// MonitorPodHealth is meant to be a long running go routine.
// MonitorPodHealth reads from a consul store to determine which
// services should be running on the host. MonitorPodHealth
// runs a CheckHealth routine to monitor the health of each
// service and kills routines for services that should no
// longer be running.
func MonitorPodHealth(config *preparer.PreparerConfig, logger *logging.Logger, shutdownCh chan struct{}) {
store, err := config.GetStore()
if err != nil {
// A bad config should have already produced a nice, user-friendly error message.
logger.WithError(err).Fatalln("error creating health monitor KV store")
}
healthManager := store.NewHealthManager(config.NodeName, *logger)
node := config.NodeName
pods := []PodWatch{}
watchQuitCh := make(chan struct{})
watchErrCh := make(chan error)
watchPodCh := make(chan []kp.ManifestResult)
go store.WatchPods(
kp.REALITY_TREE,
node,
watchQuitCh,
watchErrCh,
watchPodCh,
)
// if GetClient fails it means the certfile/keyfile/cafile were
// invalid or did not exist. It makes sense to throw a fatal error
secureClient, err := config.GetClient(time.Duration(*HEALTHCHECK_TIMEOUT) * time.Second)
if err != nil {
logger.WithError(err).Fatalln("failed to get http client for this preparer")
}
insecureClient, err := config.GetInsecureClient(time.Duration(*HEALTHCHECK_TIMEOUT) * time.Second)
if err != nil {
logger.WithError(err).Fatalln("failed to get http client for this preparer")
}
for {
select {
case results := <-watchPodCh:
// check if pods have been added or removed
// starts monitor routine for new pods
// kills monitor routine for removed pods
pods = updatePods(healthManager, secureClient, insecureClient, pods, results, node, logger)
case err := <-watchErrCh:
logger.WithError(err).Errorln("there was an error reading reality manifests for health monitor")
case <-shutdownCh:
for _, pod := range pods {
pod.shutdownCh <- true
}
healthManager.Close()
close(watchQuitCh)
return
}
}
}
func verifyHealthChecks(config *preparer.PreparerConfig, services []string) error {
store, err := config.GetStore()
if err != nil {
return err
}
time.Sleep(5 * time.Second)
// check consul for health information for each app
name, err := os.Hostname()
if err != nil {
return err
}
node := types.NodeName(name)
for _, sv := range services {
res, err := store.GetHealth(sv, node)
if err != nil {
return err
} else if (res == kp.WatchResult{}) {
return fmt.Errorf("No results for %s: \n\n %s", sv, targetLogs())
} else if res.Status != string(health.Passing) {
return fmt.Errorf("%s did not pass health check: \n\n %s", sv, targetLogs())
} else {
fmt.Println(res)
}
}
for _, sv := range services {
res, err := store.GetServiceHealth(sv)
getres, _ := store.GetHealth(sv, node)
if err != nil {
return err
}
val := res[kp.HealthPath(sv, node)]
if getres.Id != val.Id || getres.Service != val.Service || getres.Status != val.Status {
return fmt.Errorf("GetServiceHealth failed %+v: \n\n%s", res, targetLogs())
}
}
// if it reaches here it means health checks
// are being written to the KV store properly
return nil
}