langgenius / langgenius/dify-plugin-daemon
[Feature]: expose Prometheus /metrics — concurrent invokes, plugin process count, invoke latency
Nobody has claimed this yet.
- Dominant language
- Go
- Stars
- 459
- Forks
- 313
- Avg merge
- 5h 17m
- Merged PRs (30d)
- 5
Description
Self Checks
I have searched existing issues/discussions — no feature request covers runtime metrics exposure for the plugin daemon.
Background
dify-plugin-daemon carries all LLM and tool invocations in self-hosted Dify (1.0+), yet exposes zero runtime metrics. Under high-concurrency workflow runs it is impossible to answer:
- Is throughput limited by the daemon itself, a specific plugin, or the upstream model API?
- How many invokes are queued vs executing, per plugin?
- Which plugin is leaking processes or erroring silently?
We ended up correlating api-side outbound-latency alerts (which have no target-host dimension) with daemon pod restarts — a very lossy attribution path. A capacity incident is only confirmed after user-visible failures.
Requested metrics (Prometheus text format, /metrics endpoint)
| Metric | Type | Labels | Purpose |
|---|---|---|---|
plugin_daemon_invokes_in_flight |
gauge | plugin | queued/executing invokes per plugin — the saturation signal |
plugin_daemon_invoke_duration_seconds |
histogram | plugin, outcome | latency distribution per plugin |
plugin_daemon_plugin_processes |
gauge | plugin | running plugin subprocesses (detect leaks) |
plugin_daemon_invoke_errors_total |
counter | plugin, error_class | failures by class (timeout, crash, plugin error) |
Standard Go runtime metrics (goroutines, GC, FDs) and process metrics would be a welcome bonus.
Acceptance
A /metrics endpoint (server port or dedicated), standard Prometheus text format, scrape-friendly.
Why it matters
The daemon is the LLM-calling choke point of every self-hosted deployment. With per-plugin in-flight gauges and latency histograms, bottleneck attribution becomes a dashboard query instead of a forensic exercise.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the daemon's HTTP server or server-port entry point and the code handling invokes and plugin subprocesses. Determine where queued, executing, completed, failed, and process-count events can be observed, then verify that a scrape of /metrics uses standard Prometheus text format and exposes the requested labels and metric types.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- go, prometheus
- Domain
- api, observability
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100