[extension/opampextension] Fix self-reported status events dropped by non-Reporter hosts

[CodeBlackwell]

Root cause

opampAgent.Start wires the extension's internal reportFunc (used to report the extension's own status, e.g. from monitorPPID) directly to componentstatus.ReportStatus(host, event).

componentstatus.ReportStatus is a no-op when the component.Host passed to Start does not implement componentstatus.Reporter. In that case, self-reported events from the opamp extension — most notably the fatal error monitorPPID emits when the collector's parent process disappears — were silently discarded instead of reaching the extension's own status aggregator. The collector could keep reporting healthy even though it had been orphaned.

The fix

• Added an extensionID component.ID field to opampAgent, populated from extension.Settings.ID in newOpampAgent.
• In Start, build selfInstanceID := componentstatus.NewInstanceID(o.extensionID, component.KindExtension) and route reportFunc through the extension's own ComponentStatusChanged handler (o.ComponentStatusChanged(selfInstanceID, event)) instead of componentstatus.ReportStatus(host, event).

ComponentStatusChanged always forwards events into the extension's internal status aggregator, so self-reported events now reach it regardless of whether the host implements componentstatus.Reporter.

Before / after

• Before: self-reported status events (fatal error on parent-process disappearance, etc.) were dropped whenever host didn't implement componentstatus.Reporter (e.g. componenttest.NewNopHost()), so the collector's health state never reflected them.
• After: self-reported events are delivered to the extension's own status aggregator unconditionally, independent of the host's capabilities.

Test added

TestReportFuncReachesStatusAggregatorRegardlessOfHostReporter (extension/opampextension/opamp_agent_test.go) starts the agent against componenttest.NewNopHost() — which intentionally does not implement componentstatus.Reporter — reports a fatal error via o.reportFunc, and asserts the mock status aggregator receives it with StatusFatalError and the original error.

Status

Fix was produced and verified by an automated CRG (code-review-graph) debug pass: discovery → adversarial verification → TDD fix (RED before the fix, GREEN after) → final gate. Final gate: clean.

Draft PR — opening for early feedback; not requesting review yet.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: CodeBlackwell / name: LeChristopher Blackwell (32cf38d, c799e70, 9427401, ebb884c, f2df5a3)

[github-actions]

Welcome, contributor! Thank you for your contribution to opentelemetry-collector-contrib.

Important reminders:

• Read our Contributing Guidelines.
• Sign the CLA if you haven't already.
• First-time contributors should have at most one PR not marked as draft until their first PR is merged.
• If your change isn't one of our priority components, reviews may take more time.
• Give reviewers at least a few days before pinging them for feedback.
• If you need help or struggle to move your PR forward:
  • Ask questions regarding general setup, development process, or contributor etiquette in #opentelemetry-new-contributors.
  • Raise technical or Collector-specific questions in #otel-collector-dev or a Collector SIG meeting.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: eb2668a22f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Route fatal self-reports through the host reporter

With the normal collector service host, host implements componentstatus.Reporter; before this change the self-report went through that reporter, whose status notification path also sends StatusFatalError to the service AsyncErrorChannel. Calling the OpAMP watcher directly only updates this extension's internal aggregator, so when ppid is configured and the parent disappears the OpAMP health may flip but the collector no longer receives the fatal event and keeps running orphaned (and with reports_health: false this send has no initialized componentStatusCh to receive it). Please preserve the host reporter path when available and add the internal fallback for non-reporter hosts.

Useful? React with 👍 / 👎.

[CodeBlackwell]

Thanks @avy252004 and @chatgpt-codex-connector — you're both right. The first version replaced componentstatus.ReportStatus(host, event) outright, which fixed the non-Reporter (NopHost) case but regressed production: with the real service host, ReportStatus also forwards a StatusFatalError to the service's AsyncErrorChannel, so bypassing it would leave an orphaned collector running instead of shutting down.

Fixed in 706ec46 with the if/else you suggested:

o.reportFunc = func(event *componentstatus.Event) {
    if _, ok := host.(componentstatus.Reporter); ok {
        componentstatus.ReportStatus(host, event)
        return
    }
    o.ComponentStatusChanged(selfInstanceID, event)
}

So the production path is preserved when the host implements componentstatus.Reporter, and the internal aggregator fallback only kicks in for non-Reporter hosts (where ReportStatus is a silent no-op). Added TestReportFuncPrefersHostReporter for the reporter path; the existing NopHost test still covers the fallback. Changelog subtext updated to match.

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[CodeBlackwell]

/easycla

[iblancasa]

I think this branch does not preserve the real service shutdown path in production. Can you check?

I think we always fallback to o.ComponentStatusChanged

[CodeBlackwell]

You're right, thanks for catching this. Traced it against collector core v0.156.0:

• The service starts extensions with the bare graph.Host. Only pipeline components get the HostWrapper that implements componentstatus.Reporter.
• So for an extension, host.(componentstatus.Reporter) is always false. We always hit the fallback, and the ReportStatus branch was dead code. The passing test only worked because it injected a synthetic Reporter host that production never supplies.
• Pushed a commit that drops the dead branch and always routes through ComponentStatusChanged, so the event reliably reaches the aggregator and the orphan gets reported unhealthy. Removed the misleading test and changelog wording.

Remaining gap: this makes the orphan reported unhealthy but not actually shut down, and an extension has no direct line to the service's fatal-error channel, so real shutdown-on-orphan looks like a bigger design question than this fix. Want me to open a separate follow-up issue for that and keep this PR scoped to the dropped status event?

[douglascamata]

@CodeBlackwell can you explain to us what's happening here more clearly, ideally with some examples and in your own words? What is the problem you observe? What is the proposed solution? This seems heavily AI generated, as seen by the co-author commit trailer and lack of use of our PR template and I'm struggling to understand it.

[CodeBlackwell]

I'm Happy to Explain, apologies if my previous explanations were unclear. Yes I do use Ai to format and speed up responses but I understand the human to human aspect too.

In shortest words,

• TThe README says the ppid setting's error emission policy is: when the watched parent process is no longer running, the extension emits a fatal error, causing the collector to exit. That is the stated guarantee.

• The current behavior in the case illustrated in this reproduction script does not follow said policy.

Thats all this is tending to. If its a red herring, My genuine apologies - While It was brought to my attention via AI, I did my best due diligence to verify that this was a legit deviation from stated guidelines and a valid PR.

Below is a reproduction script

#!/bin/bash
# Repro: opamp ppid watchdog should make an orphaned collector exit. It doesn't.
set -u
cd "$(dirname "$0")"
VER=0.156.0

# Get the collector binary if needed
if ! ./otelcol-contrib --version 2>/dev/null; then
    OS=$(uname | tr '[:upper:]' '[:lower:]')
    ARCH=$(uname -m); [ "$ARCH" = "x86_64" ] && ARCH=amd64
    echo "Downloading otelcol-contrib v${VER} (${OS}_${ARCH})..."
    curl -sL "https://github.com/open-telemetry/opentelemetry-collector-releases/releases/download/v${VER}/otelcol-contrib_${VER}_${OS}_${ARCH}.tar.gz" | tar xz otelcol-contrib
    ./otelcol-contrib --version || exit 1
fi

# Collector config: watch PARENT_PID, poll every 1s
cat > config.yaml <<'EOF'
extensions:
  opamp:
    server:
      ws:
        endpoint: wss://127.0.0.1:4320/v1/opamp
    ppid: ${env:PARENT_PID}
    ppid_poll_interval: 1s

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 127.0.0.1:4317

exporters:
  debug:

service:
  extensions: [opamp]
  pipelines:
    traces:
      receivers: [otlp]
      exporters: [debug]
EOF

# Fake parent process
sleep 600 &
export PARENT_PID=$!
disown
echo "parent pid: $PARENT_PID"

# Start collector watching the parent
./otelcol-contrib --config config.yaml > collector.log 2>&1 &
COLLECTOR_PID=$!
sleep 3
ps -p $COLLECTOR_PID > /dev/null || { echo "collector died on startup, see collector.log"; exit 1; }
echo "collector pid: $COLLECTOR_PID"

# Orphan the collector. Docs say: fatal error + exit within ~1s.
echo "killing parent..."
kill $PARENT_PID
sleep 5

# Verdict
if ps -p $COLLECTOR_PID > /dev/null; then
    echo "BUG: collector still running 5s after being orphaned"
else
    echo "collector exited as documented (bug not reproduced)"
fi
if grep -qi orphan collector.log; then
    grep -i orphan collector.log
else
    echo "BUG: 'orphaned' fatal error never logged"
fi

kill $COLLECTOR_PID 2>/dev/null

[CodeBlackwell]

Also I'm Having an issue with the easycla. I've signed it multiple times but it wont clear the error.

[douglascamata]

@CodeBlackwell thanks for the explanation. I think I understand this better now. I also like the reproduction script, that's helpful.

I don't think your solution works though. It will push that status, yes. But it won't ultimately cause the Collector to exit when the PPID is gone. Can you check this?

[douglascamata]

Yes I do use Ai to format and speed up responses but I understand the human to human aspect too.

@CodeBlackwell no problem using AI to tidy up text or to write the code. It's also fine to write "my AI said (...)" or add a small note saying "I used AI to format the text, but the contents itself are all written by" in the PR. It's just about disclosure. 👍

[douglascamata]

Regarding EasyCLA, I don't know how to handle this. I'm not even sure how to debug it. We can try to get some help from maintainers that have more experience with it.

[opentelemetry-pr-dashboard]

Pull request dashboard status

Status last refreshed: 2026-07-27 15:23:04 UTC.

• Waiting on: Author
• Next step: Address or respond to 2 review feedback items:
  • Inline threads: 1
  • Top-level feedback: 2
  • For each item, reply to move the discussion forward, e.g. link to the commit that addresses it, explain why no change is needed, or ask a follow-up question.

This automated status or its linked feedback items may be incorrect. If something looks wrong, please report it with the result you expected. If you believe this pull request is incorrectly routed as waiting on the author, comment /dashboard route:reviewers to route it from waiting on the author to waiting on reviewers. If the last refreshed time above predates your latest reply or push, the dashboard hasn't processed it yet.

[CodeBlackwell]

You are correct, it doesn't exit. I dug through collector core v0.156.0 to check and the exit path just isn't reachable from an extension.

In shortest words:

• The service starts extensions with the bare graph.Host, which doesn't implement componentstatus.Reporter - so ReportStatus from an extension is a silent no-op.
• The machinery that actually shuts the collector down (fatal event -> AsyncErrorChannel) only hangs off HostWrapper (internal/graph/graph.go:445), and only pipeline components ever get that wrapper. Extensions never see it.

So making the collector truly exit needs a fix in core, not contrib. Best this PR can do is what it does now - the orphan shows up as unhealthy instead of the event being silently dropped.

I suppose A PR Name change?

I solved the CLA problem.

[douglascamata]

So making the collector truly exit needs a fix in core, not contrib. Best this PR can do is what it does now - the orphan shows up as unhealthy instead of the event being silently dropped.

Sounds good, @CodeBlackwell. We can move forward with this fix and open an issue in core to fix the wrapping. Then when core is fixed this logic will behave as expected. 👍

I suppose A PR Name change?

Personally I think the PR name is fine now, but feel free to change it if you have better ones in mind.

[douglascamata]

/workflow-approve

[douglascamata]

@CodeBlackwell please keep in mind we have a pull request template to be followed.