📝 document runtime sink state surface

2026-07-27 18:32:20 +00:00 · 2026-06-13 22:38:14 +08:00
parent a17f50db88
commit 1ebf31945c
4 changed files with 239 additions and 0 deletions
@@ -0,0 +1,76 @@
+---
+name: runtime-sink-dropped-count
+group: api
+category: runtime
+update-time: 20260613
+description: Read the cumulative dropped-record count from a RuntimeSink.
+key-word:
+    - runtime
+    - sink
+    - queue
+    - public
+---
+
+## Runtime-sink-dropped-count
+
+Read the cumulative dropped-record count from a `RuntimeSink`. This helper exposes the direct sink-level loss metric for queue-backed runtime variants.
+
+### Interface
+
+```moonbit
+pub fn RuntimeSink::dropped_count(self : RuntimeSink) -> Int {
+```
+
+#### input
+
+- `self : RuntimeSink` - Runtime sink whose dropped-record metric should be inspected.
+
+#### output
+
+- `Int` - Number of dropped records reported by the runtime sink.
+
+### Explanation
+
+Detailed rules explaining key parameters and behaviors
+
+- Queue-wrapped runtime variants forward to the wrapped queue sink's `dropped_count()` value.
+- Plain console and plain file runtime variants return `0` because they do not track queued record drops.
+- The counter is cumulative for the lifetime of the concrete runtime sink value.
+- This method is the direct sink-level API used by `ConfiguredLogger::dropped_count(...)`.
+
+### How to Use
+
+Here are some specific examples provided.
+
+#### When Need Direct Queue Loss Visibility
+
+When a runtime sink may discard records under pressure:
+```moonbit
+if sink.dropped_count() > 0 {
+  println("runtime sink dropped records")
+}
+```
+
+In this example, direct runtime queue loss becomes visible without inspecting inner fields manually.
+
+#### When Compare Queue Tuning Changes
+
+When queue overflow behavior should be checked operationally:
+```moonbit
+ignore(sink.dropped_count())
+```
+
+In this example, the helper exposes the metric needed to compare different queue settings.
+
+### Error Case
+
+e.g.:
+- If the runtime sink is not queue-backed, the method simply returns `0`.
+
+- If callers need the higher-level logger wrapper API, use `ConfiguredLogger::dropped_count()` instead.
+
+### Notes
+
+1. This helper reports cumulative loss, not the reason for that loss.
+
+2. Pair it with `pending_count()` and queue configuration when investigating pressure.