kubernetes_events input plugin cannot handle chunked transfer encoding

[syurevich]

Bug Report

Describe the bug

kubernetes_events input plugin cannot handle chunked transfer encoding, as required by HTTP/1.1.

With the "debug" logging level enabled in the Fluent Bit kubernetes_events input plugin, we observed lots of errors. See the logs below.

The ff4, ef7, f10, 134d, etc. are chunk sizes in hexadecimal as per HTTP/1.1 chunked transfer encoding. They are the chunked length headers used by servers (proxies/L7 load balancers) to send streaming HTTP responses when the total length is unknown in advance.

The kubernetes_events input plugin seems to be not designed to handle the chunked encoding markers, so it tries to parse ff4 as JSON, resulting in the "bad formed JSON" error (as printed by flb_plg_debug function).

Moreover, as soon as this error occurs, the watch stream gets stuck and the kubernetes_events input plugin does not receive new k8s events for about 25-40 minutes until it re-requests the watch.

To Reproduce

• Rubular link if applicable:
• Example log message if applicable:

[2025/12/03 10:11:01] [debug] [input:kubernetes_events:kubernetes_events.0] could not process payload, incomplete or bad formed JSON: ff4
{"type":"DELETED","object":{ ... }}
...
[2025/12/03 10:51:04] [debug] [input:kubernetes_events:kubernetes_events.0] could not process payload, incomplete or bad formed JSON: ef7
{"type":"DELETED","object":{ ... }}
...
[2025/12/03 11:36:05] [debug] [input:kubernetes_events:kubernetes_events.0] could not process payload, incomplete or bad formed JSON: f10
{"type":"DELETED","object":{ ... }}
...
[2025/12/03 12:31:04] [debug] [input:kubernetes_events:kubernetes_events.0] could not process payload, incomplete or bad formed JSON: 134d
{"type":"DELETED","object":{ ... }}
...

• Steps to reproduce the problem:

Expected behavior

kubernetes_events input plugin should handle chunked transfer encoding, as required by HTTP/1.1.

Screenshots

Your Environment

• Version used: fluent bit version=4.2.0
• Configuration:

    service:
      flush: 5
      log_level: "info"
      scheduler.cap: 60
      scheduler.base: 5
      daemon: false
      storage.metrics: true
      storage.max_chunks_up: 8
      storage.path: /fluent-bit/buffers
      storage.delete_irrecoverable_chunks: true
      storage.backlog.flush_on_shutdown: true
      grace: 29
      http_server: true
      http_listen: 0.0.0.0
      http_port: 2020

    pipeline:
      inputs:
        - name: kubernetes_events
          tag: k8s_events
          kube_token_ttl: 10s
          kube_request_limit: 100
          kube_retention_time: 86400
          log_level: "debug"
          storage.type: filesystem
...

• Environment name and version (e.g. Kubernetes? What version?): Kubernetes v1.32.9
• Server type and version:
• Operating System and version: gardenlinux 1592.14.0
• Filters and plugins: kubernetes_events

Additional context

The kubernetes_events input plugin stops receiving new k8s events.

[I501307]

We are also affected by this issue in our production Kubernetes clusters.

Impact:

• FluentBit kubernetes_events input plugin stops receiving new events for 25-40 minutes after encountering chunked transfer encoding markers
• This creates significant gaps in our event monitoring and alerting
• The issue occurs regularly when k8s API server or intermediate proxies use HTTP/1.1 chunked transfer encoding

Environment:

• FluentBit version: 4.2.0
• Kubernetes version: v1.32.9
• This affects multiple clusters in our infrastructure

This is blocking our ability to reliably monitor Kubernetes events. Could someone from the maintainer team please prioritize this issue? As this is a critical bug in infra.

We can provide additional logs or testing assistance if needed.

Thank you for your attention to this matter.