feat: add event-driven handlers

Signed-off-by: Matthias Riegler <matthias.riegler@ankorstore.com>
This commit is contained in:
Matthias Riegler
2023-07-19 18:16:23 +02:00
parent 95e2a8d60c
commit 974db555ff
3 changed files with 539 additions and 0 deletions

103
cmd/agent/main.go Normal file
View File

@@ -0,0 +1,103 @@
package main
import (
"context"
"fmt"
"net/http"
"os"
"os/signal"
"sync"
"syscall"
"time"
"github.com/prometheus/client_golang/prometheus/promhttp"
"github.com/xvzf/computeblade-agent/internal/agent"
"github.com/xvzf/computeblade-agent/pkg/ledengine"
"github.com/xvzf/computeblade-agent/pkg/log"
"go.uber.org/zap"
)
func main() {
var wg sync.WaitGroup
// setup logger
zapLogger := zap.Must(zap.NewDevelopment()).With(zap.String("app", "computeblade-agent"))
_ = zap.ReplaceGlobals(zapLogger.With(zap.String("scope", "global")))
baseCtx := log.IntoContext(context.Background(), zapLogger)
ctx, cancelCtx := context.WithCancelCause(baseCtx)
defer cancelCtx(context.Canceled)
agent, err := agent.NewComputeBladeAgent(agent.ComputeBladeAgentConfig{
IdleLedColor: ledengine.LedColorGreen(0.05),
IdentifyLedColor: ledengine.LedColorPurple(0.05),
CriticalLedColor: ledengine.LedColorRed(0.3),
StealthModeEnabled: false,
DefaultFanSpeed: 40,
CriticalTemperature: 60,
})
if err != nil {
log.FromContext(ctx).Error("Failed to create agent", zap.Error(err))
cancelCtx(err)
}
// setup stop signal handlers
sigs := make(chan os.Signal, 1)
signal.Notify(sigs, syscall.SIGINT, syscall.SIGTERM)
wg.Add(1)
go func() {
defer wg.Done()
// Wait for context cancel or signal
select {
case <-ctx.Done():
case sig := <-sigs:
// On signal, cancel context
cancelCtx(fmt.Errorf("signal %s received", sig))
}
}()
// Run agent
wg.Add(1)
go func() {
defer wg.Done()
err := agent.Run(ctx)
if err != nil && err != context.Canceled {
log.FromContext(ctx).Error("Failed to run agent", zap.Error(err))
cancelCtx(err)
}
}()
// setup prometheus endpoint
promHandler := http.NewServeMux()
promHandler.Handle("/metrics", promhttp.Handler())
server := &http.Server{Addr: ":9666", Handler: promHandler}
wg.Add(1)
go func() {
defer wg.Done()
err := server.ListenAndServe()
if err != nil && err != http.ErrServerClosed {
log.FromContext(ctx).Error("Failed to start prometheus server", zap.Error(err))
cancelCtx(err)
}
}()
wg.Add(1)
go func() {
defer wg.Done()
<-ctx.Done()
shutdownCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
err := server.Shutdown(shutdownCtx)
if err != nil {
log.FromContext(ctx).Error("Failed to shutdown prometheus server", zap.Error(err))
}
}()
// Wait for context cancel
wg.Wait()
if err := ctx.Err(); err != nil && err != context.Canceled {
log.FromContext(ctx).Fatal("Exiting", zap.Error(err))
} else {
log.FromContext(ctx).Info("Exiting")
}
}

334
internal/agent/agent.go Normal file
View File

@@ -0,0 +1,334 @@
package agent
import (
"context"
"errors"
"sync"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
"github.com/xvzf/computeblade-agent/pkg/hal"
"github.com/xvzf/computeblade-agent/pkg/ledengine"
"github.com/xvzf/computeblade-agent/pkg/log"
"go.uber.org/zap"
)
var (
// eventCounter is a prometheus counter that counts the number of events handled by the agent
eventCounter = promauto.NewCounterVec(prometheus.CounterOpts{
Namespace: "computeblade_agent",
Name: "events_count",
Help: "ComputeBlade Agent internal event handler statistics (handled events)",
}, []string{"type"})
// droppedEventCounter is a prometheus counter that counts the number of events dropped by the agent
droppedEventCounter = promauto.NewCounterVec(prometheus.CounterOpts{
Namespace: "computeblade_agent",
Name: "events_dropped_count",
Help: "ComputeBlade Agent internal event handler statistics (dropped events)",
}, []string{"type"})
)
type Event int
const (
NoopEvent = iota
IdentifyEvent
IdentifyConfirmEvent
CriticalEvent
CriticalResetEvent
EdgeButtonEvent
)
func (e Event) String() string {
switch e {
case NoopEvent:
return "noop"
case IdentifyEvent:
return "identify"
case IdentifyConfirmEvent:
return "identify_confirm"
case CriticalEvent:
return "critical"
case CriticalResetEvent:
return "critical_reset"
case EdgeButtonEvent:
return "edge_button"
default:
return "unknown"
}
}
type ComputeBladeAgentConfig struct {
// IdleLedColor is the color of the edge LED when the blade is idle mode
IdleLedColor hal.LedColor
// IdentifyLedColor is the color of the edge LED when the blade is in identify mode
IdentifyLedColor hal.LedColor
// CriticalLedColor is the color of the top(!) LED when the blade is in critical mode.
// In the circumstance when >1 blades are in critical mode, the identidy function can be used to find the right blade
CriticalLedColor hal.LedColor
// StealthModeEnabled indicates whether stealth mode is enabled
StealthModeEnabled bool
// DefaultFanSpeed is the default fan speed in percent. Usually 40% is sufficient
DefaultFanSpeed uint
// Critical temperature of the compute blade (used to trigger critical mode)
CriticalTemperature uint
}
type ComputeBladeAgent interface {
Run(ctx context.Context) error
}
type computeBladeAgentImpl struct {
opts ComputeBladeAgentConfig
blade hal.ComputeBladeHal
state *computebladeState
edgeLedEngine ledengine.LedEngine
topLedEngine ledengine.LedEngine
eventChan chan Event
}
func NewComputeBladeAgent(opts ComputeBladeAgentConfig) (ComputeBladeAgent, error) {
var err error
// blade, err := hal.NewCm4Hal(hal.ComputeBladeHalOpts{
blade, err := hal.NewCm4Hal(hal.ComputeBladeHalOpts{
FanUnit: hal.FanUnitStandard, // FIXME: support smart fan unit
})
if err != nil {
return nil, err
}
edgeLedEngine := ledengine.NewLedEngine(ledengine.LedEngineOpts{
LedIdx: hal.LedEdge,
Hal: blade,
})
if err != nil {
return nil, err
}
topLedEngine := ledengine.NewLedEngine(ledengine.LedEngineOpts{
LedIdx: hal.LedTop,
Hal: blade,
})
if err != nil {
return nil, err
}
return &computeBladeAgentImpl{
opts: opts,
blade: blade,
edgeLedEngine: edgeLedEngine,
topLedEngine: topLedEngine,
state: NewComputeBladeState(),
eventChan: make(chan Event, 10), // backlog of 10 events. They should process fast but we e.g. don't want to miss button presses
}, nil
}
func (a *computeBladeAgentImpl) Run(origCtx context.Context) error {
var wg sync.WaitGroup
ctx, cancelCtx := context.WithCancelCause(origCtx)
defer a.cleanup(ctx)
log.FromContext(ctx).Info("Starting ComputeBlade agent")
// Ingest noop event to initialise metrics
a.state.RegisterEvent(NoopEvent)
// Set defaults
if err := a.blade.SetFanSpeed(uint8(a.opts.DefaultFanSpeed)); err != nil {
return err
}
if err := a.blade.SetStealthMode(a.opts.StealthModeEnabled); err != nil {
return err
}
// Start edge button event handler
wg.Add(1)
go func() {
defer wg.Done()
log.FromContext(ctx).Info("Starting edge button event handler")
for {
err := a.blade.WaitForEdgeButtonPress(ctx)
if err != nil && err != context.Canceled {
log.FromContext(ctx).Error("Edge button event handler failed", zap.Error(err))
cancelCtx(err)
} else if err != nil {
return
}
select {
case a.eventChan <- Event(EdgeButtonEvent):
default:
log.FromContext(ctx).Warn("Edge button press event dropped due to backlog")
droppedEventCounter.WithLabelValues(Event(EdgeButtonEvent).String()).Inc()
}
}
}()
// Start top LED engine
wg.Add(1)
go func() {
defer wg.Done()
log.FromContext(ctx).Info("Starting top LED engine")
err := a.runTopLedEngine(ctx)
if err != nil && err != context.Canceled {
log.FromContext(ctx).Error("Top LED engine failed", zap.Error(err))
cancelCtx(err)
}
}()
// Start edge LED engine
wg.Add(1)
go func() {
defer wg.Done()
log.FromContext(ctx).Info("Starting edge LED engine")
err := a.runEdgeLedEngine(ctx)
if err != nil && err != context.Canceled {
log.FromContext(ctx).Error("Edge LED engine failed", zap.Error(err))
cancelCtx(err)
}
}()
// Start event handler
wg.Add(1)
go func() {
defer wg.Done()
log.FromContext(ctx).Info("Starting event handler")
for {
select {
case <-ctx.Done():
return
case event := <-a.eventChan:
err := a.handleEvent(ctx, event)
if err != nil && err != context.Canceled {
log.FromContext(ctx).Error("Event handler failed", zap.Error(err))
cancelCtx(err)
}
}
}
}()
wg.Wait()
return ctx.Err()
}
// cleanup restores sane defaults before exiting. Ignores canceled context!
func (a *computeBladeAgentImpl) cleanup(ctx context.Context) {
log.FromContext(ctx).Info("Exiting, restoring safe settings")
if err := a.blade.SetFanSpeed(100); err != nil {
log.FromContext(ctx).Error("Failed to set fan speed to 100%", zap.Error(err))
}
if err := a.blade.SetLed(hal.LedEdge, hal.LedColor{}); err != nil {
log.FromContext(ctx).Error("Failed to set edge LED to off", zap.Error(err))
}
if err := a.blade.SetLed(hal.LedTop, hal.LedColor{}); err != nil {
log.FromContext(ctx).Error("Failed to set edge LED to off", zap.Error(err))
}
}
func (a *computeBladeAgentImpl) handleEvent(ctx context.Context, event Event) error {
log.FromContext(ctx).Info("Handling event", zap.String("event", event.String()))
eventCounter.WithLabelValues(event.String()).Inc()
// register event in state
a.state.RegisterEvent(event)
// Dispatch incoming events to the right handler(s)
switch event {
case CriticalEvent:
// Handle critical event
return a.handleCriticalActive(ctx)
case CriticalResetEvent:
// Handle critical event
return a.handleCriticalReset(ctx)
case IdentifyEvent:
// Handle identify event
return a.handleIdentifyActive(ctx)
case IdentifyConfirmEvent:
// Handle identify event
return a.handleIdentifyClear(ctx)
case EdgeButtonEvent:
// Handle edge button press to toggle identify mode
event := Event(IdentifyEvent)
if a.state.IdentifyActive() {
event = Event(IdentifyConfirmEvent)
}
select {
case a.eventChan <- Event(event):
default:
log.FromContext(ctx).Warn("Edge button press event dropped due to backlog")
droppedEventCounter.WithLabelValues(event.String()).Inc()
}
}
return nil
}
func (a *computeBladeAgentImpl) handleIdentifyActive(ctx context.Context) error {
log.FromContext(ctx).Info("Identify active")
return a.edgeLedEngine.SetPattern(ledengine.NewBurstPattern(hal.LedColor{}, a.opts.IdentifyLedColor))
}
func (a *computeBladeAgentImpl) handleIdentifyClear(ctx context.Context) error {
log.FromContext(ctx).Info("Identify confirmed/cleared")
return a.edgeLedEngine.SetPattern(ledengine.NewStaticPattern(a.opts.IdleLedColor))
}
func (a *computeBladeAgentImpl) handleCriticalActive(ctx context.Context) error {
log.FromContext(ctx).Warn("Blade in critical state, setting fan speed to 100% and turning on LEDs")
// Set fan speed to 100%
setFanspeedError := a.blade.SetFanSpeed(100)
// Disable stealth mode (turn on LEDs)
setStealthModeError := a.blade.SetStealthMode(false)
// Set critical pattern for top LED
setPatternTopLedErr := a.topLedEngine.SetPattern(
ledengine.NewSlowBlinkPattern(hal.LedColor{}, a.opts.CriticalLedColor),
)
// Combine errors, but don't stop execution flow for now
return errors.Join(setFanspeedError, setStealthModeError, setPatternTopLedErr)
}
func (a *computeBladeAgentImpl) handleCriticalReset(ctx context.Context) error {
log.FromContext(ctx).Info("Critical state cleared, setting fan speed to default and restoring LEDs to default state")
// Set fan speed to 100%
if err := a.blade.SetFanSpeed(uint8(a.opts.DefaultFanSpeed)); err != nil {
return err
}
// Reset stealth mode
if err := a.blade.SetStealthMode(a.opts.StealthModeEnabled); err != nil {
return err
}
// Set top LED off
if err := a.topLedEngine.SetPattern(ledengine.NewStaticPattern(hal.LedColor{})); err != nil {
return err
}
return nil
}
func (a *computeBladeAgentImpl) Close() error {
return errors.Join(a.blade.Close())
}
// runTopLedEngine runs the top LED engine
func (a *computeBladeAgentImpl) runTopLedEngine(ctx context.Context) error {
// FIXME the top LED is only used to indicate emergency situations
a.edgeLedEngine.SetPattern(ledengine.NewStaticPattern(hal.LedColor{}))
return a.edgeLedEngine.Run(ctx)
}
// runEdgeLedEngine runs the edge LED engine
func (a *computeBladeAgentImpl) runEdgeLedEngine(ctx context.Context) error {
a.edgeLedEngine.SetPattern(ledengine.NewStaticPattern(a.opts.IdleLedColor))
return a.edgeLedEngine.Run(ctx)
}

102
internal/agent/state.go Normal file
View File

@@ -0,0 +1,102 @@
package agent
import (
"context"
"sync"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
)
var (
stateMetric = promauto.NewGaugeVec(prometheus.GaugeOpts{
Namespace: "computeblade_state",
Name: "state",
Help: "ComputeBlade state (label values are critical, identify, normal)",
}, []string{"state"})
)
type computebladeState struct {
mutex sync.Mutex
// identifyActive indicates whether the blade is currently in identify mode
identifyActive bool
identifyClearChan chan struct{}
// criticalActive indicates whether the blade is currently in critical mode
criticalActive bool
criticalClearChan chan struct{}
}
func NewComputeBladeState() *computebladeState {
return &computebladeState{
identifyClearChan: make(chan struct{}),
criticalClearChan: make(chan struct{}),
}
}
func (s *computebladeState) RegisterEvent(event Event) {
s.mutex.Lock()
defer s.mutex.Unlock()
switch event {
case IdentifyEvent:
s.identifyActive = true
case IdentifyConfirmEvent:
s.identifyActive = false
close(s.identifyClearChan)
s.identifyClearChan = make(chan struct{})
case CriticalEvent:
s.criticalActive = true
s.identifyActive = false
case CriticalResetEvent:
s.criticalActive = false
close(s.criticalClearChan)
s.criticalClearChan = make(chan struct{})
}
// Set identify state metric
if s.identifyActive {
stateMetric.WithLabelValues("identify").Set(1)
} else {
stateMetric.WithLabelValues("identify").Set(0)
}
// Set critical state metric
if s.criticalActive {
stateMetric.WithLabelValues("critical").Set(1)
} else {
stateMetric.WithLabelValues("critical").Set(0)
}
// Set critical state metric
if !s.criticalActive && !s.identifyActive {
stateMetric.WithLabelValues("normal").Set(1)
} else {
stateMetric.WithLabelValues("normal").Set(0)
}
}
func (s *computebladeState) IdentifyActive() bool {
return s.identifyActive
}
func (s *computebladeState) WaitForIdentifyClear(ctx context.Context) error {
select {
case <-ctx.Done():
return ctx.Err()
case <-s.identifyClearChan:
return nil
}
}
func (s *computebladeState) CriticalActive() bool {
return s.criticalActive
}
func (s *computebladeState) WaitForCriticalClear(ctx context.Context) error {
select {
case <-ctx.Done():
return ctx.Err()
case <-s.criticalClearChan:
return nil
}
}