1
0
Fork 0
mirror of https://github.com/TwiN/gatus.git synced 2024-12-15 17:51:09 +00:00
twin-gatus/watchdog/watchdog.go
I-HSIN Cheng 34313bec7e
fix(watchdog): Close dangling file descriptors on shutdown and config reload (#544)
* fix(watchdog): Add functions to avoid dangling file descriptors

* Change function name and add comment under core/endpoint.go
- change the function name of CloseHTTPConnection() to Close()
- add some comments above Close() function

* Update core/endpoint.go

* Update core/endpoint.go

---------

Co-authored-by: Richard Cheng <richard_cheng@trendmicro.com>
Co-authored-by: TwiN <twin@linux.com>
2023-08-04 18:30:15 -04:00

105 lines
4.4 KiB
Go

package watchdog
import (
"context"
"log"
"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/maintenance"
"github.com/TwiN/gatus/v5/core"
"github.com/TwiN/gatus/v5/metrics"
"github.com/TwiN/gatus/v5/storage/store"
)
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, cfg.Debug, ctx)
}
}
}
// monitor a single endpoint in a loop
func monitor(endpoint *core.Endpoint, alertingConfig *alerting.Config, maintenanceConfig *maintenance.Config, connectivityConfig *connectivity.Config, disableMonitoringLock, enabledMetrics, debug bool, ctx context.Context) {
// Run it immediately on start
execute(endpoint, alertingConfig, maintenanceConfig, connectivityConfig, disableMonitoringLock, enabledMetrics, debug)
// Loop for the next executions
for {
select {
case <-ctx.Done():
log.Printf("[watchdog][monitor] Canceling current execution of group=%s; endpoint=%s", endpoint.Group, endpoint.Name)
return
case <-time.After(endpoint.Interval):
execute(endpoint, alertingConfig, maintenanceConfig, connectivityConfig, disableMonitoringLock, enabledMetrics, debug)
}
}
}
func execute(endpoint *core.Endpoint, alertingConfig *alerting.Config, maintenanceConfig *maintenance.Config, connectivityConfig *connectivity.Config, disableMonitoringLock, enabledMetrics, debug 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() {
log.Println("[watchdog][execute] No connectivity; skipping execution")
return
}
if debug {
log.Printf("[watchdog][execute] Monitoring group=%s; endpoint=%s", endpoint.Group, endpoint.Name)
}
result := endpoint.EvaluateHealth()
if enabledMetrics {
metrics.PublishMetricsForEndpoint(endpoint, result)
}
UpdateEndpointStatuses(endpoint, result)
if debug && !result.Success {
log.Printf("[watchdog][execute] Monitored group=%s; endpoint=%s; success=%v; errors=%d; duration=%s; body=%s", endpoint.Group, endpoint.Name, result.Success, len(result.Errors), result.Duration.Round(time.Millisecond), result.Body)
} else {
log.Printf("[watchdog][execute] Monitored group=%s; endpoint=%s; success=%v; errors=%d; duration=%s", endpoint.Group, endpoint.Name, 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(endpoint, result, alertingConfig, debug)
} else if debug {
log.Println("[watchdog][execute] Not handling alerting because currently in the maintenance window")
}
if debug {
log.Printf("[watchdog][execute] Waiting for interval=%s before monitoring group=%s endpoint=%s again", endpoint.Interval, endpoint.Group, endpoint.Name)
}
}
// UpdateEndpointStatuses updates the slice of endpoint statuses
func UpdateEndpointStatuses(endpoint *core.Endpoint, result *core.Result) {
if err := store.Get().Insert(endpoint, result); err != nil {
log.Println("[watchdog][UpdateEndpointStatuses] Failed to insert data in storage:", err.Error())
}
}
// Shutdown stops monitoring all endpoints
func Shutdown(cfg *config.Config) {
// Disable all the old HTTP connections
for _, endpoint := range cfg.Endpoints {
endpoint.Close()
}
cancelFunc()
}