Intermittent metric registration error: worker_millis_per_event

**Problem**:

Intermittent warning during pipeline startup. The worker_millis_per_event metric fails to register but pipeline functions normally.

```[org.logstash.execution.AbstractPipelineExt] Metric registration error: `worker_millis_per_event` could not be registered in namespace `[:stats, :pipelines, :logstash, :plugins, :filters, :c13c16377fd9a15e6e691284665f2da3494b42273b1fc6e37b6330fff12eb5ab, :flow]```

**Expected versus actual behavior**:

Expected: All flow metrics register successfully; /_node/stats/pipelines returns full pipeline statistics.

Actual: Registration fails. Pipelines start and process events normally (status: "green", Pipelines running {count: 2}). However, /_node/stats/pipelines returns empty pipeline data: {"pipelines": {}}.

**Environment Details**:

1. Logstash version - 9.0.3 OSS Version
2. Logstash installation source (e.g. built from source, with a package manager: DEB/RPM, expanded from tar or zip archive, docker) - Built from source
3. How is Logstash being run (e.g. as a service/service manager: systemd, upstart, etc. Via command line, docker/kubernetes) - Kubernetes (running as a process with logstash binary in Docker container)

**JVM**:

bash-4.4$ java -version
openjdk version "21.0.12" 2026-07-21
OpenJDK Runtime Environment (build 21.0.12+8-suse-01500-x8664)
OpenJDK 64-Bit Server VM (build 21.0.12+8-suse-01500-x8664, mixed mode, sharing)

**OS version**:

`Linux log-transformer-597b585774 6.4.0-150600.23.42-default #1 SMP PREEMPT_DYNAMIC Fri Mar 7 09:53:00 UTC 2025 (7bf6ecd) x86_64 x86_64 x86_64 GNU/Linux`

**Steps to reproduce**:

Not consistently reproducible - Occurs intermittently during startup. The reported error is observed approximately once in every 40-50 startups. Reproduced with automated script that deploys logstash, searches for error pattern during startup, redeploys if error not found. This continues till error pattern is found.

**Additional Observations**:

1. No functional impact - Pipeline processes events normally
2. Metrics not available - Pipeline stats are empty from endpoint localhost:9600/_node/stats/pipelines
{
"host": "log-transformer",
"version": "9.0.3",
"http_address": "0.0.0.0:9600",
"id": "5fbd8fxxxxxxxxxxxx5602",
"name": "log-transformer",
"ephemeral_id": "36axxxxxxxxxxxxxxxxx2348a7",
"snapshot": false,
"status": "unknown",
"pipeline": {
"workers": 2,
"batch_size": 2048,
"batch_delay": 50
},
"pipelines": {}
}

3. Self-resolving: Logstash process restart with same configuration succeeds without error. No warning message and stats are also available in /_node/stats/pipelines endpoint.

Note - We have created a discussion in elastic community (https://discuss.elastic.co/t/intermittent-metric-registration-error-worker-millis-per-event/383836) and tried the workaround/suggestions provided there. But unfortunately, it didn't solve our issue.

**Logs**:

Full startup logs attached - [Logstash Logs - Metrics registration error · GitHub](https://gist.github.com/Mariyabanatic/70f79f24f3cb10eff92bbc24829ce12e)

Start with the attached startup logs and the automated redeploy loop described in the issue, then inspect the metric-registration path for worker_millis_per_event and the /_node/stats/pipelines endpoint. Done means the intermittent registration warning is eliminated and the endpoint consistently returns populated pipeline statistics across repeated startups.

Written by the indexing model from the issue text.