generated from rmcguire/go-server-with-otel
implement econet exporter
This commit is contained in:
@@ -0,0 +1,41 @@
|
||||
package econetmetrics
|
||||
|
||||
import "go.opentelemetry.io/otel/metric"
|
||||
|
||||
// gaugeBuilder creates observable gauges, accumulating them for a single
|
||||
// RegisterCallback and short-circuiting on the first creation error.
|
||||
//
|
||||
// Units are intentionally baked into the metric names (Prometheus idiom)
|
||||
// rather than set via metric.WithUnit — the OTEL→Prometheus exporter would
|
||||
// otherwise append a second unit suffix (e.g. econet_setpoint_fahrenheit_F).
|
||||
type gaugeBuilder struct {
|
||||
m metric.Meter
|
||||
obs []metric.Observable
|
||||
err error
|
||||
}
|
||||
|
||||
func (b *gaugeBuilder) i64(name, desc string) metric.Int64ObservableGauge {
|
||||
if b.err != nil {
|
||||
return nil
|
||||
}
|
||||
g, err := b.m.Int64ObservableGauge(name, metric.WithDescription(desc))
|
||||
if err != nil {
|
||||
b.err = err
|
||||
return nil
|
||||
}
|
||||
b.obs = append(b.obs, g)
|
||||
return g
|
||||
}
|
||||
|
||||
func (b *gaugeBuilder) f64(name, desc string) metric.Float64ObservableGauge {
|
||||
if b.err != nil {
|
||||
return nil
|
||||
}
|
||||
g, err := b.m.Float64ObservableGauge(name, metric.WithDescription(desc))
|
||||
if err != nil {
|
||||
b.err = err
|
||||
return nil
|
||||
}
|
||||
b.obs = append(b.obs, g)
|
||||
return g
|
||||
}
|
||||
@@ -0,0 +1,132 @@
|
||||
package econetmetrics
|
||||
|
||||
import (
|
||||
"context"
|
||||
|
||||
rheemcloud "github.com/kevinburke/rheemcloud-go"
|
||||
"go.opentelemetry.io/otel/attribute"
|
||||
"go.opentelemetry.io/otel/metric"
|
||||
)
|
||||
|
||||
// registerLive wires the per-device live-state gauges to a single callback
|
||||
// that snapshots client.Devices() once per scrape (cheap, in-memory).
|
||||
func (c *Collector) registerLive() error {
|
||||
b := &gaugeBuilder{m: c.meter}
|
||||
g := &liveGauges{
|
||||
setpoint: b.i64("econet_setpoint_fahrenheit", "Target water temperature (°F)"),
|
||||
setpointMin: b.i64("econet_setpoint_min_fahrenheit", "Minimum allowed setpoint (°F)"),
|
||||
setpointMax: b.i64("econet_setpoint_max_fahrenheit", "Maximum allowed setpoint (°F)"),
|
||||
connected: b.i64("econet_connected", "1 if the device is connected"),
|
||||
running: b.i64("econet_running", "1 if the device is actively heating"),
|
||||
enabled: b.i64("econet_enabled", "1 if the device is enabled"),
|
||||
away: b.i64("econet_away", "1 if away mode is on"),
|
||||
hotWater: b.i64("econet_hot_water_availability_percent", "Hot water availability (%)"),
|
||||
alerts: b.i64("econet_alert_count", "Number of active alerts"),
|
||||
wifi: b.i64("econet_wifi_signal_db", "WiFi signal strength (dB)"),
|
||||
lastUpdated: b.i64("econet_last_updated_seconds", "Unix time of last MQTT state update"),
|
||||
info: b.i64("econet_device_info", "Device metadata (value is always 1)"),
|
||||
}
|
||||
if b.err != nil {
|
||||
return b.err
|
||||
}
|
||||
_, err := c.meter.RegisterCallback(func(_ context.Context, o metric.Observer) error {
|
||||
for _, d := range c.client.Devices() {
|
||||
g.observe(o, d)
|
||||
}
|
||||
return nil
|
||||
}, b.obs...)
|
||||
return err
|
||||
}
|
||||
|
||||
// registerUsage wires the polled energy/water gauges, reading the cache the
|
||||
// poller refreshes (the underlying REST calls are too slow for a callback).
|
||||
func (c *Collector) registerUsage() error {
|
||||
b := &gaugeBuilder{m: c.meter}
|
||||
g := &usageGauges{
|
||||
energy: b.f64("econet_energy_usage_kwh", "Energy used today (kWh)"),
|
||||
cost: b.f64("econet_energy_cost_dollars", "Energy cost today in US dollars (kWh * costPerKWH)"),
|
||||
water: b.f64("econet_water_usage_gallons", "Water used today (gallons)"),
|
||||
}
|
||||
if b.err != nil {
|
||||
return b.err
|
||||
}
|
||||
_, err := c.meter.RegisterCallback(func(_ context.Context, o metric.Observer) error {
|
||||
c.mu.RLock()
|
||||
defer c.mu.RUnlock()
|
||||
for serial, u := range c.usage {
|
||||
g.observe(o, serial, u, c.cfg.CostPerKWH)
|
||||
}
|
||||
return nil
|
||||
}, b.obs...)
|
||||
return err
|
||||
}
|
||||
|
||||
type liveGauges struct {
|
||||
setpoint, setpointMin, setpointMax metric.Int64ObservableGauge
|
||||
connected, running, enabled, away metric.Int64ObservableGauge
|
||||
hotWater, alerts, wifi, lastUpdated metric.Int64ObservableGauge
|
||||
info metric.Int64ObservableGauge
|
||||
}
|
||||
|
||||
func (g *liveGauges) observe(o metric.Observer, d *rheemcloud.Device) {
|
||||
set := metric.WithAttributes(deviceAttrs(d)...)
|
||||
min, max := d.SetpointLimits()
|
||||
o.ObserveInt64(g.setpoint, int64(d.Setpoint()), set)
|
||||
o.ObserveInt64(g.setpointMin, int64(min), set)
|
||||
o.ObserveInt64(g.setpointMax, int64(max), set)
|
||||
o.ObserveInt64(g.connected, b2i(d.Connected()), set)
|
||||
o.ObserveInt64(g.running, b2i(d.Running()), set)
|
||||
o.ObserveInt64(g.enabled, b2i(d.Enabled()), set)
|
||||
o.ObserveInt64(g.away, b2i(d.Away()), set)
|
||||
o.ObserveInt64(g.hotWater, int64(d.HotWaterAvailability()), set)
|
||||
o.ObserveInt64(g.alerts, int64(d.AlertCount()), set)
|
||||
o.ObserveInt64(g.wifi, int64(d.WiFiSignal()), set)
|
||||
// LastUpdated is zero until the first MQTT update lands; skip it then
|
||||
// rather than emit a nonsensical negative epoch.
|
||||
if !d.LastUpdated.IsZero() {
|
||||
o.ObserveInt64(g.lastUpdated, d.LastUpdated.Unix(), set)
|
||||
}
|
||||
o.ObserveInt64(g.info, 1, metric.WithAttributes(infoAttrs(d)...))
|
||||
}
|
||||
|
||||
type usageGauges struct {
|
||||
energy, cost, water metric.Float64ObservableGauge
|
||||
}
|
||||
|
||||
func (g *usageGauges) observe(o metric.Observer, serial string, u usage, costPerKWH float64) {
|
||||
set := metric.WithAttributes(attribute.String("econet.serial", serial))
|
||||
o.ObserveFloat64(g.energy, u.energyKWH, metric.WithAttributes(
|
||||
attribute.String("econet.serial", serial),
|
||||
attribute.String("econet.energy_type", u.energyType),
|
||||
))
|
||||
o.ObserveFloat64(g.water, u.waterGallons, set)
|
||||
if u.energyType == "KWH" {
|
||||
o.ObserveFloat64(g.cost, u.energyKWH*costPerKWH, set)
|
||||
}
|
||||
}
|
||||
|
||||
// deviceAttrs are low-cardinality identity attributes safe for every series.
|
||||
// They follow OTEL semantic-convention style (namespaced, dotted keys) and
|
||||
// deliberately exclude anything that changes over time (e.g. timestamps).
|
||||
func deviceAttrs(d *rheemcloud.Device) []attribute.KeyValue {
|
||||
return []attribute.KeyValue{
|
||||
attribute.String("econet.serial", d.SerialNumber()),
|
||||
attribute.String("econet.device_id", d.DeviceID()),
|
||||
attribute.String("econet.friendly_name", d.FriendlyName()),
|
||||
}
|
||||
}
|
||||
|
||||
func infoAttrs(d *rheemcloud.Device) []attribute.KeyValue {
|
||||
return append(deviceAttrs(d),
|
||||
attribute.String("econet.type", d.Type().String()),
|
||||
attribute.String("econet.generic_type", d.GenericType()),
|
||||
attribute.String("econet.mode", d.Mode().String()),
|
||||
)
|
||||
}
|
||||
|
||||
func b2i(b bool) int64 {
|
||||
if b {
|
||||
return 1
|
||||
}
|
||||
return 0
|
||||
}
|
||||
@@ -0,0 +1,12 @@
|
||||
package econetmetrics
|
||||
|
||||
import "testing"
|
||||
|
||||
func TestB2I(t *testing.T) {
|
||||
if got := b2i(true); got != 1 {
|
||||
t.Errorf("b2i(true) = %d, want 1", got)
|
||||
}
|
||||
if got := b2i(false); got != 0 {
|
||||
t.Errorf("b2i(false) = %d, want 0", got)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,123 @@
|
||||
// Package econetmetrics registers OTEL observable gauges for Rheem EcoNet
|
||||
// device state and periodically polls energy/water usage history so the
|
||||
// values are available at Prometheus scrape time.
|
||||
package econetmetrics
|
||||
|
||||
import (
|
||||
"context"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
rheemcloud "github.com/kevinburke/rheemcloud-go"
|
||||
"github.com/rs/zerolog"
|
||||
"go.opentelemetry.io/otel/metric"
|
||||
|
||||
"gitea.libretechconsulting.com/rmcguire/go-app/pkg/otel"
|
||||
|
||||
"gitea.libretechconsulting.com/rmcguire/econet-exporter/pkg/config"
|
||||
)
|
||||
|
||||
// usage holds the most recent polled energy/water totals for a device.
|
||||
type usage struct {
|
||||
energyKWH float64
|
||||
energyType string
|
||||
waterGallons float64
|
||||
}
|
||||
|
||||
type Collector struct {
|
||||
ctx context.Context
|
||||
cfg *config.ServiceConfig
|
||||
log *zerolog.Logger
|
||||
client *rheemcloud.Client
|
||||
meter metric.Meter
|
||||
|
||||
mu sync.RWMutex
|
||||
usage map[string]usage // keyed by serial number
|
||||
|
||||
stop chan struct{}
|
||||
}
|
||||
|
||||
func NewCollector(ctx context.Context, cfg *config.ServiceConfig, client *rheemcloud.Client) *Collector {
|
||||
return &Collector{
|
||||
ctx: ctx,
|
||||
cfg: cfg,
|
||||
log: zerolog.Ctx(ctx),
|
||||
client: client,
|
||||
meter: otel.GetMeter(ctx, "econet"),
|
||||
usage: make(map[string]usage),
|
||||
stop: make(chan struct{}),
|
||||
}
|
||||
}
|
||||
|
||||
// Start registers the gauge callbacks and launches the usage poller and
|
||||
// event drain goroutines. The rheemcloud client keeps live state fresh over
|
||||
// MQTT; draining its event channel keeps the cap-64 buffer from filling.
|
||||
func (c *Collector) Start() error {
|
||||
if err := c.registerLive(); err != nil {
|
||||
return err
|
||||
}
|
||||
if err := c.registerUsage(); err != nil {
|
||||
return err
|
||||
}
|
||||
go c.drainEvents()
|
||||
go c.pollUsage()
|
||||
return nil
|
||||
}
|
||||
|
||||
func (c *Collector) Stop() { close(c.stop) }
|
||||
|
||||
func (c *Collector) drainEvents() {
|
||||
events := c.client.Subscribe()
|
||||
for {
|
||||
select {
|
||||
case <-c.stop:
|
||||
return
|
||||
case <-c.ctx.Done():
|
||||
return
|
||||
case ev, ok := <-events:
|
||||
if !ok {
|
||||
return
|
||||
}
|
||||
c.log.Debug().Str("event", ev.Kind.String()).Msg("econet event")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func (c *Collector) pollUsage() {
|
||||
c.refreshUsage()
|
||||
t := time.NewTicker(c.cfg.GetUsageInterval())
|
||||
defer t.Stop()
|
||||
for {
|
||||
select {
|
||||
case <-c.stop:
|
||||
return
|
||||
case <-c.ctx.Done():
|
||||
return
|
||||
case <-t.C:
|
||||
c.refreshUsage()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// refreshUsage polls each device's energy/water usage for the current day
|
||||
// and caches the totals for the usage gauge callback to read.
|
||||
func (c *Collector) refreshUsage() {
|
||||
now := time.Now()
|
||||
start := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())
|
||||
for serial, d := range c.client.Devices() {
|
||||
u := usage{}
|
||||
if e, err := c.client.EnergyUsage(c.ctx, d, start, now); err == nil {
|
||||
u.energyKWH, u.energyType = e.Total, e.EnergyType
|
||||
} else {
|
||||
c.log.Debug().Err(err).Str("serial", serial).Msg("energy usage poll failed")
|
||||
}
|
||||
if w, err := c.client.WaterUsage(c.ctx, d, start, now); err == nil {
|
||||
u.waterGallons = w.Total
|
||||
} else {
|
||||
c.log.Debug().Err(err).Str("serial", serial).Msg("water usage poll failed")
|
||||
}
|
||||
c.mu.Lock()
|
||||
c.usage[serial] = u
|
||||
c.mu.Unlock()
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user