istio.io/content/help/ops/misc/index.md

title	description	weight	force_inline_toc
Miscellaneous	Advice on tackling common problems with Istio.	90	true

503 errors while reconfiguring service routes

When setting route rules to direct traffic to specific versions (subsets) of a service, care must be taken to ensure that the subsets are available before they are used in the routes. Otherwise, calls to the service may return 503 errors during a reconfiguration period.

Creating both the VirtualServices and DestinationRules that define the corresponding subsets using a single kubectl call (e.g., kubectl apply -f myVirtualServiceAndDestinationRule.yaml is not sufficient because the resources propagate (from the configuration server, i.e., Kubernetes API server) to the Pilot instances in an eventually consistent manner. If the VirtualService using the subsets arrives before the DestinationRule where the subsets are defined, the Envoy configuration generated by Pilot would refer to non-existent upstream pools, resulting in HTTP 503 errors until all configuration objects are available to Pilot.

To make sure services will have zero down-time when configuring routes with subsets, follow a "make-before-break" process as described below:

• When adding new subsets:

  1. Update DestinationRules to add a new subset first, before updating any VirtualServices that use it. Apply the rule using kubectl or any platform-specific tooling.

  2. Wait a few seconds for the DestinationRule configuration to propagate to the Envoys

  3. Update the VirtualService to refer to the newly added subsets.

• When removing subsets:

  1. Update VirtualServices to remove any references to a subset, before removing the subset from a DestinationRule.

  2. Wait a few seconds for the VirtualService configuration to propagate to the Envoys

  3. Update the DestinationRule to remove the unused subsets

Route rules have no effect on ingress gateway requests

Let's assume you are using an ingress Gateway and corresponding VirtualSerive to access an internal service. For example, your VirtualService looks something like this:

{{< text yaml >}} apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: myapp spec: hosts:

• "myapp.com" # or maybe "*" if you are testing without DNS using the ingress-gateway IP (e.g., http://1.2.3.4/hello) gateways:
• myapp-gateway http:
• match:
  • uri: prefix: /hello route:
  • destination: host: helloworld.default.svc.cluster.local
• match: ... {{< /text >}}

You also have a VirtualService which routes traffic for the helloworld service to a particular subset:

{{< text yaml >}} apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: helloworld spec: hosts:

• helloworld.default.svc.cluster.local http:
• route:
  • destination: host: helloworld.default.svc.cluster.local subset: v1 {{< /text >}}

In this situation you will notice that requests to the helloworld service via the ingress gateway will not be directed to subset v1 but instead will continue to use default round-robin routing.

The ingress requests are using the gateway host (e.g., myapp.com) which will activate the rules in the myapp VirtualService that routes to any endpoint in the helloworld service. On the other hand, internal requests with host helloworld.default.svc.cluster.local will use the helloworld VirtualService which directs traffic exclusively to subset v1.

To control the traffic from the gateway, you need to include the subset rule in the myapp VirtualService:

{{< text yaml >}} apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: myapp spec: hosts:

• "myapp.com" # or maybe "*" if you are testing without DNS using the ingress-gateway IP (e.g., http://1.2.3.4/hello) gateways:
• myapp-gateway http:
• match:
  • uri: prefix: /hello route:
  • destination: host: helloworld.default.svc.cluster.local subset: v1
• match: ... {{< /text >}}

Alternatively, you can combine both VirtualServices into one unit if possible:

{{< text yaml >}} apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: myapp spec: hosts:

• myapp.com # cannot use "*" here since this is being combined with the mesh services
• helloworld.default.svc.cluster.local gateways:
• mesh # applies internally as well as externally
• myapp-gateway http:
• match:
  • uri: prefix: /hello gateways:
    • myapp-gateway #restricts this rule to apply only to ingress gateway route:
  • destination: host: helloworld.default.svc.cluster.local subset: v1
• match:
  • gateways:
    • mesh # applies to all services inside the mesh route:
  • destination: host: helloworld.default.svc.cluster.local subset: v1 {{< /text >}}

Route rules have no effect on my application

If route rules are working perfectly for the Bookinfo sample, but similar version routing rules have no effect on your own application, it may be that your Kubernetes services need to be changed slightly.

Kubernetes services must adhere to certain restrictions in order to take advantage of Istio's L7 routing features. Refer to the Requirements for Pods and Services for details.

Verifying connectivity to Istio Pilot

Verifying connectivity to Pilot is a useful troubleshooting step. Every proxy container in the service mesh should be able to communicate with Pilot. This can be accomplished in a few simple steps:

1. Get the name of the Istio Ingress pod:

   {{< text bash >}} INGRESS_POD_NAME=(kubectl get po -n istio-system | grep ingressgateway- | awk '{print$1}'); echo ${INGRESS_POD_NAME}; {{< /text >}}

2. Exec into the Istio Ingress pod:

   {{< text bash >}} $ kubectl exec -it $INGRESS_POD_NAME -n istio-system /bin/bash {{< /text >}}

3. Test connectivity to Pilot using cURL. The following example cURL's the v1 registration API using default Pilot configuration parameters and mutual TLS enabled:

   {{< text bash >}} $ curl -k --cert /etc/certs/cert-chain.pem --cacert /etc/certs/root-cert.pem --key /etc/certs/key.pem https://istio-pilot:15003/v1/registration {{< /text >}}

   If mutual TLS is disabled:

   {{< text bash >}} $ curl http://istio-pilot:15003/v1/registration {{< /text >}}

You should receive a response listing the "service-key" and "hosts" for each service in the mesh.

No traces appearing in Zipkin when running Istio locally on Mac

Istio is installed and everything seems to be working except there are no traces showing up in Zipkin when there should be.

This may be caused by a known Docker issue where the time inside containers may skew significantly from the time on the host machine. If this is the case, when you select a very long date range in Zipkin you will see the traces appearing as much as several days too early.

You can also confirm this problem by comparing the date inside a docker container to outside:

{{< text bash >}} $ docker run --entrypoint date gcr.io/istio-testing/ubuntu-16-04-slave:latest Sun Jun 11 11:44:18 UTC 2017 {{< /text >}}

{{< text bash >}} $ date -u Thu Jun 15 02:25:42 UTC 2017 {{< /text >}}

To fix the problem, you'll need to shutdown and then restart Docker before reinstalling Istio.

Envoy won't connect to my HTTP/1.0 service

Envoy requires HTTP/1.1 or HTTP/2 traffic for upstream services. For example, when using NGINX for serving traffic behind Envoy, you will need to set the proxy_http_version directive in your NGINX configuration to be "1.1", since the NGINX default is 1.0

Example configuration:

{{< text plain >}} upstream http_backend { server 127.0.0.1:8080;

keepalive 16;

}

server { ...

location /http/ {
    proxy_pass http://http_backend;
    proxy_http_version 1.1;
    proxy_set_header Connection "";
    ...
}

} {{< /text >}}

No Grafana output when connecting from a local web client to Istio remotely hosted

Validate the client and server date and time match.

The time of the web client (e.g. Chrome) affects the output from Grafana. A simple solution to this problem is to verify a time synchronization service is running correctly within the Kubernetes cluster and the web client machine also is correctly using a time synchronization service. Some common time synchronization systems are NTP and Chrony. This is especially problematic is engineering labs with firewalls. In these scenarios, NTP may not be configured properly to point at the lab-based NTP services.

Where are the metrics for my service?

The expected flow of metrics is:

1. Envoy reports attributes to Mixer in batch (asynchronously from requests)
2. Mixer translates the attributes from Mixer into instances based on operator-provided configuration.
3. The instances are handed to Mixer adapters for processing and backend storage.
4. The backend storage systems record metrics data.

The default installations of Mixer ship with a Prometheus adapter, as well as configuration for generating a basic set of metric values and sending them to the Prometheus adapter. The Prometheus add-on also supplies configuration for an instance of Prometheus to scrape Mixer for metrics.

If you do not see the expected metrics in the Istio Dashboard and/or via Prometheus queries, there may be an issue at any of the steps in the flow listed above. Below is a set of instructions to troubleshoot each of those steps.

Verify Mixer is receiving Report calls

Mixer generates metrics for monitoring the behavior of Mixer itself. Check these metrics.

1. Establish a connection to the Mixer self-monitoring endpoint.

   In Kubernetes environments, execute the following command:

   {{< text bash >}} $ kubectl -n istio-system port-forward 9093 & {{< /text >}}

2. Verify successful report calls.

   On the Mixer self-monitoring endpoint, search for grpc_server_handled_total.

   You should see something like:

   {{< text plain >}} grpc_server_handled_total{grpc_code="OK",grpc_method="Report",grpc_service="istio.mixer.v1.Mixer",grpc_type="unary"} 68 {{< /text >}}

If you do not see any data for grpc_server_handled_total with a grpc_method="Report", then Mixer is not being called by Envoy to report telemetry. In this case, ensure that the services have been properly integrated into the mesh (either by via automatic or manual sidecar injection).

Verify Mixer metrics configuration exists

1. Verify Mixer rules exist.

   In Kubernetes environments, issue the following command:

   {{< text bash >}} $ kubectl get rules --all-namespaces NAMESPACE NAME KIND istio-system promhttp rule.v1alpha2.config.istio.io istio-system promtcp rule.v1alpha2.config.istio.io istio-system stdio rule.v1alpha2.config.istio.io {{< /text >}}

   If you do not see anything named promhttp or promtcp, then there is no Mixer configuration for sending metric instances to a Prometheus adapter. You will need to supply configuration for rules that connect Mixer metric instances to a Prometheus handler.

1. Verify the Prometheus handler configuration exists.

   In Kubernetes environments, issue the following command:

   {{< text bash >}} $ kubectl get prometheuses.config.istio.io --all-namespaces NAMESPACE NAME KIND istio-system handler prometheus.v1alpha2.config.istio.io {{< /text >}}

   If there are no prometheus handlers configured, you will need to reconfigure Mixer with the appropriate handler configuration.

1. Verify the Mixer metric instance configuration exists.

   In Kubernetes environments, issue the following command:

   {{< text bash >}} $ kubectl get metrics.config.istio.io --all-namespaces NAMESPACE NAME KIND istio-system requestcount metric.v1alpha2.config.istio.io istio-system requestduration metric.v1alpha2.config.istio.io istio-system requestsize metric.v1alpha2.config.istio.io istio-system responsesize metric.v1alpha2.config.istio.io istio-system stackdriverrequestcount metric.v1alpha2.config.istio.io istio-system stackdriverrequestduration metric.v1alpha2.config.istio.io istio-system stackdriverrequestsize metric.v1alpha2.config.istio.io istio-system stackdriverresponsesize metric.v1alpha2.config.istio.io istio-system tcpbytereceived metric.v1alpha2.config.istio.io istio-system tcpbytesent metric.v1alpha2.config.istio.io {{< /text >}}

   If there are no metric instances configured, you will need to reconfigure Mixer with the appropriate instance configuration.

1. Verify Mixer configuration resolution is working for your service.

   1. Establish a connection to the Mixer self-monitoring endpoint.

      Setup a port-forward to the Mixer self-monitoring port as described in Verify Mixer is receiving Report calls.

   2. On the Mixer self-monitoring port, search for mixer_config_resolve_count.

      You should find something like:

      {{< text plain >}} mixer_config_resolve_count{error="false",target="details.default.svc.cluster.local"} 56 mixer_config_resolve_count{error="false",target="ingress.istio-system.svc.cluster.local"} 67 mixer_config_resolve_count{error="false",target="mongodb.default.svc.cluster.local"} 18 mixer_config_resolve_count{error="false",target="productpage.default.svc.cluster.local"} 59 mixer_config_resolve_count{error="false",target="ratings.default.svc.cluster.local"} 26 mixer_config_resolve_count{error="false",target="reviews.default.svc.cluster.local"} 54 {{< /text >}}

   3. Validate that there are values for mixer_config_resolve_count where target="<your service>" and error="false".

      If there are only instances where error="true" where target=<your service>, there is likely an issue with Mixer configuration for your service. Logs information is needed to further debug.

      In Kubernetes environments, retrieve the Mixer logs via:

      {{< text bash >}} $ kubectl -n istio-system logs -c mixer {{< /text >}}

      Look for errors related to your configuration or your service in the returned logs.

More on viewing Mixer configuration can be found here

Verify Mixer is sending metric instances to the Prometheus adapter

1. Establish a connection to the Mixer self-monitoring endpoint.

   Setup a port-forward to the Mixer self-monitoring port as described in Verify Mixer is receiving Report calls.

2. On the Mixer self-monitoring port, search for mixer_adapter_dispatch_count.

   You should find something like:

   {{< text plain >}} mixer_adapter_dispatch_count{adapter="prometheus",error="false",handler="handler.prometheus.istio-system",meshFunction="metric",response_code="OK"} 114 mixer_adapter_dispatch_count{adapter="prometheus",error="true",handler="handler.prometheus.default",meshFunction="metric",response_code="INTERNAL"} 4 mixer_adapter_dispatch_count{adapter="stdio",error="false",handler="handler.stdio.istio-system",meshFunction="logentry",response_code="OK"} 104 {{< /text >}}

3. Validate that there are values for mixer_adapter_dispatch_count where adapter="prometheus" and error="false".

   If there are are no recorded dispatches to the Prometheus adapter, there is likely a configuration issue. Please see Verify Mixer metrics configuration exists.

   If dispatches to the Prometheus adapter are reporting errors, check the Mixer logs to determine the source of the error. Most likely, there is a configuration issue for the handler listed in mixer_adapter_dispatch_count.

   In Kubernetes environment, check the Mixer logs via:

   {{< text bash >}} $ kubectl -n istio-system logs -c mixer {{< /text >}}

   Filter for lines including something like Report 0 returned with: INTERNAL (1 error occurred: (with some surrounding context) to find more information regarding Report dispatch failures.

Verify Prometheus configuration

1. Connect to the Prometheus UI and verify that it can successfully scrape Mixer.

   In Kubernetes environments, setup port-forwarding as follows:

   {{< text bash >}} kubectl -n istio-system port-forward(kubectl -n istio-system get pod -l app=prometheus -o jsonpath='{.items[0].metadata.name}') 9090:9090 & {{< /text >}}

2. Visit http://localhost:9090/targets and confirm that the target istio-mesh has a status of UP.

3. Visit http://localhost:9090/config and confirm that an entry exists that looks like:

   {{< text yaml >}}

   • job_name: istio-mesh scrape_interval: 5s scrape_timeout: 5s metrics_path: /metrics scheme: http kubernetes_sd_configs:
   • api_server: null role: endpoints namespaces: names: [] relabel_configs:
   • source_labels: [__meta_kubernetes_namespace, __meta_kubernetes_service_name, __meta_kubernetes_endpoint_port_name] separator: ; regex: istio-system;istio-telemetry;prometheus replacement: $1 action: keep {{< /text >}}

How can I debug issues with the service mesh?

With istioctl

istioctl allows you to inspect the current xDS of a given Envoy from its admin interface (locally) or from Pilot using the proxy-config or pc command.

For example, to retrieve the configured clusters in an Envoy via the admin interface run the following command:

{{< text bash >}} $ istioctl proxy-config endpoint clusters {{< /text >}}

To retrieve endpoints for a given pod in the application namespace from Pilot run the following command:

{{< text bash >}} $ istioctl proxy-config pilot -n application eds {{< /text >}}

The proxy-config command also allows you to retrieve the state of the entire mesh from Pilot using the following command:

{{< text bash >}} $ istioctl proxy-config pilot mesh ads {{< /text >}}

With GDB

To debug Istio with gdb, you will need to run the debug images of Envoy / Mixer / Pilot. A recent gdb and the golang extensions (for Mixer/Pilot or other golang components) is required.

1. kubectl exec -it PODNAME -c [proxy | mixer | pilot]

2. Find process ID: ps ax

3. gdb -p PID binary

4. For go: info goroutines, goroutine x bt

With Tcpdump

Tcpdump doesn't work in the sidecar pod - the container doesn't run as root. However any other container in the same pod will see all the packets, since the network namespace is shared. iptables will also see the pod-wide configuration.

Communication between Envoy and the app happens on 127.0.0.1, and is not encrypted.

Envoy is crashing under load

Check your ulimit -a. Many systems have a 1024 open file descriptor limit by default which will cause Envoy to assert and crash with:

{{< text plain >}} [2017-05-17 03:00:52.735][14236][critical][assert] assert failure: fd_ != -1: external/envoy/source/common/network/connection_impl.cc:58 {{< /text >}}

Make sure to raise your ulimit. Example: ulimit -n 16384

Headless TCP services losing connection from Istiofied containers

If istio-citadel is deployed, Envoy is restarted every 15 minutes to refresh certificates. This causes the disconnection of TCP streams or long-running connections between services.

You should build resilience into your application for this type of disconnect, but if you still want to prevent the disconnects from happening, you will need to disable mutual TLS and the istio-citadel deployment.

First, edit your istio configuration to disable mutual TLS:

{{< text bash >}} $ kubectl edit configmap -n istio-system istio $ kubectl delete pods -n istio-system -l istio=pilot {{< /text >}}

Next, scale down the istio-citadel deployment to disable Envoy restarts:

{{< text bash >}} $ kubectl scale --replicas=0 deploy/istio-citadel -n istio-system {{< /text >}}

This should stop Istio from restarting Envoy and disconnecting TCP connections.

Envoy process has high CPU usage

For larger clusters, the default configuration that comes with Istio refreshes the Envoy configuration every 1 second. This can cause high CPU usage, even when Envoy isn't doing anything. In order to bring the CPU usage down for larger deployments, increase the refresh interval for Envoy to something higher, like 30 seconds.

{{< text bash >}} $ kubectl edit configmap -n istio-system istio $ kubectl delete pods -n istio-system -l istio=pilot {{< /text >}}

Also make sure to reinject the sidecar into all of your pods, as their configuration needs to be updated as well.

Afterwards, you should see CPU usage fall back to 0-1% while idling. Make sure to tune these values for your specific deployment.

Warning:: Changes created by routing rules will take up to 2x refresh interval to propagate to the sidecars. While the larger refresh interval will reduce CPU usage, updates caused by routing rules may cause a period of HTTP 404s (up to 2x the refresh interval) until the Envoy sidecars get all relevant configuration.

Automatic sidecar injection will fail if the kube-apiserver has proxy settings

When the Kube-apiserver included proxy settings such as:

{{< text yaml >}} env:

• name: http_proxy value: http://proxy-wsa.esl.foo.com:80
• name: https_proxy value: http://proxy-wsa.esl.foo.com:80
• name: no_proxy value: 127.0.0.1,localhost,dockerhub.foo.com,devhub-docker.foo.com,10.84.100.125,10.84.100.126,10.84.100.127 {{< /text >}}

The sidecar injection would fail. The only related failure logs was in the kube-apiserver log:

{{< text plain >}} W0227 21:51:03.156818 1 admission.go:257] Failed calling webhook, failing open sidecar-injector.istio.io: failed calling admission webhook "sidecar-injector.istio.io": Post https://istio-sidecar-injector.istio-system.svc:443/inject: Service Unavailable {{< /text >}}

Make sure both pod and service CIDRs are not proxied according to *_proxy variables. Check the kube-apiserver files and logs to verify the configuration and whether any requests are being proxied.

A workaround is to remove the proxy settings from the kube-apiserver manifest and restart the server or use a later version of Kubernetes.

An issue was filed with Kubernetes related to this and has since been closed. https://github.com/kubernetes/kubeadm/issues/666 https://github.com/kubernetes/kubernetes/pull/58698#discussion_r163879443

What Envoy version is istio using?

To find out the envoy version, you can follow below steps:

1. kubectl exec -it PODNAME -c istio-proxy -n NAMESPACE /bin/bash

2. curl localhost:15000/server_info