[receiver/k8sobjectsreceiver] watcher not restarting when kubernetes hangs up.

[stokerjon]

Component(s)

receiver/k8sobjects

What happened?

Description

Kubernetes periodically hangs up watch connections, the receiver does not then restart them.

Steps to Reproduce

Create a k8sobjectsreceiver watching for kubernetes events.

Expected Result

k8sobjectsreceiver will continue watching for events and when kubernetes hangs up restart the watch

Actual Result

Kubernetes hangs up k8sobjectsreceiver watch. but does not restart

Collector version

0.68.0

Environment information

Environment

OS: Amazon Linux

OpenTelemetry Collector configuration

exporters:
  splunk_hec/platform_logs:
    disable_compression: true
    endpoint: OMITTED
    index: OMITTED
    max_connections: 200
    profiling_data_enabled: false
    retry_on_failure:
      enabled: true
      initial_interval: 5s
      max_elapsed_time: 300s
      max_interval: 30s
    sending_queue:
      enabled: true
      num_consumers: 10
      queue_size: 5000
    source: kubernetes
    splunk_app_name: splunk-otel-collector
    splunk_app_version: 0.68.0
    timeout: 10s
    tls:
      insecure_skip_verify: false
    token: OMITTED
extensions:
      health_check: null
      memory_ballast:
        size_mib: OMITTED
processors:
  batch: null
  memory_limiter:
    check_interval: 2s
    limit_mib: OMITTED
  resource:
    attributes:
    - action: insert
      key: metric_source
      value: kubernetes
    - action: upsert
      key: k8s.cluster.name
      value: OMITTED
  resource/add_environment:
    attributes:
    - action: insert
      key: deployment.environment
      value: OMITTED
  resourcedetection:
    detectors:
    - env
    - eks
    - ec2
    - system
    override: true
    timeout: 10s
  transform/add_sourcetype:
    log_statements:
    - context: log
      statements:
      - set(resource.attributes["com.splunk.sourcetype"], Concat(["kube:object:",
        attributes["k8s.resource.name"]], ""))
receivers:
  k8sobjects:
    auth_type: serviceAccount
    objects:
    - group: events.k8s.io
      mode: watch
      name: events
service:
  extensions:
  - health_check
  - memory_ballast
  pipelines:
    logs/objects:
      exporters:
      - splunk_hec/platform_logs
      processors:
      - memory_limiter
      - batch
      - resourcedetection
      - resource
      - transform/add_sourcetype
      - resource/add_environment
      receivers:
      - k8sobjects

Log output

2023-01-27T10:12:45.167Z        warn    [email protected]/receiver.go:162      Watch channel closed unexpectedly       {"kind": "receiver", "name": "k8sobjects", "pipeline": "logs", "resource": "events.k8s.io/v1, Resource=events"}

Additional context

No response

[github-actions]

Pinging code owners:

• receiver/k8sobjects: @dmitryax @harshit-splunk

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[nicolastakashi]

Hey, @atoulme and @stokerjon I faced the same issue, and I'm available to help with that fix.

I'm not sure how to force reproduce that issue, do you know how we can simulate this behavior?

[atoulme]

No idea. Maybe create a mock k8s object api and restart it?

[stokerjon]

Setting a short timeoutSeconds in ListOptions for the watch should have the same affect.
The hang up isn't random its happening every 30 min, so its probably a default timeout in the api server that hasn't been accounted for.

[nicolastakashi]

I managed it!
Just adjusting the unit test before opening a PR, if you want you can assign this issue to me @stokerjon

[dmitryax]

Thanks @nicolastakashi