kubernetes-event-exporter
This tool is presented at KubeCon 2019 San Diego.
This tool allows exporting the often missed Kubernetes events to various outputs so that they can be used for observability or alerting purposes. You won't believe what you are missing.
Deployment
Head on to deploy/
folder and apply the YAMLs in the given filename order. Do not forget to modify the
deploy/01-config.yaml
file to your configuration needs. The additional information for configuration is as follows:
Configuration
Configuration is done via a YAML file, when run in Kubernetes, ConfigMap. The tool watches all the events and user has to option to filter out some events, according to their properties. Critical events can be routed to alerting tools such as Opsgenie, or all events can be dumped to an Elasticsearch instance. You can use namespaces, labels on the related object to route some Pod related events to owners via Slack. The final routing is a tree which allows flexibility. It generally looks like following:
route:
# Main route
routes:
# This route allows dumping all events because it has no fields to match and no drop rules.
- match:
- receiver: dump
# This starts another route, drops all the events in *test* namespaces and Normal events
# for capturing critical events
- drop:
- namespace: "*test*"
- type: "Normal"
match:
- receiver: "critical-events-queue"
# This a final route for user messages
- match:
- kind: "Pod|Deployment|ReplicaSet"
labels:
version: "dev"
receiver: "slack"
receivers:
# See below for configuring the receivers
- A
match
rule is exclusive, all conditions must be matched to the event. - During processing a route,
drop
rules are executed first to filter out events. - The
match
rules in a route are independent of each other. If an event matches a rule, it goes down it's subtree. - If all the
match
rules are matched, the event is passed to thereceiver
. - A route can have many sub-routes, forming a tree.
- Routing starts from the root route.
Opsgenie
Opsgenie is an alerting and on-call management tool. kubernetes-event-exporter can push to events to Opsgenie so that you can notify the on-call when something critical happens. Alerting should be precise and actionable, so you should carefully design what kind of alerts you would like in Opsgenie. A good starting point might be filtering out Normal type of events, while some additional filtering can help. Below is an example configuration.
# ...
receivers:
- name: "alerts"
opsgenie:
apiKey: xxx
priority: "P3"
message: "Event {{ .Reason }} for {{ .InvolvedObject.Namespace }}/{{ .InvolvedObject.Name }} on K8s cluster"
alias: "{{ .UID }}"
description: "<pre>{{ toPrettyJson . }}</pre>"
tags:
- "event"
- "{{ .Reason }}"
- "{{ .InvolvedObject.Kind }}"
- "{{ .InvolvedObject.Name }}"
Webhooks/HTTP
Webhooks are the easiest way of integrating this tool to external systems. It allows templating & custom headers which allows you to push events to many possible sources out there. See [Customizing Payload] for more information.
# ...
receivers:
- name: "alerts"
webhook:
endpoint: "https://my-super-secret-service.com"
headers:
X-API-KEY: "123"
User-Agent: kube-event-exporter 1.0
layout: # Optional
Elasticsearch
Elasticsearch is a full-text, distributed search engine which can also do powerful aggregations. You may decide to push all events to Elasticsearch and do some interesting queries over time to find out which images are pulled, how often pod schedules happen etc. You can watch the presentation in Kubecon to see what else you can do with aggregation and reporting.
# ...
receivers:
- name: "dump"
elasticsearch:
hosts:
- http://localhost:9200
index: kube-events
# Ca be used optionally for time based indices, accepts Go time formatting directives
indexFormat: "kube-events-{2006-01-02}"
username: # optional
password: # optional
cloudID: # optional
apiKey: # optional
# If set to true, it allows updating the same document in ES (might be useful handling count)
useEventID: true|false
# Type should be only used for clusters Version 6 and lower.
# type: kube-event
# If set to true, all dots in labels and annotation keys are replaced by underscores. Defaults false
deDot: true|false
layout: # Optional
tls: # optional, advanced options for tls
insecureSkipVerify: true|false # optional, if set to true, the tls cert won't be verified
serverName: # optional, the domain, the certificate was issued for, in case it doesn't match the hostname used for the connection
caFile: # optional, path to the CA file of the trusted authority the cert was signed with
Slack
Slack is a cloud-based instant messaging platform where many people use it for integrations and getting notified by software such as Jira, Opsgenie, Google Calendar etc. and even some implement ChatOps on it. This tool also allows exporting events to Slack channels or direct messages to persons. If your objects in Kubernetes, such as Pods, Deployments have real owners, you can opt-in to notify them via important events by using the labels of the objects. If a Pod sandbox changes and it's restarted, or it cannot find the Docker image, you can immediately notify the owner.
# ...
receivers:
- name: "slack"
slack:
token: YOUR-API-TOKEN-HERE
channel: "@{{ .InvolvedObject.Labels.owner }}"
message: "{{ .Message }}"
color: # optional
title: # optional
author_name: # optional
footer: # optional
fields:
namespace: "{{ .Namespace }}"
reason: "{{ .Reason }}"
object: "{{ .Namespace }}"
Kinesis
Kinesis is an AWS service allows to collect high throughput messages and allow it to be used in stream processing.
# ...
receivers:
- name: "kinesis"
kineis:
streamName: "events-pipeline"
region: us-west-2
layout: # Optional
Firehose
Firehose is an AWS service providing high throughput message collection for use in stream processing.
# ...
receivers:
- name: "firehose"
firehose:
deliveryStreamName: "events-pipeline"
region: us-west-2
layout: # Optional
SNS
SNS is an AWS service for highly durable pub/sub messaging system.
# ...
receivers:
- name: "sns"
sns:
topicARN: "arn:aws:sns:us-east-1:1234567890123456:mytopic"
region: "us-west-2"
layout: # Optional
SQS
SQS is an AWS service for message queuing that allows high throughput messaging.
# ...
receivers:
- name: "sqs"
sqs:
queueName: "/tmp/dump"
region: us-west-2
layout: # Optional
File
For some debugging purposes, you might want to push the events to files. Or you can already have a logging tool that can ingest these files and it might be a good idea to just use plain old school files as an integration point.
# ...
receivers:
- name: "file"
file:
path: "/tmp/dump"
layout: # Optional
Stdout
Standard out is also another file in Linux. logLevel
refers to the application logging severity - available levels
trace
, debug
, info
, warn
, error
, fatal
and panic
. When not specified, default level is set to info
. You
can use the following configuration as an example.
By default, events emit with eventime > 5seconds since catching are not collected. You can set this period with trottlePeriod in seconds. Consider to increase time of seconds to catch more events like "Backoff".
logLevel: error
logFormat: json
trottlePeriod: 5
route:
routes:
- match:
- receiver: "dump"
receivers:
- name: "dump"
stdout: { }
Kafka
Kafka is a popular tool used for real-time data pipelines. You can combine it with other tools for further analysis.
receivers:
- name: "kafka"
kafka:
clientId: "kubernetes"
topic: "kube-event"
brokers:
- "localhost:9092"
compressionCodec: "snappy"
tls:
enable: true
certFile: "kafka-client.crt"
keyFile: "kafka-client.key"
caFile: "kafka-ca.crt"
sasl:
enable: true
username: "kube-event-producer"
passsord: "kube-event-producer-password"
layout: #optionnal
kind: {{ .InvolvedObject.Kind }}
namespace: {{ .InvolvedObject.Namespace }}
name: {{ .InvolvedObject.Name }}
reason: {{ .Reason }}
message: {{ .Message }}
type: {{ .Type }}
createdAt: {{ .GetTimestampISO8601 }}
OpsCenter
OpsCenter provides a central location where operations engineers and IT professionals can view, investigate, and resolve operational work items (OpsItems) related to AWS resources. OpsCenter is designed to reduce mean time to resolution for issues impacting AWS resources. This Systems Manager capability aggregates and standardizes OpsItems across services while providing contextual investigation data about each OpsItem, related OpsItems, and related resources. OpsCenter also provides Systems Manager Automation documents (runbooks) that you can use to quickly resolve issues. You can specify searchable, custom data for each OpsItem. You can also view automatically-generated summary reports about OpsItems by status and source.
# ...
receivers:
- name: "alerts"
opscenter:
title: "{{ .Message }}",
category: "{{ .Reason }}", # Optional
description: "Event {{ .Reason }} for {{ .InvolvedObject.Namespace }}/{{ .InvolvedObject.Name }} on K8s cluster",
notifications: # Optional: SNS ARN
- "sns1"
- "sns2"
operationalData: # Optional
- Reason: ""{ { .Reason } }"}"
priority: "6", # Optional
region: "us-east1",
relatedOpsItems: # Optional: OpsItems ARN
- "ops1"
- "ops2"
severity: "6" # Optional
source: "production"
tags: # Optional
- ENV: "{{ .InvolvedObject.Namespace }}"
Customizing Payload
Some receivers allow customizing the payload. This can be useful to integrate it to external systems that require the data be in some format. It is designed to reduce the need for code writing. It allows mapping an event using Go templates, with sprig library additions. It supports a recursive map definition, so that you can create virtually any kind of JSON to be pushed to a webhook, a Kinesis stream, SQS queue etc.
# ...
receivers:
- name: pipe
kinesis:
region: us-west-2
streamName: event-pipeline
layout:
region: "us-west-2"
eventType: "kubernetes-event"
createdAt: "{{ .GetTimestampMs }}"
details:
message: "{{ .Message }}"
reason: "{{ .Reason }}"
type: "{{ .Type }}"
count: "{{ .Count }}"
kind: "{{ .InvolvedObject.Kind }}"
name: "{{ .InvolvedObject.Name }}"
namespace: "{{ .Namespace }}"
component: "{{ .Source.Component }}"
host: "{{ .Source.Host }}"
labels: "{{ toJson .InvolvedObject.Labels}}"
Pubsub
Pub/Sub is a fully-managed real-time messaging service that allows you to send and receive messages between independent applications.
receivers:
- name: "pubsub"
pubsub:
gcloud_project_id: "my-project"
topic: "kube-event"
create_topic: False
Teams
Microsoft Teams is your hub for teamwork in Office 365. All your team conversations, files, meetings, and apps live together in a single shared workspace, and you can take it with you on your favorite mobile device.
# ...
receivers:
- name: "ms_teams"
teams:
endpoint: "https://outlook.office.com/webhook/..."
layout: # Optional
Syslog
Syslog sink support enables to write k8s-events to syslog daemon server over tcp/udp. This can also be consumed by rsyslog.
# ...
receivers:
- name: "syslog"
syslog:
network: "tcp"
address: "127.0.0.1:11514"
tag: "k8s.event"
BigQuery
Google's query thing
receivers:
- name: "my-big-query"
bigquery:
location:
project:
dataset:
table:
credentials_path:
batch_size:
max_retries:
interval_seconds:
timeout_seconds:
Pipe
pipe output directly into some file descriptor
receivers:
- name: "my_pipe"
pipe:
path: "/dev/stdout"
AWS EventBridge
receivers:
- name: "eventbridge"
eventbridge:
detailType: "deployment"
source: "cd"
eventBusName: "default"
region: "ap-southeast-1"
details:
message: "{{ .Message }}"
namespace: "{{ .Namespace }}"
reason: "{{ .Reason }}"
object: "{{ .Namespace }}"