gatus/watchdog/watchdog.go
TwiN 01131755bc
fix(logging): Replace log-level parameter by GATUS_LOG_LEVEL env var (#895)
* fix(logging): Replace log-level parameter by GATUS_LOG_LEVEL env var

* Improve log message if GATUS_LOG_LEVEL isn't set
2024-11-13 23:54:00 -05:00

105 lines
4.6 KiB
Go

package watchdog
import (
"context"
"sync"
"time"
"github.com/TwiN/gatus/v5/alerting"
"github.com/TwiN/gatus/v5/config"
"github.com/TwiN/gatus/v5/config/connectivity"
"github.com/TwiN/gatus/v5/config/endpoint"
"github.com/TwiN/gatus/v5/config/maintenance"
"github.com/TwiN/gatus/v5/metrics"
"github.com/TwiN/gatus/v5/storage/store"
"github.com/TwiN/logr"
)
var (
// monitoringMutex is used to prevent multiple endpoint from being evaluated at the same time.
// Without this, conditions using response time may become inaccurate.
monitoringMutex sync.Mutex
ctx context.Context
cancelFunc context.CancelFunc
)
// Monitor loops over each endpoint and starts a goroutine to monitor each endpoint separately
func Monitor(cfg *config.Config) {
ctx, cancelFunc = context.WithCancel(context.Background())
for _, endpoint := range cfg.Endpoints {
if endpoint.IsEnabled() {
// To prevent multiple requests from running at the same time, we'll wait for a little before each iteration
time.Sleep(777 * time.Millisecond)
go monitor(endpoint, cfg.Alerting, cfg.Maintenance, cfg.Connectivity, cfg.DisableMonitoringLock, cfg.Metrics, ctx)
}
}
}
// monitor a single endpoint in a loop
func monitor(ep *endpoint.Endpoint, alertingConfig *alerting.Config, maintenanceConfig *maintenance.Config, connectivityConfig *connectivity.Config, disableMonitoringLock bool, enabledMetrics bool, ctx context.Context) {
// Run it immediately on start
execute(ep, alertingConfig, maintenanceConfig, connectivityConfig, disableMonitoringLock, enabledMetrics)
// Loop for the next executions
for {
select {
case <-ctx.Done():
logr.Warnf("[watchdog.monitor] Canceling current execution of group=%s; endpoint=%s; key=%s", ep.Group, ep.Name, ep.Key())
return
case <-time.After(ep.Interval):
execute(ep, alertingConfig, maintenanceConfig, connectivityConfig, disableMonitoringLock, enabledMetrics)
}
}
// Just in case somebody wandered all the way to here and wonders, "what about ExternalEndpoints?"
// Alerting is checked every time an external endpoint is pushed to Gatus, so they're not monitored
// periodically like they are for normal endpoints.
}
func execute(ep *endpoint.Endpoint, alertingConfig *alerting.Config, maintenanceConfig *maintenance.Config, connectivityConfig *connectivity.Config, disableMonitoringLock bool, enabledMetrics bool) {
if !disableMonitoringLock {
// By placing the lock here, we prevent multiple endpoints from being monitored at the exact same time, which
// could cause performance issues and return inaccurate results
monitoringMutex.Lock()
defer monitoringMutex.Unlock()
}
// If there's a connectivity checker configured, check if Gatus has internet connectivity
if connectivityConfig != nil && connectivityConfig.Checker != nil && !connectivityConfig.Checker.IsConnected() {
logr.Infof("[watchdog.execute] No connectivity; skipping execution")
return
}
logr.Debugf("[watchdog.execute] Monitoring group=%s; endpoint=%s; key=%s", ep.Group, ep.Name, ep.Key())
result := ep.EvaluateHealth()
if enabledMetrics {
metrics.PublishMetricsForEndpoint(ep, result)
}
UpdateEndpointStatuses(ep, result)
if logr.GetThreshold() == logr.LevelDebug && !result.Success {
logr.Debugf("[watchdog.execute] Monitored group=%s; endpoint=%s; key=%s; success=%v; errors=%d; duration=%s; body=%s", ep.Group, ep.Name, ep.Key(), result.Success, len(result.Errors), result.Duration.Round(time.Millisecond), result.Body)
} else {
logr.Infof("[watchdog.execute] Monitored group=%s; endpoint=%s; key=%s; success=%v; errors=%d; duration=%s", ep.Group, ep.Name, ep.Key(), result.Success, len(result.Errors), result.Duration.Round(time.Millisecond))
}
if !maintenanceConfig.IsUnderMaintenance() {
// TODO: Consider moving this after the monitoring lock is unlocked? I mean, how much noise can a single alerting provider cause...
HandleAlerting(ep, result, alertingConfig)
} else {
logr.Debug("[watchdog.execute] Not handling alerting because currently in the maintenance window")
}
logr.Debugf("[watchdog.execute] Waiting for interval=%s before monitoring group=%s endpoint=%s (key=%s) again", ep.Interval, ep.Group, ep.Name, ep.Key())
}
// UpdateEndpointStatuses updates the slice of endpoint statuses
func UpdateEndpointStatuses(ep *endpoint.Endpoint, result *endpoint.Result) {
if err := store.Get().Insert(ep, result); err != nil {
logr.Errorf("[watchdog.UpdateEndpointStatuses] Failed to insert result in storage: %s", err.Error())
}
}
// Shutdown stops monitoring all endpoints
func Shutdown(cfg *config.Config) {
// Disable all the old HTTP connections
for _, ep := range cfg.Endpoints {
ep.Close()
}
cancelFunc()
}