Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import { describe, expect, it } from "vitest";
import { GfsMetrics } from "./metrics.js";
/**
* P5-S02 — gfsc SLIs. /metrics exposes writer availability, p99 read/write
* latency, mount failures, read-after-write lag, quota usage, cache hit rate.
*/
describe("GfsMetrics", () => {
it("computes p99 latency from observations", () => {
const m = new GfsMetrics();
for (let i = 1; i <= 100; i++) m.recordRead(i);
expect(m.snapshot().readLatencyP99Ms).toBe(99);
});
it("tracks mount failures, cache hit rate, and writer availability", () => {
const m = new GfsMetrics();
m.recordMountFailure();
m.recordMountFailure();
m.recordCache(true);
m.recordCache(true);
m.recordCache(false); // 2/3 hit rate
m.setWriterAvailable(false);
const s = m.snapshot();
expect(s.mountFailures).toBe(2);
expect(s.cacheHitRate).toBeCloseTo(2 / 3, 5);
expect(s.writerAvailable).toBe(0);
});
it("renders every SLI in Prometheus text", () => {
const m = new GfsMetrics();
m.recordWrite(10);
m.setQuotaUsageRatio(0.5);
m.setReadAfterWriteLagMs(7);
const text = m.render();
for (const metric of [
"gfs_writer_available",
"gfs_read_latency_p99_ms",
"gfs_write_latency_p99_ms",
"gfs_mount_failures_total",
"gfs_read_after_write_lag_ms",
"gfs_quota_usage_ratio",
"gfs_cache_hit_rate",
"gfs_blob_orphan_candidates",
"gfs_blob_orphan_bytes",
"gfs_blob_cleanup_failures_total",
]) {
expect(text).toContain(metric);
}
expect(text).toContain("gfs_quota_usage_ratio 0.5");
expect(text).toContain("gfs_read_after_write_lag_ms 7");
});
it("tracks conservative blob reconciliation candidates and failures", () => {
const m = new GfsMetrics();
m.setOrphanCandidates(3, 4096);
m.recordBlobCleanupFailure();
expect(m.snapshot()).toMatchObject({
orphanCandidates: 3,
orphanBytes: 4096,
blobCleanupFailures: 1,
});
});
it("is zero-valued before any observation (empty p99 = 0, no division by zero)", () => {
const s = new GfsMetrics().snapshot();
expect(s.readLatencyP99Ms).toBe(0);
expect(s.cacheHitRate).toBe(0);
expect(s.writerAvailable).toBe(1); // available by default until marked down
});
});