http_server: built-in HTTP server deadlocks when exec input curls Fluent Bit internal endpoints

[jackfletch]

Bug Report

Describe the bug
In Fluent Bit v5.0.x, the built-in HTTP server (port 2020) deadlocks when any pipeline component makes an HTTP request to it from within the main event loop.

To Reproduce
Minimal config:

[SERVICE]
    HTTP_Server  On
    HTTP_Listen  0.0.0.0
    HTTP_Port    2020
    Flush        1

[INPUT]
    Name         exec
    Tag          test
    Command      curl -s http://127.0.0.1:2020/api/v1/metrics/prometheus
    Interval_Sec 30

[OUTPUT]
    Name  null
    Match *

1. Start Fluent Bit, confirm the health check works:

curl http://127.0.0.1:2020/api/v1/uptime   # works fine

2. Wait 30s for exec to fire, then try again:

curl http://127.0.0.1:2020/api/v1/uptime   # hangs forever

This works fine on v1.9.x and v4.2.x, but breaks on every v5.0.x release I've tested (v5.0.0 through v5.0.4).

Expected behavior
The built-in HTTP server should remain responsive while the in_exec plugin is running, as it did in v1.9.x and v4.2.x.

Screenshots
N/A

Your Environment

• Version used: broken in v5.0.0 through v5.0.4, works in v1.9.10 and v4.2.2
• Configuration: see minimal config above
• Environment name and version: ECS Fargate (ARM64), also reproduced locally on macOS
• Server type and version: N/A
• Operating System and version: Amazon Linux 2023 (Fargate), macOS 15 (local)
• Filters and plugins: in_exec, out_null (minimal repro)

Additional context
My use case is emitting Fluent Bit internal metrics out as CloudWatch metrics. The flow is (1) scrape Fluent Bit metrics endpoint with in_exec, (2) transform metrics into CloudWatch embedded metric format via filter_lua, and then (3) emit to CloudWatch as logs via out_cloudwatch_logs, which CloudWatch automatically converts to metrics.

[jackfletch]

I dug into this a bit and am working on a PR.