diff --git a/.commitlintrc.js b/.commitlintrc.js
index 84dcb12..0025947 100644
--- a/.commitlintrc.js
+++ b/.commitlintrc.js
@@ -1,3 +1,23 @@
module.exports = {
extends: ['@commitlint/config-conventional'],
+ rules: {
+ 'body-max-line-length': [0],
+ 'type-enum': [2, 'always', [
+ 'build',
+ 'chore',
+ 'ci',
+ 'docs',
+ 'wiki',
+ 'feat',
+ 'fix',
+ 'perf',
+ 'refactor',
+ 'revert',
+ 'style',
+ 'test',
+ 'vibe'
+ 'plan'
+ 'vision',
+ ]],
+ },
};
diff --git a/README.md b/README.md
index 859e304..4016519 100644
--- a/README.md
+++ b/README.md
@@ -48,9 +48,20 @@ export PSSTD_GOSSIP=":7947" # peer sync listen address, de
export PSSTD_SEEDS="10.0.1.20:7946,10.0.1.21:7946" # explicit peer sync addresses
export PSSTD_DB="./data" # local state directory
export PSSTD_WEB="true" # set false for sync-only nodes
+export PSSTD_NODE_NAME="rack-a-01" # optional stable node identity override
+export PSSTD_NODE_TTL="15s" # how long this node's heartbeat stays online
./psstd
```
+By default, psstd uses the OS hostname as the node identity. Set
+`PSSTD_NODE_NAME` for cloned hosts, containers, or multiple test instances that
+would otherwise publish the same hostname. The override must be non-empty and
+must not contain whitespace.
+
+Each node publishes its own heartbeat TTL with `PSSTD_NODE_TTL`. Shorter values
+make stale/offline indication react faster; longer values are better for slow or
+lossy networks. The default is `15s`, and values must be at least `2s`.
+
## Discovery
| Environment | How nodes find each other |
diff --git a/discovery.go b/discovery.go
index c2730e5..a3b6209 100644
--- a/discovery.go
+++ b/discovery.go
@@ -63,7 +63,6 @@ func discoverPeers() []string {
continue
}
peer := fmt.Sprintf("%s:%d", addr.String(), entry.Port)
- log.Printf("mDNS: discovered peer %s (%s)", entry.Name, peer)
peers = append(peers, peer)
}
return peers
diff --git a/gossip.go b/gossip.go
index 1bbb0f4..6fce1ea 100644
--- a/gossip.go
+++ b/gossip.go
@@ -5,7 +5,6 @@ import (
"errors"
"log"
"sync"
- "time"
"github.com/cockroachdb/pebble/v2"
"github.com/hashicorp/memberlist"
@@ -14,14 +13,15 @@ import (
var appVersion = "dev"
type NodeStats struct {
- Name string `json:"name"`
- Version string `json:"version,omitempty"`
- WebURL string `json:"web,omitempty"`
- CPU []float64 `json:"cpu"`
- MemUsed uint64 `json:"mu"`
- MemTotal uint64 `json:"mt"`
- Load [3]float64 `json:"ld"`
- UpdatedAt int64 `json:"ts"` // unix nano, LWW key
+ Name string `json:"name"`
+ Version string `json:"version,omitempty"`
+ WebURL string `json:"web,omitempty"`
+ TTLSeconds int `json:"ttl,omitempty"`
+ CPU []float64 `json:"cpu"`
+ MemUsed uint64 `json:"mu"`
+ MemTotal uint64 `json:"mt"`
+ Load [3]float64 `json:"ld"`
+ UpdatedAt int64 `json:"ts"` // unix nano, LWW key
}
var errStaleVersion = errors.New("stale version")
@@ -107,7 +107,7 @@ func purgeOfflineDifferentVersion(db *pebble.DB, version string) error {
}
func nodeRecordOffline(s NodeStats) bool {
- return s.UpdatedAt == 0 || time.Since(time.Unix(0, s.UpdatedAt)) > 15*time.Second
+ return nodeHealth(s).State == healthOffline
}
// ── Delegate ──────────────────────────────────────────────────────────────────
diff --git a/main.go b/main.go
index 6c7ae25..0576ef5 100644
--- a/main.go
+++ b/main.go
@@ -2,6 +2,7 @@ package main
import (
"encoding/json"
+ "flag"
"fmt"
"log"
"net"
@@ -15,19 +16,34 @@ import (
"github.com/hashicorp/memberlist"
)
+type cliOptions struct {
+ List bool
+}
+
const (
- envDB = "PSSTD_DB"
- envHTTP = "PSSTD_HTTP"
- envGossip = "PSSTD_GOSSIP"
- envSeeds = "PSSTD_SEEDS"
- envHTTPAd = "PSSTD_ADVERTISE_HTTP"
- envWeb = "PSSTD_WEB" // "true" to enable HTTP, default true
- gossipPort = 7946
- httpPort = 8080
+ envDB = "PSSTD_DB"
+ envHTTP = "PSSTD_HTTP"
+ envGossip = "PSSTD_GOSSIP"
+ envSeeds = "PSSTD_SEEDS"
+ envHTTPAd = "PSSTD_ADVERTISE_HTTP"
+ envWeb = "PSSTD_WEB" // "true" to enable HTTP, default true
+ envNodeName = "PSSTD_NODE_NAME"
+ envNodeTTL = "PSSTD_NODE_TTL"
+ gossipPort = 7946
+ httpPort = 8080
)
func main() {
+ opts := parseCLI(os.Args[1:])
hostname, _ := os.Hostname()
+ nodeName, err := nodeNameFromEnv(hostname)
+ if err != nil {
+ log.Fatalf("node name: %v", err)
+ }
+ nodeTTL, err := nodeTTLFromEnv()
+ if err != nil {
+ log.Fatalf("node ttl: %v", err)
+ }
dbPath := envOr(envDB, "./data")
httpAddr := envOr(envHTTP, fmt.Sprintf(":%d", httpPort))
@@ -47,7 +63,7 @@ func main() {
if err != nil {
if pebbleLockHeld(err) {
log.Printf("psstd already appears to own %s; starting terminal mirror instead", dbPath)
- runTerminalMirror(hostname, gossipAddr, seeds)
+ runTerminalMirror(nodeName, gossipAddr, seeds, opts.List)
return
}
log.Fatalf("pebble open: %v", err)
@@ -65,7 +81,7 @@ func main() {
delegate := newKVDelegate(db, appVersion)
cfg := memberlist.DefaultLANConfig()
- cfg.Name = hostname
+ cfg.Name = nodeName
cfg.BindAddr, cfg.BindPort = splitHostPort(gossipAddr)
cfg.Delegate = delegate
cfg.Events = newEventDelegate(db, appVersion)
@@ -79,7 +95,7 @@ func main() {
}
db = nil
log.Printf("psstd already appears to be listening on %s; starting terminal mirror instead", gossipAddr)
- runTerminalMirror(hostname, gossipAddr, seeds)
+ runTerminalMirror(nodeName, gossipAddr, seeds, opts.List)
return
}
log.Fatalf("memberlist create: %v", err)
@@ -88,39 +104,87 @@ func main() {
// ── Discovery ────────────────────────────────────────────────────────────
// 1. Register ourselves via mDNS so peers can find us on LAN
- stopMDNS := registerMDNS(hostname, cfg.BindPort)
+ stopMDNS := registerMDNS(nodeName, cfg.BindPort)
defer stopMDNS()
// 2. Scan for existing peers (mDNS + any explicit seeds)
discovered := discoverPeers()
allSeeds := append(seeds, discovered...)
+ joinedPeers := 0
+ joinErr := error(nil)
if len(allSeeds) > 0 {
- if n, err := list.Join(allSeeds); err != nil {
- log.Printf("join warning (joined %d): %v", n, err)
- } else {
- log.Printf("joined cluster, %d peer(s)", n)
- }
- } else {
- log.Println("no peers found — running solo, will be discovered by others")
+ joinedPeers, joinErr = list.Join(allSeeds)
}
+ logStartupConfig(startupConfig{
+ NodeName: nodeName,
+ DBPath: dbPath,
+ HTTPAddr: httpAddr,
+ WebURL: webURL,
+ GossipAddr: gossipAddr,
+ WebEnabled: webEnabled,
+ Version: appVersion,
+ NodeTTL: nodeTTL,
+ SeedCount: len(seeds),
+ MDNSCount: len(discovered),
+ JoinedPeers: joinedPeers,
+ JoinErr: joinErr,
+ })
// ── Stats heartbeat ─────────────────────────────────────────────────────
- go statsLoop(hostname, webURL, appVersion, db, delegate)
+ go statsLoop(nodeName, webURL, appVersion, nodeTTL, db, delegate)
// ── HTTP ─────────────────────────────────────────────────────────────────
if webEnabled {
mux := http.NewServeMux()
- mux.HandleFunc("/", makeHandler(db, hostname))
- log.Printf("psstd version=%s node=%s http=%s advertise=%s gossip=%s web=true", appVersion, hostname, httpAddr, webURL, gossipAddr)
+ mux.HandleFunc("/", makeHandler(db, nodeName))
if err := http.ListenAndServe(httpAddr, mux); err != nil {
log.Fatalf("http: %v", err)
}
} else {
- log.Printf("psstd version=%s node=%s gossip=%s web=false", appVersion, hostname, gossipAddr)
select {} // block forever
}
}
+func parseCLI(args []string) cliOptions {
+ fs := flag.NewFlagSet("psstd", flag.ExitOnError)
+ fs.SetOutput(os.Stderr)
+ var opts cliOptions
+ fs.BoolVar(&opts.List, "l", false, "render terminal mirror as a vertical node list")
+ fs.BoolVar(&opts.List, "list", false, "render terminal mirror as a vertical node list")
+ _ = fs.Parse(args)
+ return opts
+}
+
+type startupConfig struct {
+ NodeName string
+ DBPath string
+ HTTPAddr string
+ WebURL string
+ GossipAddr string
+ WebEnabled bool
+ Version string
+ NodeTTL time.Duration
+ SeedCount int
+ MDNSCount int
+ JoinedPeers int
+ JoinErr error
+}
+
+func logStartupConfig(cfg startupConfig) {
+ log.Print(startupSummary(cfg))
+}
+
+func startupSummary(cfg startupConfig) string {
+ join := "solo"
+ if cfg.JoinErr != nil {
+ join = fmt.Sprintf("warning joined=%d error=%q", cfg.JoinedPeers, cfg.JoinErr)
+ } else if cfg.JoinedPeers > 0 {
+ join = fmt.Sprintf("joined=%d", cfg.JoinedPeers)
+ }
+ return fmt.Sprintf("psstd startup: version=%s node=%s db=%s web=%t http=%s advertise=%s gossip=%s ttl=%s seeds=%d mdns=%d join=%s",
+ cfg.Version, cfg.NodeName, cfg.DBPath, cfg.WebEnabled, cfg.HTTPAddr, cfg.WebURL, cfg.GossipAddr, cfg.NodeTTL, cfg.SeedCount, cfg.MDNSCount, join)
+}
+
func pebbleLockHeld(err error) bool {
msg := strings.ToLower(err.Error())
return strings.Contains(msg, "lock") &&
@@ -137,7 +201,7 @@ func addressInUse(err error) bool {
strings.Contains(msg, "bind: only one usage of each socket address")
}
-func runTerminalMirror(hostname, gossipAddr string, seeds []string) {
+func runTerminalMirror(hostname, gossipAddr string, seeds []string, listMode bool) {
tmpDir, err := os.MkdirTemp("", "psstd-view-*")
if err != nil {
log.Fatalf("terminal mirror temp db: %v", err)
@@ -172,7 +236,7 @@ func runTerminalMirror(hostname, gossipAddr string, seeds []string) {
log.Printf("terminal mirror joined cluster, %d peer(s)", n)
}
- terminalRenderLoop(db)
+ terminalRenderLoop(db, listMode)
}
func terminalMirrorSeeds(gossipAddr string, seeds []string) []string {
@@ -188,11 +252,11 @@ func terminalMirrorSeeds(gossipAddr string, seeds []string) []string {
// ── Stats loop ───────────────────────────────────────────────────────────────
-func statsLoop(hostname, webURL, version string, db *pebble.DB, d *kvDelegate) {
+func statsLoop(hostname, webURL, version string, ttl time.Duration, db *pebble.DB, d *kvDelegate) {
ticker := time.NewTicker(2 * time.Second)
defer ticker.Stop()
for range ticker.C {
- stats, err := collectStats(hostname, webURL, version)
+ stats, err := collectStats(hostname, webURL, version, ttl)
if err != nil {
log.Printf("stats error: %v", err)
continue
@@ -255,6 +319,38 @@ func envOr(key, def string) string {
return def
}
+func nodeNameFromEnv(hostname string) (string, error) {
+ override, ok := os.LookupEnv(envNodeName)
+ if !ok || override == "" {
+ if strings.TrimSpace(hostname) == "" {
+ return "", fmt.Errorf("hostname is empty; set %s", envNodeName)
+ }
+ return hostname, nil
+ }
+ if strings.TrimSpace(override) != override || override == "" {
+ return "", fmt.Errorf("%s must not be empty or have leading/trailing whitespace", envNodeName)
+ }
+ if strings.ContainsAny(override, " \t\r\n") {
+ return "", fmt.Errorf("%s must not contain whitespace", envNodeName)
+ }
+ return override, nil
+}
+
+func nodeTTLFromEnv() (time.Duration, error) {
+ value, ok := os.LookupEnv(envNodeTTL)
+ if !ok || value == "" {
+ return defaultNodeTTL, nil
+ }
+ ttl, err := time.ParseDuration(value)
+ if err != nil {
+ return 0, fmt.Errorf("%s must be a duration such as 15s or 1m: %w", envNodeTTL, err)
+ }
+ if ttl < 2*time.Second {
+ return 0, fmt.Errorf("%s must be at least 2s", envNodeTTL)
+ }
+ return ttl, nil
+}
+
func splitCSV(s string) []string {
var out []string
for _, p := range strings.Split(s, ",") {
diff --git a/main_test.go b/main_test.go
new file mode 100644
index 0000000..f37d62b
--- /dev/null
+++ b/main_test.go
@@ -0,0 +1,133 @@
+package main
+
+import (
+ "errors"
+ "os"
+ "strings"
+ "testing"
+ "time"
+)
+
+func withoutNodeNameEnv(t *testing.T) {
+ t.Helper()
+ old, hadOld := os.LookupEnv(envNodeName)
+ if err := os.Unsetenv(envNodeName); err != nil {
+ t.Fatalf("unset %s: %v", envNodeName, err)
+ }
+ t.Cleanup(func() {
+ if hadOld {
+ _ = os.Setenv(envNodeName, old)
+ } else {
+ _ = os.Unsetenv(envNodeName)
+ }
+ })
+}
+
+func TestNodeNameFromEnvUsesHostnameByDefault(t *testing.T) {
+ withoutNodeNameEnv(t)
+
+ got, err := nodeNameFromEnv("host-a")
+ if err != nil {
+ t.Fatalf("nodeNameFromEnv: %v", err)
+ }
+ if got != "host-a" {
+ t.Fatalf("name = %q, want host-a", got)
+ }
+}
+
+func TestNodeNameFromEnvUsesOverride(t *testing.T) {
+ t.Setenv(envNodeName, "test-node")
+
+ got, err := nodeNameFromEnv("host-a")
+ if err != nil {
+ t.Fatalf("nodeNameFromEnv: %v", err)
+ }
+ if got != "test-node" {
+ t.Fatalf("name = %q, want test-node", got)
+ }
+}
+
+func TestNodeNameFromEnvRejectsWhitespaceOverride(t *testing.T) {
+ for _, value := range []string{" test-node", "test-node ", "test node", "\t"} {
+ t.Run(value, func(t *testing.T) {
+ t.Setenv(envNodeName, value)
+ if got, err := nodeNameFromEnv("host-a"); err == nil {
+ t.Fatalf("name = %q, want error", got)
+ }
+ })
+ }
+}
+
+func TestNodeTTLFromEnvDefaultAndOverride(t *testing.T) {
+ t.Setenv(envNodeTTL, "")
+ got, err := nodeTTLFromEnv()
+ if err != nil {
+ t.Fatalf("default ttl: %v", err)
+ }
+ if got != defaultNodeTTL {
+ t.Fatalf("default ttl = %s, want %s", got, defaultNodeTTL)
+ }
+
+ t.Setenv(envNodeTTL, "45s")
+ got, err = nodeTTLFromEnv()
+ if err != nil {
+ t.Fatalf("override ttl: %v", err)
+ }
+ if got != 45*time.Second {
+ t.Fatalf("override ttl = %s, want 45s", got)
+ }
+}
+
+func TestNodeTTLFromEnvRejectsInvalidValues(t *testing.T) {
+ for _, value := range []string{"soon", "1s"} {
+ t.Run(value, func(t *testing.T) {
+ t.Setenv(envNodeTTL, value)
+ if got, err := nodeTTLFromEnv(); err == nil {
+ t.Fatalf("ttl = %s, want error", got)
+ }
+ })
+ }
+}
+
+func TestStartupSummaryIncludesJoinOutcome(t *testing.T) {
+ base := startupConfig{
+ Version: "v1",
+ NodeName: "node-a",
+ DBPath: "./data",
+ HTTPAddr: ":8080",
+ WebURL: "http://node-a:8080",
+ GossipAddr: ":7946",
+ WebEnabled: true,
+ NodeTTL: 15 * time.Second,
+ SeedCount: 2,
+ MDNSCount: 1,
+ }
+
+ joined := base
+ joined.JoinedPeers = 3
+ if got := startupSummary(joined); !strings.Contains(got, "join=joined=3") {
+ t.Fatalf("joined summary missing outcome: %s", got)
+ }
+
+ solo := base
+ if got := startupSummary(solo); !strings.Contains(got, "join=solo") {
+ t.Fatalf("solo summary missing outcome: %s", got)
+ }
+
+ warn := base
+ warn.JoinedPeers = 1
+ warn.JoinErr = errors.New("partial join")
+ if got := startupSummary(warn); !strings.Contains(got, `join=warning joined=1 error="partial join"`) {
+ t.Fatalf("warning summary missing outcome: %s", got)
+ }
+}
+
+func TestParseCLIListFlag(t *testing.T) {
+ for _, args := range [][]string{{"-l"}, {"--list"}} {
+ t.Run(strings.Join(args, " "), func(t *testing.T) {
+ if opts := parseCLI(args); !opts.List {
+ t.Fatalf("List = false, want true")
+ }
+ })
+ }
+}
diff --git a/render.go b/render.go
index baf5d1b..f6061f5 100644
--- a/render.go
+++ b/render.go
@@ -21,6 +21,8 @@ import (
"github.com/cockroachdb/pebble/v2"
)
+const defaultNodeTTL = 15 * time.Second
+
//go:embed templates/dashboard.html
var templateFS embed.FS
@@ -34,7 +36,7 @@ func init() {
// ── Stats collection ──────────────────────────────────────────────────────────
-func collectStats(hostname, webURL, version string) (NodeStats, error) {
+func collectStats(hostname, webURL, version string, ttl time.Duration) (NodeStats, error) {
cpuPcts, err := cpu.Percent(200*time.Millisecond, true)
if err != nil {
return NodeStats{}, err
@@ -48,14 +50,15 @@ func collectStats(hostname, webURL, version string) (NodeStats, error) {
return NodeStats{}, err
}
return NodeStats{
- Name: hostname,
- Version: version,
- WebURL: webURL,
- CPU: cpuPcts,
- MemUsed: vmStat.Used,
- MemTotal: vmStat.Total,
- Load: [3]float64{loadStat.Load1, loadStat.Load5, loadStat.Load15},
- UpdatedAt: time.Now().UnixNano(),
+ Name: hostname,
+ Version: version,
+ WebURL: webURL,
+ TTLSeconds: int(ttl / time.Second),
+ CPU: cpuPcts,
+ MemUsed: vmStat.Used,
+ MemTotal: vmStat.Total,
+ Load: [3]float64{loadStat.Load1, loadStat.Load5, loadStat.Load15},
+ UpdatedAt: time.Now().UnixNano(),
}, nil
}
@@ -91,6 +94,46 @@ var (
}
)
+type healthState string
+
+const (
+ healthFresh healthState = "fresh"
+ healthStale healthState = "stale"
+ healthOffline healthState = "offline"
+)
+
+type nodeHealthInfo struct {
+ State healthState
+ Age time.Duration
+ TTL time.Duration
+}
+
+func nodeTTL(s NodeStats) time.Duration {
+ if s.TTLSeconds > 0 {
+ return time.Duration(s.TTLSeconds) * time.Second
+ }
+ return defaultNodeTTL
+}
+
+func nodeHealth(s NodeStats) nodeHealthInfo {
+ ttl := nodeTTL(s)
+ if s.UpdatedAt == 0 {
+ return nodeHealthInfo{State: healthOffline, TTL: ttl}
+ }
+ age := time.Since(time.Unix(0, s.UpdatedAt))
+ if age > ttl {
+ return nodeHealthInfo{State: healthOffline, Age: age, TTL: ttl}
+ }
+ if age > ttl/2 {
+ return nodeHealthInfo{State: healthStale, Age: age, TTL: ttl}
+ }
+ return nodeHealthInfo{State: healthFresh, Age: age, TTL: ttl}
+}
+
+func nodeOnline(s NodeStats) bool {
+ return nodeHealth(s).State != healthOffline
+}
+
func pctBar(pct float64, width int, segments []barSegment) string {
filled := int(math.Round(pct / 100.0 * float64(width)))
if filled > width {
@@ -120,20 +163,26 @@ func segmentStyle(pct float64, segments []barSegment) lipgloss.Style {
func renderANSI(s NodeStats) string {
var sb strings.Builder
- age := time.Since(time.Unix(0, s.UpdatedAt))
- offline := s.UpdatedAt == 0 || age > 15*time.Second
+ health := nodeHealth(s)
status := styleGreen.Render("●")
- if offline {
+ if health.State == healthStale {
+ status = styleYellow.Render("●")
+ }
+ if health.State == healthOffline {
status = styleRed.Render("●")
}
sb.WriteString(fmt.Sprintf("%s %s\n", status, s.Name))
- if offline {
+ if health.State == healthOffline {
sb.WriteString(styleDim.Render(" offline"))
sb.WriteByte('\n')
return sb.String()
}
- sb.WriteString(fmt.Sprintf(" updated %.0fs ago\n", age.Seconds()))
+ if health.State == healthStale {
+ sb.WriteString(fmt.Sprintf(" stale %.0fs ago\n", health.Age.Seconds()))
+ } else {
+ sb.WriteString(fmt.Sprintf(" updated %.0fs ago\n", health.Age.Seconds()))
+ }
sb.WriteString(styleDim.Render(strings.Repeat("─", barWidth+14)))
sb.WriteByte('\n')
@@ -211,8 +260,14 @@ func avgCPU(s NodeStats) float64 {
return sum / float64(len(s.CPU))
}
-func nodeOnline(s NodeStats) bool {
- return s.UpdatedAt != 0 && time.Since(time.Unix(0, s.UpdatedAt)) <= 15*time.Second
+func maxCPU(s NodeStats) float64 {
+ max := 0.0
+ for _, v := range s.CPU {
+ if v > max {
+ max = v
+ }
+ }
+ return max
}
func computeRefreshIntervalMs(nodes []NodeStats) int {
@@ -265,6 +320,51 @@ func findBestNodeHint(nodes []NodeStats) string {
return fmt.Sprintf("lowest-load node: %s (%.0f%% cpu, %.2f load)", best.Name, avgCPU(*best), best.Load[0])
}
+type clusterSummary struct {
+ Fresh int
+ Stale int
+ Offline int
+ Hottest string
+ HotCPU float64
+ HotLoad float64
+ HasHot bool
+}
+
+func summarizeCluster(nodes []NodeStats) clusterSummary {
+ var summary clusterSummary
+ hotScore := -1.0
+ for _, s := range nodes {
+ switch nodeHealth(s).State {
+ case healthFresh:
+ summary.Fresh++
+ case healthStale:
+ summary.Stale++
+ continue
+ default:
+ summary.Offline++
+ continue
+ }
+ cpu := avgCPU(s)
+ score := cpu + s.Load[0]*10
+ if score > hotScore {
+ hotScore = score
+ summary.Hottest = s.Name
+ summary.HotCPU = cpu
+ summary.HotLoad = s.Load[0]
+ summary.HasHot = true
+ }
+ }
+ return summary
+}
+
+func (s clusterSummary) TerminalHeader() string {
+ hot := "hottest: none"
+ if s.HasHot {
+ hot = fmt.Sprintf("hottest: %s %.0f%% cpu %.2f load", s.Hottest, s.HotCPU, s.HotLoad)
+ }
+ return fmt.Sprintf("online %d - stale %d - offline %d - %s", s.Fresh, s.Stale, s.Offline, hot)
+}
+
func nodeScore(s NodeStats) float64 {
return avgCPU(s) + s.Load[0]*10
}
@@ -360,11 +460,20 @@ func displayQuery(r *http.Request, winW, winH int) url.Values {
// ── HTTP handler ──────────────────────────────────────────────────────────────
type cellData struct {
- Name string
- URL string
- HTML template.HTML
- Offline bool
- Link bool
+ Name string
+ URL string
+ HTML template.HTML
+ State healthState
+ CPUAvg float64
+ CPUMax float64
+ MemPct float64
+ MemUsed uint64
+ MemTotal uint64
+ Load1 float64
+ Load5 float64
+ Load15 float64
+ Age float64
+ Link bool
}
type pageData struct {
@@ -374,6 +483,7 @@ type pageData struct {
RefreshLabel string
RefreshURL string
BestHint string
+ Summary clusterSummary
}
func makeHandler(db *pebble.DB, selfName string) http.HandlerFunc {
@@ -392,18 +502,31 @@ func makeHandler(db *pebble.DB, selfName string) http.HandlerFunc {
cells := make([]cellData, 0, len(nodes))
for _, s := range nodes {
htmlBytes := ansihtml.ConvertToHTML([]byte(renderANSI(s)))
- offline := s.UpdatedAt == 0 || time.Since(time.Unix(0, s.UpdatedAt)) > 15*time.Second
+ health := nodeHealth(s)
nodeURL := ""
if s.WebURL != "" {
nodeURL = pageURL(s.WebURL, displayQuery(r, winW, winH))
}
+ memPct := 0.0
+ if s.MemTotal > 0 {
+ memPct = float64(s.MemUsed) / float64(s.MemTotal) * 100
+ }
cells = append(cells, cellData{
- Name: s.Name,
- URL: nodeURL,
- HTML: template.HTML(htmlBytes),
- Offline: offline,
- Link: nodeURL != "",
+ Name: s.Name,
+ URL: nodeURL,
+ HTML: template.HTML(htmlBytes),
+ State: health.State,
+ CPUAvg: avgCPU(s),
+ CPUMax: maxCPU(s),
+ MemPct: memPct,
+ MemUsed: s.MemUsed,
+ MemTotal: s.MemTotal,
+ Load1: s.Load[0],
+ Load5: s.Load[1],
+ Load15: s.Load[2],
+ Age: health.Age.Seconds(),
+ Link: nodeURL != "",
})
}
@@ -423,6 +546,7 @@ func makeHandler(db *pebble.DB, selfName string) http.HandlerFunc {
RefreshLabel: fmt.Sprintf("%.1fs", float64(refreshMs)/1000),
RefreshURL: refreshURL,
BestHint: bestHint,
+ Summary: summarizeCluster(nodes),
}); err != nil {
http.Error(w, "template error", 500)
return
diff --git a/render_test.go b/render_test.go
index 51330f1..8393716 100644
--- a/render_test.go
+++ b/render_test.go
@@ -53,3 +53,115 @@ func TestDashboardDoesNotLinkNodesWithoutWebURL(t *testing.T) {
t.Fatalf("web node was not rendered as a link:\n%s", body)
}
}
+
+func TestNodeHealthDistinguishesFreshStaleOffline(t *testing.T) {
+ now := time.Now()
+ cases := []struct {
+ name string
+ node NodeStats
+ want healthState
+ }{
+ {
+ name: "fresh",
+ node: NodeStats{UpdatedAt: now.Add(-2 * time.Second).UnixNano(), TTLSeconds: 10},
+ want: healthFresh,
+ },
+ {
+ name: "stale",
+ node: NodeStats{UpdatedAt: now.Add(-7 * time.Second).UnixNano(), TTLSeconds: 10},
+ want: healthStale,
+ },
+ {
+ name: "offline",
+ node: NodeStats{UpdatedAt: now.Add(-11 * time.Second).UnixNano(), TTLSeconds: 10},
+ want: healthOffline,
+ },
+ {
+ name: "zero timestamp offline",
+ node: NodeStats{},
+ want: healthOffline,
+ },
+ }
+ for _, tc := range cases {
+ t.Run(tc.name, func(t *testing.T) {
+ if got := nodeHealth(tc.node).State; got != tc.want {
+ t.Fatalf("state = %s, want %s", got, tc.want)
+ }
+ })
+ }
+}
+
+func TestSummarizeClusterCountsStatesAndHottestOnlineNode(t *testing.T) {
+ now := time.Now()
+ summary := summarizeCluster([]NodeStats{
+ {Name: "fresh-hot", UpdatedAt: now.UnixNano(), TTLSeconds: 10, CPU: []float64{20}, Load: [3]float64{0.5}},
+ {Name: "stale-hot", UpdatedAt: now.Add(-7 * time.Second).UnixNano(), TTLSeconds: 10, CPU: []float64{70}, Load: [3]float64{1.2}},
+ {Name: "offline", UpdatedAt: 0, TTLSeconds: 10, CPU: []float64{99}, Load: [3]float64{9}},
+ })
+
+ if summary.Fresh != 1 || summary.Stale != 1 || summary.Offline != 1 {
+ t.Fatalf("counts = fresh %d stale %d offline %d, want 1/1/1", summary.Fresh, summary.Stale, summary.Offline)
+ }
+ if !summary.HasHot || summary.Hottest != "fresh-hot" {
+ t.Fatalf("hottest = %q has=%t, want fresh-hot", summary.Hottest, summary.HasHot)
+ }
+}
+
+func TestDashboardRendersClusterSummaryAndStateData(t *testing.T) {
+ db := openTestDB(t)
+ now := time.Now()
+ nodes := []NodeStats{
+ {Name: "fresh", Version: appVersion, UpdatedAt: now.UnixNano(), TTLSeconds: 10, CPU: []float64{20}, MemTotal: 100, MemUsed: 30},
+ {Name: "stale", Version: appVersion, UpdatedAt: now.Add(-7 * time.Second).UnixNano(), TTLSeconds: 10, CPU: []float64{80}, MemTotal: 100, MemUsed: 40},
+ {Name: "offline", Version: appVersion, UpdatedAt: 0, TTLSeconds: 10, CPU: []float64{90}, MemTotal: 100, MemUsed: 50},
+ }
+ for _, node := range nodes {
+ if err := dbSet(db, node); err != nil {
+ t.Fatalf("set %s: %v", node.Name, err)
+ }
+ }
+
+ req := httptest.NewRequest("GET", "/?theme=dark&palette=monochrome", nil)
+ rr := httptest.NewRecorder()
+ makeHandler(db, "fresh").ServeHTTP(rr, req)
+
+ body := rr.Body.String()
+ for _, want := range []string{
+ `online 1`,
+ `stale 1`,
+ `offline 1`,
+ `data-state="stale"`,
+ `id="sort-select"`,
+ `id="hide-offline"`,
+ `value="cpuAvg"`,
+ `value="cpuMax"`,
+ `value="memPct"`,
+ `value="memUsed"`,
+ `value="memTotal"`,
+ `value="load1"`,
+ `value="load5"`,
+ `value="load15"`,
+ `data-cpu-avg="20.000"`,
+ `data-cpu-max="20.000"`,
+ `data-mem-pct="30.000"`,
+ `data-mem-used="30"`,
+ `data-mem-total="100"`,
+ `data-load1="0.000"`,
+ `data-load5="0.000"`,
+ `data-load15="0.000"`,
+ } {
+ if !strings.Contains(body, want) {
+ t.Fatalf("dashboard missing %q:\n%s", want, body)
+ }
+ }
+}
+
+func TestCollectStatsUsesConfiguredTTL(t *testing.T) {
+ stats, err := collectStats("node-a", "http://node-a:8080", "v1", 42*time.Second)
+ if err != nil {
+ t.Fatalf("collectStats: %v", err)
+ }
+ if stats.TTLSeconds != 42 {
+ t.Fatalf("ttl seconds = %d, want 42", stats.TTLSeconds)
+ }
+}
diff --git "a/tasks/0_planning/\360\237\222\241-\360\237\224\224-\360\237\233\251\357\270\217-selectable_metrics_and_sensors.md" "b/tasks/0_planning/\360\237\222\241-\360\237\224\224-\360\237\233\251\357\270\217-selectable_metrics_and_sensors.md"
new file mode 100644
index 0000000..1f4e59a
--- /dev/null
+++ "b/tasks/0_planning/\360\237\222\241-\360\237\224\224-\360\237\233\251\357\270\217-selectable_metrics_and_sensors.md"
@@ -0,0 +1,41 @@
+---
+id: task-20260518-selectable-metrics-sensors
+title: Plan selectable metrics and host sensors
+status: 0_planning
+type: feature
+priority: normal
+effort: trip
+creator: codex
+owner: ""
+created: 2026-05-18
+---
+
+## Problem
+
+psstd currently renders a fixed set of metrics for every node. Real hosts may
+have useful extra sensors, such as temperature, fan speed, power, battery, UPS,
+HID, or I2C-attached devices, and not every node will support the same readings.
+
+Operators should eventually be able to choose which metrics appear in the
+dashboard and terminal view without making unsupported node sensors look broken.
+
+## Planning questions
+
+- Which metrics are core and always shown?
+- Which metrics are optional per-node capabilities?
+- Should metric selection be a browser preference, node-local config, cluster-wide config, or a mix?
+- How should the UI represent a selected metric that only some nodes can report?
+- Which operating-system sensor APIs are reliable enough on Linux, macOS, Windows, and containers?
+- Which Go libraries or native command integrations should be considered for sensor discovery and reading?
+- How often should slow or expensive sensor reads run compared with CPU/memory/load heartbeats?
+- How should sensor values be normalized, named, unit-tagged, and versioned in the gossip payload?
+
+## Done when
+
+- Existing CPU, memory, load, age, and health metrics have an explicit display-selection model
+- Optional metrics can be hidden or shown without adding server API surface unless the design justifies it
+- Sensor capability discovery is defined separately from sensor reading
+- Unsupported metrics render as unavailable, not stale/offline
+- Sensor payloads include stable names, units, values, and collection timestamps
+- The first implementation scope is small enough for one MR
+- Follow-up tickets exist for OS-specific sensor support after research
diff --git "a/tasks/0_planning/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\217\225\357\270\217-cli_mode_flags.md" "b/tasks/0_planning/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\217\225\357\270\217-cli_mode_flags.md"
new file mode 100644
index 0000000..c3480d5
--- /dev/null
+++ "b/tasks/0_planning/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\217\225\357\270\217-cli_mode_flags.md"
@@ -0,0 +1,30 @@
+---
+id: task-20260518-cli-mode-flags
+title: Design explicit CLI mode flags
+status: 0_planning
+type: maintenance
+priority: normal
+effort: night
+creator: codex
+owner: ""
+created: 2026-05-18
+---
+
+## Problem
+
+The requested `-t`, `-v`, `-tv`, and `-r` modes change psstd startup semantics:
+today the primary process owns the local database, participates in gossip, serves
+HTTP by default, and only falls back to terminal mirror mode when another local
+instance already owns the store or gossip port.
+
+Making terminal mirror mode the default needs a clearer mode model so service
+deployments, web serving, read-only cluster observation, and log streaming remain
+predictable.
+
+## Done when
+
+- CLI modes are explicitly named and documented
+- Default behavior for interactive shells and services is decided
+- `-t`, `-v`, `-tv`, and `-r` have non-overlapping behavior
+- Read-only observer mode cannot accidentally publish a duplicate node
+- Tests cover flag parsing and mode selection
diff --git "a/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_health_ttl_configuration.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_health_ttl_configuration.md"
new file mode 100644
index 0000000..c0de1df
--- /dev/null
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_health_ttl_configuration.md"
@@ -0,0 +1,34 @@
+---
+id: task-20260518-node-health-ttl-config
+title: Make node health TTL configurable
+status: 3_done
+type: feature
+priority: normal
+effort: cake
+creator: codex
+owner: ""
+created: 2026-05-18
+forked_from: task-20260517-health
+---
+
+## Problem
+
+Fresh, stale, and offline rendering now uses a TTL published by the origin node,
+but the origin currently only publishes psstd's built-in default. Operators
+still cannot tune how long a node wants its last heartbeat to be considered
+fresh or stale.
+
+## Done when
+
+- A node-local configuration value controls the TTL published in that node's heartbeat
+- Empty configuration preserves the current default
+- Invalid TTL values fail clearly or fall back with an explicit warning
+- README documents when and how to tune the TTL
+- Tests cover default, configured, and invalid TTL behavior
+
+## Result
+
+- Added `PSSTD_NODE_TTL` duration parsing with a `15s` default and clear invalid-value failures
+- Published the configured TTL in each node heartbeat
+- Documented TTL tuning in README
+- Added unit coverage for default, configured, invalid, and heartbeat-published TTL behavior
diff --git "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md"
similarity index 71%
rename from "tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md"
rename to "tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md"
index 0004d39..0984e0d 100644
--- "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md"
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\215\260-node_name_override.md"
@@ -1,7 +1,7 @@
---
id: task-20260517-node-name-override
title: Add optional node name override
-status: 1_todo
+status: 3_done
type: feature
priority: normal
effort: cake
@@ -21,3 +21,9 @@ psstd uses the OS hostname as the node identity. That is good for zero-config in
- Node name is validated enough to avoid empty names and confusing whitespace
- README documents when to use the override
- Tests cover default hostname and override behavior where practical
+
+## Result
+
+- Implemented `PSSTD_NODE_NAME` for memberlist identity, mDNS registration, stats heartbeat, and terminal mirror fallback naming
+- Added validation and README coverage
+- Added focused unit tests for default, override, and whitespace rejection
diff --git "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md"
similarity index 75%
rename from "tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md"
rename to "tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md"
index 2c90022..3fa6385 100644
--- "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md"
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-cluster_summary_band.md"
@@ -1,7 +1,7 @@
---
id: task-20260517-cluster-summary-band
title: Add a compact cluster summary band
-status: 1_todo
+status: 3_done
type: feature
priority: normal
effort: walk
@@ -20,3 +20,9 @@ The dashboard is strongest as an at-a-glance cluster htop. A small summary can a
- Summary identifies the hottest online node by CPU or load
- Terminal mirror includes the same counts in its header
- The summary uses existing node snapshots only and adds no API surface
+
+## Result
+
+- Added a compact web summary with online, stale, offline, and hottest-node details
+- Added the same summary to the terminal mirror header
+- Summary is calculated from existing node snapshots
diff --git "a/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_metric_sort_completeness.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_metric_sort_completeness.md"
new file mode 100644
index 0000000..fe15244
--- /dev/null
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_metric_sort_completeness.md"
@@ -0,0 +1,33 @@
+---
+id: task-20260518-dashboard-metric-sort-completeness
+title: Complete dashboard metric sorting
+status: 3_done
+type: feature
+priority: normal
+effort: walk
+creator: codex
+owner: ""
+created: 2026-05-18
+forked_from: task-20260517-client-sort
+---
+
+## Problem
+
+The dashboard now sorts by name, average CPU, memory percentage, load1, and age.
+The original ticket said the dashboard should sort by any populated metrics, but
+the implementation does not yet cover the full node snapshot shape or explicitly
+define which metrics are intentionally sortable.
+
+## Done when
+
+- The sortable metric set is explicitly defined in code and UI
+- Populated node metrics are either sortable or intentionally excluded in the ticket result
+- Load sorting covers the load values users expect, not only load1 unless that is the intended scope
+- CPU sorting handles the desired aggregate, such as average and/or max core
+- Tests cover rendered sort data for each supported metric
+
+## Result
+
+- Sort controls now cover name, CPU average, CPU max, memory percent, memory used, memory total, load 1m, load 5m, load 15m, and age
+- Dashboard cards render sortable data attributes for each supported metric
+- Tests cover the rendered sort options and metric data
diff --git "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md"
similarity index 58%
rename from "tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md"
rename to "tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md"
index 9bcdd0a..dfd0bb7 100644
--- "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md"
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-dashboard_sort_filter.md"
@@ -1,7 +1,7 @@
---
id: task-20260517-client-sort
title: Add dashboard sort and filter controls
-status: 0_planning
+status: 3_done
type: feature
priority: normal
effort: walk
@@ -21,4 +21,12 @@ Large clusters benefit from sorting by CPU% or memory to spot hot nodes quickly.
- Optional filter can hide offline or stale nodes
- Preferences are stored in the browser, not the server
+## Result
+- Added client-side sort controls for name, CPU, memory, load, and age
+- Added browser-persisted stale/offline filters
+- Kept the implementation in the dashboard template without adding server API surface
+
+## Forked follow-up
+
+- `task-20260518-dashboard-metric-sort-completeness`: finish or explicitly define the full set of populated metrics that should be sortable
diff --git "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md" "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md"
similarity index 60%
rename from "tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md"
rename to "tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md"
index 72762cd..3012f80 100644
--- "a/tasks/1_todo/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md"
+++ "b/tasks/3_done/\360\237\222\241-\360\237\224\224-\360\237\233\235-stale_node_indication.md"
@@ -1,7 +1,7 @@
---
id: task-20260517-health
title: Distinguish stale nodes from offline nodes
-status: 1_todo
+status: 3_done
type: feature
priority: normal
effort: walk
@@ -20,3 +20,13 @@ Offline nodes and recently stale nodes can look too similar. Operators should qu
- The staleness threshold is controlled by the origin node (each node controls the 'time' it'd like to be known for)
- Terminal and web views use the same state calculation
- Existing offline purge/version behavior still works
+
+## Result
+
+- Added node heartbeat TTL metadata and shared fresh/stale/offline state calculation
+- Terminal and web rendering now use the shared health state
+- Offline purge/version behavior still flows through the same `nodeRecordOffline` helper
+
+## Forked follow-up
+
+- `task-20260518-node-health-ttl-config`: expose and validate the origin node's TTL as configuration instead of only publishing the built-in default
diff --git "a/tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\215\260-startup_join_outcome_summary.md" "b/tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\215\260-startup_join_outcome_summary.md"
new file mode 100644
index 0000000..b11929a
--- /dev/null
+++ "b/tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\215\260-startup_join_outcome_summary.md"
@@ -0,0 +1,33 @@
+---
+id: task-20260518-startup-join-outcome-summary
+title: Include join outcome in startup summary
+status: 3_done
+type: maintenance
+priority: normal
+effort: cake
+creator: codex
+owner: ""
+created: 2026-05-18
+forked_from: task-20260517-cli
+---
+
+## Problem
+
+Startup logging now prints a concise configuration summary and quiets repeated
+mDNS discovery lines, but the actual join outcome is still reported separately.
+The original operator-facing problem asks for the startup summary to make it
+obvious whether psstd joined peers or is running solo.
+
+## Done when
+
+- The startup summary includes joined peer count or solo status
+- Join warnings still include enough detail for troubleshooting
+- Seed count, mDNS discovery count, and join result are not repeated noisily
+- Tests cover startup summary formatting for joined, solo, and warning cases where practical
+
+## Result
+
+- Startup summary now logs solo, joined, or warning join status in the same operator-facing line as configuration
+- Join warning summaries include joined count and error text
+- Removed the separate joined/solo startup log line to avoid repetition
+- Added tests for joined, solo, and warning startup summary formatting
diff --git "a/tasks/1_todo/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md" "b/tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md"
similarity index 54%
rename from "tasks/1_todo/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md"
rename to "tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md"
index 5fc9c12..1c71b70 100644
--- "a/tasks/1_todo/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md"
+++ "b/tasks/3_done/\360\237\233\240\357\270\217-\360\237\224\224-\360\237\233\235-startup_configuration.md"
@@ -1,7 +1,7 @@
---
id: task-20260517-cli
title: Print a clear startup configuration
-status: 1_todo
+status: 3_done
type: maintenance
priority: normal
effort: walk
@@ -19,12 +19,11 @@ psstd is easiest to like when it feels obvious what it is doing: which database
- Startup logs show DB path, HTTP listen address, advertised URL, gossip listen address, web enabled state, and version
- Seed and mDNS discovery results are summarized without noisy repetition
-## Problem 2
-Just a general cleanup of cli flags and provide some of the basics like help text.
+## Result
-## Done when
-- Terminal mirror mode should be default mode (-t)
-- If a -v flag is provided, then it prints the log instead
-- If a -tv flag is provided, then we print both the terminal rendering and a log stream (similar to how screen does window splits)
-- A -r flag does "read only" and '-t' is still default, and -v and -tv still do the same, just as a read-only 'observer' of the cluster
-- Existing log lines remain useful for troubleshooting
+- Added a concise startup summary including version, node, DB path, web state, HTTP listen address, advertised URL, gossip address, explicit seed count, and mDNS discovery count
+- Split the CLI mode flag cleanup into `task-20260518-cli-mode-flags` for planning
+
+## Forked follow-up
+
+- `task-20260518-startup-join-outcome-summary`: include the actual join outcome in the operator-facing startup summary
diff --git a/templates/dashboard.html b/templates/dashboard.html
index 335f6a7..549f3d9 100644
--- a/templates/dashboard.html
+++ b/templates/dashboard.html
@@ -15,6 +15,8 @@
--text: #f7f7f7;
--muted: #9ca3af;
--offline: #f97316;
+ --stale: #facc15;
+ --fresh: #22c55e;
--accent: #60a5fa;
--toolbar: #111111;
--toolbar-border: #2a2a2a;
@@ -28,6 +30,8 @@
--text: #0f172a;
--muted: #475569;
--offline: #b91c1c;
+ --stale: #b45309;
+ --fresh: #15803d;
--accent: #2563eb;
--toolbar: #eef2ff;
--toolbar-border: #cbd5e1;
@@ -40,6 +44,8 @@
--text: #93a1a1;
--muted: #657b83;
--offline: #dc322f;
+ --stale: #b58900;
+ --fresh: #859900;
--accent: #b58900;
--toolbar: #00212b;
--toolbar-border: #586e75;
@@ -52,6 +58,8 @@
--text: #d0d0d0;
--muted: #94a3b8;
--offline: #ef4444;
+ --stale: #facc15;
+ --fresh: #8ae234;
--accent: #ef8b00;
--toolbar: #20242b;
--toolbar-border: #4b5263;
@@ -64,6 +72,8 @@
--text: #eceff4;
--muted: #d8dee9;
--offline: #bf616a;
+ --stale: #ebcb8b;
+ --fresh: #a3be8c;
--accent: #88c0d0;
--toolbar: #242933;
--toolbar-border: #4c566a;
@@ -76,6 +86,8 @@
--text: #ebdbb2;
--muted: #a89984;
--offline: #fb4934;
+ --stale: #fabd2f;
+ --fresh: #b8bb26;
--accent: #fabd2f;
--toolbar: #1d2021;
--toolbar-border: #665c54;
@@ -88,6 +100,8 @@
--text: #f8f8f2;
--muted: #bd93f9;
--offline: #ff5555;
+ --stale: #f1fa8c;
+ --fresh: #50fa7b;
--accent: #50fa7b;
--toolbar: #21222c;
--toolbar-border: #6272a4;
@@ -100,6 +114,8 @@
--text: #cdd6f4;
--muted: #a6adc8;
--offline: #f38ba8;
+ --stale: #f9e2af;
+ --fresh: #a6e3a1;
--accent: #89b4fa;
--toolbar: #181825;
--toolbar-border: #585b70;
@@ -112,6 +128,8 @@
--text: #d3c6aa;
--muted: #9da9a0;
--offline: #e67e80;
+ --stale: #dbbc7f;
+ --fresh: #a7c080;
--accent: #a7c080;
--toolbar: #232a2e;
--toolbar-border: #56635f;
@@ -124,6 +142,8 @@
--text: #e5e7eb;
--muted: #93c5fd;
--offline: #fb7185;
+ --stale: #fde047;
+ --fresh: #34d399;
--accent: #22d3ee;
--toolbar: #030712;
--toolbar-border: #374151;
@@ -136,6 +156,8 @@
--text: #d4ffd9;
--muted: #6ee787;
--offline: #ff6b6b;
+ --stale: #faff00;
+ --fresh: #00ff66;
--accent: #00ff66;
--toolbar: #000000;
--toolbar-border: #1f6b3a;
@@ -148,6 +170,8 @@
--text: #ffe9bd;
--muted: #f4b860;
--offline: #ff5c3a;
+ --stale: #ffcf33;
+ --fresh: #83ff66;
--accent: #ffb000;
--toolbar: #0b0703;
--toolbar-border: #7c4a12;
@@ -161,6 +185,8 @@
--text: #22201b;
--muted: #6f6758;
--offline: #b42318;
+ --stale: #a16207;
+ --fresh: #166534;
--accent: #006d77;
--toolbar: #eee7d5;
--toolbar-border: #c8bfa8;
@@ -173,6 +199,8 @@
--text: #ffffff;
--muted: #cfcfcf;
--offline: #ff3b30;
+ --stale: #ffff00;
+ --fresh: #00ff66;
--accent: #00e5ff;
--toolbar: #000000;
--toolbar-border: #ffffff;
@@ -205,6 +233,24 @@
align-items: center;
flex-wrap: wrap;
}
+.summary {
+ color: var(--muted);
+ display: flex;
+ gap: 10px;
+ flex-wrap: wrap;
+ align-items: center;
+}
+.summary strong { color: var(--text); }
+.summary .fresh { color: var(--fresh); }
+.summary .stale { color: var(--stale); }
+.summary .offline { color: var(--offline); }
+.nodes {
+ width: 100%;
+ display: flex;
+ flex-wrap: wrap;
+ align-content: flex-start;
+ gap: 6px;
+}
.toolbar label {
font-size: 0.95rem;
color: var(--muted);
@@ -247,6 +293,7 @@
.cell-header a:hover {
text-decoration: underline;
}
+.cell.stale { border-color: var(--stale); }
.cell.offline { opacity: 0.55; border-color: var(--offline); }
pre { line-height: 1.4; white-space: pre; color: var(--text); }
@@ -258,6 +305,9 @@
const theme = params.get('theme') || storedTheme;
const palette = params.get('palette') || storedPalette;
const refreshMs = {{.RefreshMs}};
+ const storedSort = localStorage.getItem('psstd-sort') || 'name';
+ const storedHideStale = localStorage.getItem('psstd-hide-stale') === 'true';
+ const storedHideOffline = localStorage.getItem('psstd-hide-offline') === 'true';
document.documentElement.dataset.theme = theme;
document.documentElement.dataset.palette = palette;
@@ -290,14 +340,44 @@
document.documentElement.dataset.palette = paletteValue;
localStorage.setItem('psstd-palette', paletteValue);
updateQuery('palette', paletteValue);
+ },
+ applyDashboardControls() {
+ const sortSelect = document.getElementById('sort-select');
+ const hideStale = document.getElementById('hide-stale');
+ const hideOffline = document.getElementById('hide-offline');
+ const nodes = document.getElementById('nodes');
+ if (!sortSelect || !hideStale || !hideOffline || !nodes) return;
+
+ localStorage.setItem('psstd-sort', sortSelect.value);
+ localStorage.setItem('psstd-hide-stale', hideStale.checked ? 'true' : 'false');
+ localStorage.setItem('psstd-hide-offline', hideOffline.checked ? 'true' : 'false');
+
+ const cards = Array.from(nodes.querySelectorAll('.cell'));
+ const sortKey = sortSelect.value;
+ cards.sort((a, b) => {
+ if (sortKey === 'name') return a.dataset.name.localeCompare(b.dataset.name);
+ return Number(b.dataset[sortKey] || 0) - Number(a.dataset[sortKey] || 0);
+ });
+ for (const card of cards) {
+ const state = card.dataset.state;
+ card.hidden = (hideStale.checked && state === 'stale') || (hideOffline.checked && state === 'offline');
+ nodes.appendChild(card);
+ }
}
};
window.addEventListener('DOMContentLoaded', () => {
const themeSelect = document.getElementById('theme-select');
const paletteSelect = document.getElementById('palette-select');
+ const sortSelect = document.getElementById('sort-select');
+ const hideStale = document.getElementById('hide-stale');
+ const hideOffline = document.getElementById('hide-offline');
if (themeSelect) themeSelect.value = theme;
if (paletteSelect) paletteSelect.value = palette;
+ if (sortSelect) sortSelect.value = storedSort;
+ if (hideStale) hideStale.checked = storedHideStale;
+ if (hideOffline) hideOffline.checked = storedHideOffline;
+ window.psstd.applyDashboardControls();
});
setTimeout(() => location.replace({{.RefreshURL}}), refreshMs);
@@ -331,14 +411,38 @@
- Auto-refresh {{.RefreshLabel}} · {{.BestHint}}
+
+
+
+
+
+
+ Cluster
+ online {{.Summary.Fresh}}
+ stale {{.Summary.Stale}}
+ offline {{.Summary.Offline}}
+ {{if .Summary.HasHot}}hottest {{.Summary.Hottest}} {{printf "%.0f" .Summary.HotCPU}}% CPU {{printf "%.2f" .Summary.HotLoad}} load{{else}}hottest none{{end}}
+ Auto-refresh {{.RefreshLabel}} · {{.BestHint}}
+