {ticks.map((e, i) => { - const step = Math.max(1, Math.round(ticks.length / 4)) - const mark = - i === 0 || - i === ticks.length - 1 || - i % step === 0 - return ( - - - {mark && ( - {formatTickShort(e.ts)} - )} - - ) - })} + {(() => { + const last = ticks.length - 1 + return ticks.map((e, i) => { + const mark = (last - i) % 20 === 0 + return ( + + + {mark && ( + {formatTickShort(e.ts)} + )} + + ) + }) + })()}
)} {hist && hist.length > 0 && ( diff --git a/src/dark.css b/src/dark.css index 25b374d..9d17718 100644 --- a/src/dark.css +++ b/src/dark.css @@ -41,7 +41,7 @@ --border-ink: #3a403a; --border: 2px solid var(--border-ink); --border-muted: #3a403a; - --shadow-color: #000; + --shadow-color: #0004; --shadow: 0.35rem 0.35rem 0 var(--shadow-color); --shadow-lg: 0.55rem 0.55rem 0 var(--shadow-color); --overlay: rgba(0, 0, 0, 0.6); diff --git a/src/light.css b/src/light.css index 20fbdd3..0b31099 100644 --- a/src/light.css +++ b/src/light.css @@ -45,7 +45,7 @@ --border-ink: #111; --border: 2px solid var(--border-ink); --border-muted: #bbb; - --shadow-color: #888; + --shadow-color: #0004; --shadow: 0.35rem 0.35rem 0 var(--shadow-color); --shadow-lg: 0.55rem 0.55rem 0 var(--shadow-color); --overlay: rgba(0, 0, 0, 0.35); diff --git a/src/types.ts b/src/types.ts index 8e51b16..cad7dc2 100644 --- a/src/types.ts +++ b/src/types.ts @@ -33,6 +33,7 @@ export interface Machine { icmp: Check tcp: Check lastPing: string | null + uptime?: number sshConfigured: boolean ssh: SshResult | null metrics: MetricStatus[]