Apigee Metrics pods in CrashLoopBackOff status

You're viewing Apigee and Apigee hybrid documentation.
There is no equivalent Apigee Edge documentation for this topic.

Symptom

When starting up, the Metrics pods remain in the CrashLoopBackoff state. This may cause periodic gaps in your metrics/graphs as the pods restart. This may also cause discrepancies with Analytics data as some sections of data are missing.

This issue can occur if your hybrid installation produces a large amount of metrics data. A large amount of data can be produced due to a high traffic load (resulting in a large number of underlying resources, for example MPs emitting metrics), or a large number of Apigee resources (for example, proxies, targets, envs, policies etc) being monitored.

Error messages

When you use kubectl to view the pod states, you see that one or more Metric pods are in the CrashLoopBackoff state. For example:

kubectl get pods -n NAMESPACE
NAME                                              READY   STATUS            RESTARTS   AGE

apigee-metrics-default-telemetry-proxy-1a2b3c4    0/1     CrashLoopBackoff  10         10m
apigee-metrics-adapter-apigee-telemetry-a2b3c4d   0/1     CrashLoopBackoff  10         10m
...

Possible causes

Cause Description Troubleshooting instructions applicable for
Metrics pods are out of memory Telemetry pods are in CrashLoopBackoff because of insufficient memory Apigee hybrid

Cause 1

Metric Pods are Out of Memory (OOM) with error reason OOMKilled.

Diagnosis

Check that the issue is occurring by inspecting the pod logs:

  1. List the pods to get the ID of the Metrics pod that is failing: 
    kubectl get pods -n APIGEE_NAMESPACE -l "app in (app, proxy, collector)"
  2. Check the failing pod's log: 
    kubectl -n APIGEE_NAMESPACE describe pods POD_NAME

For Example:

kubectl describe -n apigee pods apigee-metrics-default-telemetry-proxy-1a2b3c4

Investigate the apigee-prometheus-agg section of the output. Output like the following indicates that the container is repeatedly hitting an OOM condition: 

Containers:
  apigee-prometheus-agg:
    Container ID:  docker://cd893dbb06c2672c41a7d6f3f7d0de4d76742e68cef70d4250bf2d5cdfcdeae6
    Image:         us.gcr.io/apigee-saas-staging-repo/thirdparty/apigee-prom-prometheus/master:v2.9.2
    Image ID:      docker-pullable://us.gcr.io/apigee-saas-staging-repo/thirdparty/apigee-prom-prometheus/master@sha256:05350e0d1a577674442046961abf56b3e883dcd82346962f9e73f00667958f6b
    Port:          19090/TCP
    Host Port:     0/TCP
    Args:
      --config.file=/etc/prometheus/agg/prometheus.yml
      --storage.tsdb.path=/prometheus/agg/
      --storage.tsdb.retention=48h
      --web.enable-admin-api
      --web.listen-address=127.0.0.1:19090
    State:          Waiting
      Reason:       CrashLoopBackOff
    Last State:     Terminated
      Reason:       OOMKilled
      Exit Code:    137
      Started:      Wed, 21 Oct 2020 16:53:42 +0000
      Finished:     Wed, 21 Oct 2020 16:54:28 +0000
    Ready:          False
    Restart Count:  1446
    Limits:
      cpu:     500m
      memory:  512Mi
    Requests:
      cpu:     100m
      memory:  256Mi

Resolution

  1. Check the current container limits with the following command 
    kubectl -n APIGEE_NAMESPACE describe pods POD_NAME
  2. Configure the metrics pod limits in your overrides.yaml file with the following properties: 
    metrics:
  aggregator: # The apigee-prometheus-agg container in the "proxy" pod
    resources:
      limits:
        memory: 32Gi # default: 3Gi
  app: # The apigee-prometheus-app container in the "app" pod
    resources:
      limits:
        memory: 16Gi # default: 1Gi
  3. Apply the changes with the helm upgrade: 
    helm upgrade telemetry apigee-telemetry/ \
  --install \
  --namespace APIGEE_NAMESPACE \
  -f OVERRIDES_FILE

    If the OOM error is still encountered after increasing the limit, you can upsize the underlying nodes to allow for more memory usage.

Manage runtime plane components

Must gather diagnostic information

Gather the following diagnostic information, and then contact Apigee Support:

Gather Data from Prometheus Container for Troubleshooting

Start port-forwarding for the prometheus container. Repeat for both app and proxy pods. For example:

kubectl port-forward -n apigee apigee-metrics-apigee-telemetry-app-1a2-b3c4-d5ef 8081:9090

Use the script below within their cluster to collect data:

#!/bin/bash

set -e

# check if jq is installed
jq --version &> /dev/null
if [ $? -ne 0 ]; then
  echo "jq not installed"
  exit 1
fi

# check if curl is installed
curl --version &> /dev/null
if [ $? -ne 0 ] ; then
  echo "curl not installed"
  exit 1
fi

# Simple check for missing arguments
if [[ $# -eq 0 ]] ; then
  echo 'No arguments provided'
  exit 1
fi

# Simple check for missing arguments
if [[ $# -ne 3 ]]; then
  echo 'Illegal number of arguments'
  exit 1
fi

FORWARDED_PORT=${1}
DEST_DIR=${2}
CASE_NUMBER=${3}
DIR_FULL_PATH=${DEST_DIR}/${CASE_NUMBER}_$(date +%Y_%m_%d_%H_%M_%S)
CURRENT_DATE=$(date +%Y-%m-%d)

# we set the default start date for query at 10 days before current date
START_DATE=$(date +%Y-%m-%d -d "10 days ago")

mkdir -pv ${DIR_FULL_PATH}

set -x
curl -s '127.0.0.1:'${FORWARDED_PORT}'/status' | tee ${DIR_FULL_PATH}/prometheus_status_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).txt
curl -s '127.0.0.1:'${FORWARDED_PORT}'/config' | tee ${DIR_FULL_PATH}/prometheus_config_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).txt
curl -s '127.0.0.1:'${FORWARDED_PORT}'/api/v1/targets' | tee ${DIR_FULL_PATH}/prometheus_targets_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).json
curl -s '127.0.0.1:'${FORWARDED_PORT}'/api/v1/status/config' | jq . | tee ${DIR_FULL_PATH}/prometheus_status_config_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).json
curl -s '127.0.0.1:'${FORWARDED_PORT}'/debug/pprof/heap' --output ${DIR_FULL_PATH}/prometheus_heap_$(date +%Y.%m.%d_%H.%M.%S).hprof
curl -s '127.0.0.1:'${FORWARDED_PORT}'/debug/pprof/heap?debug=1' | tee ${DIR_FULL_PATH}/prometheus_heap_$(date +%Y.%m.%d_%H.%M.%S).txt
curl -s '127.0.0.1:'${FORWARDED_PORT}'/debug/pprof/goroutine' --output ${DIR_FULL_PATH}/prometheus_goroutine_$(date +%Y.%m.%d_%H.%M.%S)
curl -s '127.0.0.1:'${FORWARDED_PORT}'/debug/pprof/goroutine?debug=1' | tee ${DIR_FULL_PATH}/prometheus_goroutine_$(date +%Y.%m.%d_%H.%M.%S).txt
curl -s '127.0.0.1:'${FORWARDED_PORT}'/debug/pprof/profile?seconds=10' --output ${DIR_FULL_PATH}/prometheus_profile_10_seconds_$(date +%Y.%m.%d_%H.%M.%S)
curl -s '127.0.0.1:'${FORWARDED_PORT}'/api/v1/query?query=topk(30%2C%20count%20by%20(__name__)(%7B__name__%3D~%22.%2B%22%7D))&timeout=5s&start='${START_DATE}'T00:00:00.000Z&end='${CURRENT_DATE}'T23:59:59.00Z&step=15s' | jq . | tee ${DIR_FULL_PATH}/prometheus_topk_count_by_name_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).json
curl -s '127.0.0.1:'${FORWARDED_PORT}'/api/v1/query?query=topk(30%2C%20count%20by%20(__name__%2C%20job)(%7B__name__%3D~%22.%2B%22%7D))&timeout=5s&start='${START_DATE}'T00:00:00.000Z&end='${CURRENT_DATE}'T23:59:59.00Z&step=15s' | jq . | tee ${DIR_FULL_PATH}/prometheus_topk_group_by_job_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).json
curl -s '127.0.0.1:'${FORWARDED_PORT}'/api/v1/query?query=topk(30%2C%20count%20by%20(job)(%7B__name__%3D~%22.%2B%22%7D))&timeout=5s&start='${START_DATE}'T00:00:00.000Z&end='${CURRENT_DATE}'T23:59:59.00Z&step=15s' | jq . | tee ${DIR_FULL_PATH}/prometheus_topk_job_most_timeseries_$(hostname)-$(date +%Y.%m.%d_%H.%M.%S).json
set +x

ls -latrh ${DIR_FULL_PATH}

tar -cvzf ${DIR_FULL_PATH}.tar.gz ${DIR_FULL_PATH}/

exit 0

Arguments:

The script takes three positional arguments.

  1. Port number: Set to the port on which you forwarded from (e.g. 8081).
  2. Directory: Base directory for output files.
  3. Case number: Case number, used for the generated file name.

Example Usage

./prometheus_gather.sh 8081 . 1510679