func (d *ClusterRegistry) getRegistryPods(service *kapi.Service, r types.DiagnosticResult) []*kapi.Pod {
runningPods := []*kapi.Pod{}
pods, err := d.KubeClient.Pods(kapi.NamespaceDefault).List(labels.SelectorFromSet(service.Spec.Selector), fields.Everything())
if err != nil {
r.Error("DClu1005", err, fmt.Sprintf("Finding pods for '%s' service failed. This should never happen. Error: (%T) %[2]v", registryName, err))
return runningPods
} else if len(pods.Items) < 1 {
r.Error("DClu1006", nil, fmt.Sprintf(clRegNoPods, registryName))
return runningPods
} else if len(pods.Items) > 1 {
// multiple registry pods using EmptyDir will be inconsistent
for _, volume := range pods.Items[0].Spec.Volumes {
if volume.Name == registryVolume && volume.EmptyDir != nil {
r.Error("DClu1007", nil, fmt.Sprintf(clRegMultiPods, registryName))
break
}
}
}
for _, pod := range pods.Items {
r.Debug("DClu1008", fmt.Sprintf("Found %s pod with name %s", registryName, pod.ObjectMeta.Name))
if pod.Status.Phase != kapi.PodRunning {
r.Warn("DClu1009", nil, fmt.Sprintf(clRegPodDown, pod.ObjectMeta.Name, registryName))
} else {
runningPods = append(runningPods, &pod)
// Check the logs for that pod for common issues (credentials, DNS resolution failure)
d.checkRegistryLogs(&pod, r)
}
}
return runningPods
}
func (d *ClusterRegistry) checkRegistryLogs(pod *kapi.Pod, r types.DiagnosticResult) {
// pull out logs from the pod
readCloser, err := d.KubeClient.RESTClient.Get().
Namespace("default").Name(pod.ObjectMeta.Name).
Resource("pods").SubResource("log").
Param("follow", "false").
Param("container", pod.Spec.Containers[0].Name).
Stream()
if err != nil {
r.Warn("DClu1010", nil, fmt.Sprintf(clRegPodLog, pod.ObjectMeta.Name, registryName, fmt.Sprintf("(%T) %[1]v", err)))
return
}
defer readCloser.Close()
clientError := ""
registryError := ""
scanner := bufio.NewScanner(readCloser)
for scanner.Scan() {
logLine := scanner.Text()
// TODO: once the logging API gets "since" and "tail" and "limit", limit to more recent log entries
// https://github.com/kubernetes/kubernetes/issues/12447
if strings.Contains(logLine, `level=error msg="client error:`) {
clientError = logLine // end up showing only the most recent client error
} else if strings.Contains(logLine, "level=error msg=") {
registryError += "\n" + logLine // gather generic errors
}
}
if clientError != "" {
r.Error("DClu1011", nil, fmt.Sprintf(clRegPodConn, pod.ObjectMeta.Name, registryName, clientError))
}
if registryError != "" {
r.Warn("DClu1012", nil, fmt.Sprintf(clRegPodErr, pod.ObjectMeta.Name, registryName, registryError))
}
}
func (d *ClusterRegistry) verifyRegistryImageStream(service *kapi.Service, r types.DiagnosticResult) {
if d.PreventModification {
r.Info("DClu1021", "Skipping creating an ImageStream to test registry service address, because you requested no API modifications.")
return
}
imgStream, err := d.OsClient.ImageStreams(kapi.NamespaceDefault).Create(&osapi.ImageStream{ObjectMeta: kapi.ObjectMeta{GenerateName: "diagnostic-test"}})
if err != nil {
r.Error("DClu1015", err, fmt.Sprintf("Creating test ImageStream failed. Error: (%T) %[1]v", err))
return
}
defer func() { // delete what we created, or notify that we couldn't
if err := d.OsClient.ImageStreams(kapi.NamespaceDefault).Delete(imgStream.ObjectMeta.Name); err != nil {
r.Warn("DClu1016", err, fmt.Sprintf(clRegISDelFail, imgStream.ObjectMeta.Name, fmt.Sprintf("(%T) %[1]s", err)))
}
}()
imgStream, err = d.OsClient.ImageStreams(kapi.NamespaceDefault).Get(imgStream.ObjectMeta.Name) // status is filled in post-create
if err != nil {
r.Error("DClu1017", err, fmt.Sprintf("Getting created test ImageStream failed. Error: (%T) %[1]v", err))
return
}
r.Debug("DClu1018", fmt.Sprintf("Created test ImageStream: %[1]v", imgStream))
cacheHost := strings.SplitN(imgStream.Status.DockerImageRepository, "/", 2)[0]
serviceHost := fmt.Sprintf("%s:%d", service.Spec.ClusterIP, service.Spec.Ports[0].Port)
if cacheHost != serviceHost {
r.Error("DClu1019", nil, fmt.Sprintf(clRegISMismatch, registryName, serviceHost, cacheHost))
}
}
// authenticateToMaster tests whether we can use the serviceaccount token
// to reach the master and authenticate
func (d PodCheckAuth) authenticateToMaster(token string, r types.DiagnosticResult) {
clientConfig := &clientcmd.Config{
MasterAddr: flagtypes.Addr{Value: d.MasterUrl}.Default(),
KubernetesAddr: flagtypes.Addr{Value: d.MasterUrl}.Default(),
CommonConfig: restclient.Config{
TLSClientConfig: restclient.TLSClientConfig{CAFile: d.MasterCaPath},
BearerToken: token,
},
}
oclient, _, _, err := clientConfig.Clients()
if err != nil {
r.Error("DP1002", err, fmt.Sprintf("could not create API clients from the service account client config: %v", err))
return
}
rchan := make(chan error, 1) // for concurrency with timeout
go func() {
_, err := oclient.Users().Get("~")
rchan <- err
}()
select {
case <-time.After(time.Second * 4): // timeout per query
r.Warn("DP1005", nil, "A request to the master timed out.\nThis could be temporary but could also indicate network or DNS problems.")
case err := <-rchan:
if err != nil {
r.Error("DP1003", err, fmt.Sprintf("Could not authenticate to the master with the service account credentials: %v", err))
} else {
r.Info("DP1004", "Service account token successfully authenticated to master")
}
}
return
}
func (d PodCheckAuth) authenticateToRegistry(token string, r types.DiagnosticResult) {
resolvConf, err := getResolvConf(r)
if err != nil {
return // any errors have been reported via "r", env is very borked, test cannot proceed
}
msg := new(dns.Msg)
msg.SetQuestion(registryHostname+".", dns.TypeA)
msg.RecursionDesired = false
result, completed := dnsQueryWithTimeout(msg, resolvConf.Servers[0], 2)
switch {
case !completed:
r.Error("DP1006", nil, fmt.Sprintf("DNS resolution for registry address %s timed out; this could indicate problems with DNS resolution or networking", registryHostname))
return
case result.err != nil:
r.Error("DP1016", nil, fmt.Sprintf("DNS resolution for registry address %s returned an error; container DNS is likely incorrect. The error was: %v", registryHostname, result.err))
return
case result.in == nil, len(result.in.Answer) == 0:
r.Warn("DP1007", nil, fmt.Sprintf("DNS resolution for registry address %s returned no results; either the integrated registry is not deployed, or container DNS configuration is incorrect.", registryHostname))
return
}
// first try the secure connection in case they followed directions to secure the registry
// (https://docs.openshift.org/latest/install_config/install/docker_registry.html#securing-the-registry)
cacert, err := ioutil.ReadFile(d.MasterCaPath) // TODO: we assume same CA as master - better choice?
if err != nil {
r.Error("DP1008", err, fmt.Sprintf("Failed to read CA cert file %s:\n%v", d.MasterCaPath, err))
return
}
pool := x509.NewCertPool()
if !pool.AppendCertsFromPEM(cacert) {
r.Error("DP1009", err, fmt.Sprintf("Could not use cert from CA cert file %s:\n%v", d.MasterCaPath, err))
return
}
noSecClient := http.Client{
CheckRedirect: func(req *http.Request, via []*http.Request) error {
return fmt.Errorf("no redirect expected")
},
Timeout: time.Second * 2,
}
secClient := noSecClient
secClient.Transport = knet.SetTransportDefaults(&http.Transport{TLSClientConfig: &tls.Config{RootCAs: pool}})
secError := processRegistryRequest(&secClient, fmt.Sprintf("https://%s:%s/v2/", registryHostname, registryPort), token, r)
if secError == nil {
return // made the request successfully enough to diagnose
}
switch {
case strings.Contains(secError.Error(), "tls: oversized record received"),
strings.Contains(secError.Error(), "server gave HTTP response to HTTPS"):
r.Debug("DP1015", "docker-registry not secured; falling back to cleartext connection")
if nosecError := processRegistryRequest(&noSecClient, fmt.Sprintf("http://%s:%s/v2/", registryHostname, registryPort), token, r); nosecError != nil {
r.Error("DP1013", nosecError, fmt.Sprintf("Unexpected error authenticating to the integrated registry:\n(%T) %[1]v", nosecError))
}
default:
r.Error("DP1013", secError, fmt.Sprintf("Unexpected error authenticating to the integrated registry:\n(%T) %[1]v", secError))
}
}
func (d *ClusterRegistry) getRegistryService(r types.DiagnosticResult) *kapi.Service {
service, err := d.KubeClient.Services(kapi.NamespaceDefault).Get(registryName)
if err != nil && reflect.TypeOf(err) == reflect.TypeOf(&kerrs.StatusError{}) {
r.Warn("DClu1002", err, fmt.Sprintf(clGetRegNone, registryName, kapi.NamespaceDefault))
return nil
} else if err != nil {
r.Error("DClu1003", err, fmt.Sprintf(clGetRegFailed, err))
return nil
}
r.Debug("DClu1004", fmt.Sprintf("Found %s service with ports %v", registryName, service.Spec.Ports))
return service
}
func (d *ClusterRouter) getRouterDC(r types.DiagnosticResult) *osapi.DeploymentConfig {
dc, err := d.OsClient.DeploymentConfigs(kapi.NamespaceDefault).Get(routerName)
if err != nil && reflect.TypeOf(err) == reflect.TypeOf(&kerrs.StatusError{}) {
r.Warn("DClu2001", err, fmt.Sprintf(clGetRtNone, routerName))
return nil
} else if err != nil {
r.Error("DClu2002", err, fmt.Sprintf(clGetRtFailed, routerName, err))
return nil
}
r.Debug("DClu2003", fmt.Sprintf("Found default router DC"))
return dc
}
func (d *ClusterRegistry) getRegistryPods(service *kapi.Service, r types.DiagnosticResult) []*kapi.Pod {
runningPods := []*kapi.Pod{}
pods, err := d.KubeClient.Core().Pods(kapi.NamespaceDefault).List(kapi.ListOptions{LabelSelector: labels.SelectorFromSet(service.Spec.Selector)})
if err != nil {
r.Error("DClu1005", err, fmt.Sprintf("Finding pods for '%s' service failed. This should never happen. Error: (%T) %[2]v", registryName, err))
return runningPods
} else if len(pods.Items) < 1 {
r.Error("DClu1006", nil, fmt.Sprintf(clRegNoPods, registryName))
return runningPods
} else if len(pods.Items) > 1 {
emptyDir := false // multiple registry pods using EmptyDir will be inconsistent
customConfig := false // ... unless the user has configured them for e.g. S3
configPath := "/config.yml"
// look through the pod volumes to see if that might be a problem
podSpec := pods.Items[0].Spec
container := podSpec.Containers[0]
for _, volume := range podSpec.Volumes {
if volume.Name == registryVolume && volume.EmptyDir != nil {
emptyDir = true
}
}
for _, env := range container.Env {
if env.Name == "REGISTRY_CONFIGURATION_PATH" {
configPath = env.Value // look for custom config here
}
}
for _, vmount := range container.VolumeMounts {
if strings.HasPrefix(configPath, vmount.MountPath) {
customConfig = true // if something's mounted there, assume custom config.
}
}
if emptyDir {
if customConfig { // assume they know what they're doing
r.Info("DClu1020", fmt.Sprintf(clRegMultiCustomCfg, registryName, configPath))
} else { // assume they scaled up with ephemeral storage
r.Error("DClu1007", nil, fmt.Sprintf(clRegMultiPods, registryName))
}
}
}
for _, pod := range pods.Items {
r.Debug("DClu1008", fmt.Sprintf("Found %s pod with name %s", registryName, pod.ObjectMeta.Name))
if pod.Status.Phase != kapi.PodRunning {
r.Warn("DClu1009", nil, fmt.Sprintf(clRegPodDown, pod.ObjectMeta.Name, registryName))
} else {
runningPods = append(runningPods, &pod)
// Check the logs for that pod for common issues (credentials, DNS resolution failure)
d.checkRegistryLogs(&pod, r)
}
}
return runningPods
}
func (d *ClusterRegistry) checkRegistryLogs(pod *kapi.Pod, r types.DiagnosticResult) {
// pull out logs from the pod
readCloser, err := d.KubeClient.RESTClient.Get().
Namespace("default").Name(pod.ObjectMeta.Name).
Resource("pods").SubResource("log").
Param("follow", "false").
Param("container", pod.Spec.Containers[0].Name).
Stream()
if err != nil {
r.Warn("DClu1010", nil, fmt.Sprintf(clRegPodLog, pod.ObjectMeta.Name, registryName, fmt.Sprintf("(%T) %[1]v", err)))
return
}
defer readCloser.Close()
// Indicator that selinux is blocking the registry from writing to disk:
selinuxErrorRegex, _ := regexp.Compile(".*level=error.*mkdir.*permission denied.*")
// If seen after the above error regex, we know the problem has since been fixed:
selinuxSuccessRegex, _ := regexp.Compile(".*level=info.*response completed.*http.request.method=PUT.*")
clientError := ""
registryError := ""
selinuxError := ""
scanner := bufio.NewScanner(readCloser)
for scanner.Scan() {
logLine := scanner.Text()
// TODO: once the logging API gets "since" and "tail" and "limit", limit to more recent log entries
// https://github.com/kubernetes/kubernetes/issues/12447
if strings.Contains(logLine, `level=error msg="client error:`) {
clientError = logLine // end up showing only the most recent client error
} else if selinuxErrorRegex.MatchString(logLine) {
selinuxError = logLine
} else if selinuxSuccessRegex.MatchString(logLine) {
// Check for a successful registry push, if this occurs after a selinux error
// we can safely clear it, the problem has already been fixed.
selinuxError = ""
} else if strings.Contains(logLine, "level=error msg=") {
registryError += "\n" + logLine // gather generic errors
}
}
if clientError != "" {
r.Error("DClu1011", nil, fmt.Sprintf(clRegPodConn, pod.ObjectMeta.Name, registryName, clientError))
}
if selinuxError != "" {
r.Error("DClu1020", nil, fmt.Sprintf(clRegSelinuxErr, pod.ObjectMeta.Name, registryName, selinuxError))
}
if registryError != "" {
r.Warn("DClu1012", nil, fmt.Sprintf(clRegPodErr, pod.ObjectMeta.Name, registryName, registryError))
}
}
func (d *ClusterRegistry) checkRegistryEndpoints(pods []*kapi.Pod, r types.DiagnosticResult) bool {
endPoint, err := d.KubeClient.Endpoints(kapi.NamespaceDefault).Get(registryName)
if err != nil {
r.Error("DClu1013", err, fmt.Sprintf(`Finding endpoints for "%s" service failed. This should never happen. Error: (%[2]T) %[2]v`, registryName, err))
return false
}
numEP := 0
for _, subs := range endPoint.Subsets {
numEP += len(subs.Addresses)
}
if numEP != len(pods) {
r.Warn("DClu1014", nil, fmt.Sprintf(clRegNoEP, registryName, len(pods), numEP))
return false
}
return true
}
func retrieveLoggingProject(r types.DiagnosticResult, masterCfg *configapi.MasterConfig, osClient *client.Client) string {
r.Debug("AGL0010", fmt.Sprintf("masterConfig.AssetConfig.LoggingPublicURL: '%s'", masterCfg.AssetConfig.LoggingPublicURL))
projectName := ""
if len(masterCfg.AssetConfig.LoggingPublicURL) == 0 {
r.Debug("AGL0017", "masterConfig.AssetConfig.LoggingPublicURL is empty")
return projectName
}
loggingUrl, err := url.Parse(masterCfg.AssetConfig.LoggingPublicURL)
if err != nil {
r.Error("AGL0011", err, fmt.Sprintf("Unable to parse the loggingPublicURL from the masterConfig '%s'", masterCfg.AssetConfig.LoggingPublicURL))
return projectName
}
routeList, err := osClient.Routes(kapi.NamespaceAll).List(kapi.ListOptions{LabelSelector: loggingSelector.AsSelector()})
if err != nil {
r.Error("AGL0012", err, fmt.Sprintf("There was an error while trying to find the route associated with '%s' which is probably transient: %s", loggingUrl, err))
return projectName
}
for _, route := range routeList.Items {
r.Debug("AGL0013", fmt.Sprintf("Comparing URL to route.Spec.Host: %s", route.Spec.Host))
if loggingUrl.Host == route.Spec.Host {
if len(projectName) == 0 {
projectName = route.ObjectMeta.Namespace
r.Info("AGL0015", fmt.Sprintf("Found route '%s' matching logging URL '%s' in project: '%s'", route.ObjectMeta.Name, loggingUrl.Host, projectName))
} else {
r.Warn("AGL0019", nil, fmt.Sprintf("Found additional route '%s' matching logging URL '%s' in project: '%s'. This could mean you have multiple logging deployments.", route.ObjectMeta.Name, loggingUrl.Host, projectName))
}
}
}
if len(projectName) == 0 {
message := fmt.Sprintf("Unable to find a route matching the loggingPublicURL defined in the master config '%s'. Check that the URL is correct and aggregated logging is deployed.", loggingUrl)
r.Error("AGL0014", errors.New(message), message)
return ""
}
project, err := osClient.Projects().Get(projectName)
if err != nil {
r.Error("AGL0018", err, fmt.Sprintf("There was an error retrieving project '%s' which is most likely a transient error: %s", projectName, err))
return ""
}
nodeSelector, ok := project.ObjectMeta.Annotations["openshift.io/node-selector"]
if !ok || len(nodeSelector) != 0 {
r.Warn("AGL0030", nil, fmt.Sprintf(projectNodeSelectorWarning, projectName))
}
return projectName
}
// getResolvConf reads a clientConfig from resolv.conf and complains about any errors
func getResolvConf(r types.DiagnosticResult) (*dns.ClientConfig, error) {
resolvConf, err := dns.ClientConfigFromFile("/etc/resolv.conf")
if err != nil {
r.Error("DP3001", err, fmt.Sprintf("could not load/parse resolver file /etc/resolv.conf: %v", err))
return nil, err
}
if len(resolvConf.Servers) == 0 {
r.Error("DP3002", nil, "could not find any nameservers defined in /etc/resolv.conf")
return nil, err
}
if len(resolvConf.Search) == 0 {
r.Warn("DP3011", nil, "could not find any search domains defined in /etc/resolv.conf")
resolvConf.Search = nil
}
r.Debug("DP3012", fmt.Sprintf("Pod /etc/resolv.conf contains:\nnameservers: %v\nsearch domains: %v", resolvConf.Servers, resolvConf.Search))
return resolvConf, nil
}
func resolveSearch(resolvConf *dns.ClientConfig, r types.DiagnosticResult) {
foundDomain := false
randomString := func() string {
b := make([]byte, 20)
for i := range b {
b[i] = letterBytes[rand.Intn(len(letterBytes))]
}
return string(b)
}()
seenDP2014 := sets.String{}
seenDP2015 := sets.String{}
for _, domain := range resolvConf.Search {
if domain == "svc.cluster.local" {
foundDomain = true // this will make kubernetes.default work
}
// put together a DNS query to configured nameservers for each search domain
msg := new(dns.Msg)
msg.SetQuestion("wildcard."+randomString+"."+domain+".", dns.TypeA)
msg.RecursionDesired = true // otherwise we just get the authority section for the TLD
for _, server := range resolvConf.Servers {
result, completed := dnsQueryWithTimeout(msg, server, 2)
switch {
case !completed:
if !seenDP2014.Has(server) {
r.Warn("DP2014", nil, fmt.Sprintf("A request to the nameserver %s timed out.\nThis could be temporary but could also indicate network or DNS problems.", server))
seenDP2014.Insert(server) // no need to keep warning about the same server for every domain
}
case result.err != nil:
if !seenDP2015.Has(server) {
r.Warn("DP2015", result.err, fmt.Sprintf("Error querying nameserver %s:\n %v\nThis may indicate a problem with DNS.", server, result.err))
seenDP2015.Insert(server) // don't repeat the error for the same nameserver; chances are it's the same error
}
case result.in.Answer == nil, len(result.in.Answer) == 0:
r.Debug("DP2017", fmt.Sprintf("Nameserver %s responded to wildcard with no answer, which is expected.\n%v", server, result.in))
default: // the random domain is not supposed to resolve
r.Error("DP2016", nil, fmt.Sprintf(txtDP2016, server, domain, result.in))
}
}
}
if !foundDomain {
r.Error("DP2019", nil, "Did not find svc.cluster.local among the configured search domains in /etc/resolv.conf.\nThis is likely to cause problems with certain components that expect to use partial cluster addresses.")
}
}
func (d *ClusterRouter) checkRouterLogs(pod *kapi.Pod, r types.DiagnosticResult) {
scanner, err := d.getPodLogScanner(pod)
if err != nil {
r.Warn("DClu2008", err, fmt.Sprintf(clRtPodLog, pod.ObjectMeta.Name, fmt.Sprintf("(%T) %[1]v", err)))
return
}
defer scanner.Close()
for scanner.Scan() {
matches := regexp.MustCompile(`^(\S+).*Failed to list \*api.Route: (.*)`).FindStringSubmatch(scanner.Text())
if len(matches) > 0 {
stamp, err := time.Parse(referenceTimestampLayout, matches[1])
// router checks every second. error only if failure is recent.
// of course... we cannot always trust the local clock.
if err == nil && time.Since(stamp).Seconds() < 30.0 {
r.Error("DClu2009", nil, fmt.Sprintf(clRtPodConn, pod.ObjectMeta.Name, matches[2], matches[1]))
break
}
}
}
}
func connectAndResolve(resolvConf *dns.ClientConfig, r types.DiagnosticResult) {
for serverIndex, server := range resolvConf.Servers {
// put together a DNS query to configured nameservers for kubernetes.default
msg := new(dns.Msg)
msg.SetQuestion("kubernetes.default.svc.cluster.local.", dns.TypeA)
msg.RecursionDesired = false
if result, completed := dnsQueryWithTimeout(msg, server, 2); !completed {
if serverIndex == 0 { // in a pod, master (SkyDNS) IP is injected as first nameserver
r.Warn("DP2009", nil, fmt.Sprintf("A request to the master (SkyDNS) nameserver %s timed out.\nThis could be temporary but could also indicate network or DNS problems.\nThis nameserver is critical for resolving cluster DNS names.", server))
} else {
r.Warn("DP2010", nil, fmt.Sprintf("A request to the nameserver %s timed out.\nThis could be temporary but could also indicate network or DNS problems.", server))
}
} else {
in, err := result.in, result.err
if serverIndex == 0 { // in a pod, master (SkyDNS) IP is injected as first nameserver
if err != nil {
r.Error("DP2003", err, fmt.Sprintf("The first /etc/resolv.conf nameserver %s\ncould not resolve kubernetes.default.svc.cluster.local.\nError: %v\nThis nameserver points to the master's SkyDNS which is critical for\nresolving cluster names, e.g. for Services.", server, err))
} else if len(in.Answer) == 0 {
r.Error("DP2006", err, fmt.Sprintf("The first /etc/resolv.conf nameserver %s\ncould not resolve kubernetes.default.svc.cluster.local.\nReturn code: %v\nThis nameserver points to the master's SkyDNS which is critical for\nresolving cluster names, e.g. for Services.", server, dns.RcodeToString[in.MsgHdr.Rcode]))
} else {
r.Debug("DP2007", fmt.Sprintf("The first /etc/resolv.conf nameserver %s\nresolved kubernetes.default.svc.cluster.local. to:\n %s", server, in.Answer[0]))
}
} else if err != nil {
r.Warn("DP2004", err, fmt.Sprintf("Error querying nameserver %s:\n %v\nThis may indicate a problem with non-cluster DNS.", server, err))
} else {
rcode := in.MsgHdr.Rcode
switch rcode {
case dns.RcodeSuccess, dns.RcodeNameError: // aka NXDOMAIN
r.Debug("DP2005", fmt.Sprintf("Successful query to nameserver %s", server))
default:
r.Warn("DP2008", nil, fmt.Sprintf("Received unexpected return code '%s' from nameserver %s:\nThis may indicate a problem with non-cluster DNS.", dns.RcodeToString[rcode], server))
}
}
}
}
}
func (d *DiagnosticPod) runDiagnosticPod(service *kapi.Service, r types.DiagnosticResult) {
loglevel := d.Level
if loglevel > 2 {
loglevel = 2 // need to show summary at least
}
imageName := d.ImageTemplate.ExpandOrDie("deployer")
pod, err := d.KubeClient.Core().Pods(d.Namespace).Create(&kapi.Pod{
ObjectMeta: kapi.ObjectMeta{GenerateName: "pod-diagnostic-test-"},
Spec: kapi.PodSpec{
RestartPolicy: kapi.RestartPolicyNever,
Containers: []kapi.Container{
{
Name: "pod-diagnostics",
Image: imageName,
Command: []string{"openshift", "infra", "diagnostic-pod", "-l", strconv.Itoa(loglevel)},
},
},
},
})
if err != nil {
r.Error("DCli2001", err, fmt.Sprintf("Creating diagnostic pod with image %s failed. Error: (%[2]T) %[2]v", imageName, err))
return
}
defer func() { // delete what we created, or notify that we couldn't
zero := int64(0)
delOpts := kapi.DeleteOptions{TypeMeta: pod.TypeMeta, GracePeriodSeconds: &zero}
if err := d.KubeClient.Core().Pods(d.Namespace).Delete(pod.ObjectMeta.Name, &delOpts); err != nil {
r.Error("DCl2002", err, fmt.Sprintf("Deleting diagnostic pod '%s' failed. Error: %s", pod.ObjectMeta.Name, fmt.Sprintf("(%T) %[1]s", err)))
}
}()
pod, err = d.KubeClient.Core().Pods(d.Namespace).Get(pod.ObjectMeta.Name) // status is filled in post-create
if err != nil {
r.Error("DCli2003", err, fmt.Sprintf("Retrieving the diagnostic pod definition failed. Error: (%T) %[1]v", err))
return
}
r.Debug("DCli2004", fmt.Sprintf("Created diagnostic pod named %v running image %s.", pod.ObjectMeta.Name, imageName))
bytelim := int64(1024000)
podLogsOpts := &kapi.PodLogOptions{
TypeMeta: pod.TypeMeta,
Container: "pod-diagnostics",
Follow: true,
LimitBytes: &bytelim,
}
req, err := d.Factory.LogsForObject(pod, podLogsOpts)
if err != nil {
r.Error("DCli2005", err, fmt.Sprintf("The request for diagnostic pod logs failed unexpectedly. Error: (%T) %[1]v", err))
return
}
// wait for pod to be started and logs available
var scanner *bufio.Scanner
var lastError error
for times := 1; true; times++ {
if times <= 25 {
readCloser, err := req.Stream()
if err != nil {
lastError = err
r.Debug("DCli2010", fmt.Sprintf("Could not get diagnostic pod logs (loop %d): (%T[2]) %[2]v", times, err))
time.Sleep(time.Duration(times*100) * time.Millisecond)
continue
}
defer readCloser.Close()
// make sure we can actually get something from the stream before going on.
// it seems the creation of docker logs can trail the container start a bit.
lineScanner := bufio.NewScanner(readCloser)
if lineScanner.Scan() {
scanner = lineScanner
break // success - drop down to reading the logs.
}
// no luck - try, try again
lastError = fmt.Errorf("Diagnostics pod is ready but not its logs (loop %d). Retry.", times)
r.Debug("DCli2010", lastError.Error())
time.Sleep(time.Duration(times*100) * time.Millisecond)
continue
}
// after 25 times trying:
r.Warn("DCli2006", err, fmt.Sprintf("Timed out preparing diagnostic pod logs for streaming, so this diagnostic cannot run.\nIt is likely that the image '%s' was not pulled and running yet.\nLast error: (%T[2]) %[2]v", pod.Spec.Containers[0].Image, lastError))
return
}
// then watch logs and wait until it exits
podLogs, warnings, errors := "", 0, 0
errorRegex := regexp.MustCompile(`^\[Note\]\s+Errors\s+seen:\s+(\d+)`)
warnRegex := regexp.MustCompile(`^\[Note\]\s+Warnings\s+seen:\s+(\d+)`)
// keep in mind one test line was already scanned, so scan after the loop runs once
for scanned := true; scanned; scanned = scanner.Scan() {
line := scanner.Text()
podLogs += line + "\n"
if matches := errorRegex.FindStringSubmatch(line); matches != nil {
errors, _ = strconv.Atoi(matches[1])
} else if matches := warnRegex.FindStringSubmatch(line); matches != nil {
warnings, _ = strconv.Atoi(matches[1])
}
}
if err := scanner.Err(); err != nil { // Scan terminated abnormally
r.Error("DCli2009", err, fmt.Sprintf("Unexpected error reading diagnostic pod logs: (%T) %[1]v\nLogs are:\n%[2]s", err, podLogs))
} else {
if errors > 0 {
r.Error("DCli2012", nil, "See the errors below in the output from the diagnostic pod:\n"+podLogs)
} else if warnings > 0 {
r.Warn("DCli2013", nil, "See the warnings below in the output from the diagnostic pod:\n"+podLogs)
} else {
r.Info("DCli2008", fmt.Sprintf("Output from the diagnostic pod (image %s):\n", imageName)+podLogs)
}
}
}