| 1 | import * as NodeServices from "@effect/platform-node/NodeServices"; |
| 2 | import { assert, describe, it } from "@effect/vitest"; |
| 3 | import * as Effect from "effect/Effect"; |
| 4 | import * as FileSystem from "effect/FileSystem"; |
| 5 | import * as Layer from "effect/Layer"; |
| 6 | import * as Logger from "effect/Logger"; |
| 7 | import * as Tracer from "effect/Tracer"; |
| 8 | import * as TestClock from "effect/testing/TestClock"; |
| 9 | |
| 10 | import * as EventLoopMonitor from "./EventLoopMonitor.ts"; |
| 11 | |
| 12 | const ms = (value: number) => value * 1e6; |
| 13 | |
| 14 | // Node reports a stall of S as a gap of up to S + 1 s, the histogram resolution. |
| 15 | const stalled: EventLoopMonitor.EventLoopReadings = { |
| 16 | delayMaxNs: ms(5_950), |
| 17 | activeMs: 6_200, |
| 18 | utilization: 0.176, |
| 19 | usage: { |
| 20 | userCPUTime: 310_400, |
| 21 | systemCPUTime: 95_600, |
| 22 | majorPageFault: 8_412, |
| 23 | minorPageFault: 20_031, |
| 24 | involuntaryContextSwitches: 57, |
| 25 | }, |
| 26 | rssBytes: 1536 * 1024 * 1024, |
| 27 | }; |
| 28 | // Over the threshold as read, but not once the resolution is subtracted. |
| 29 | const quiet: EventLoopMonitor.EventLoopReadings = { ...stalled, delayMaxNs: ms(2_950) }; |
| 30 | |
| 31 | // The first sample covers startup, so the monitor discards it. |
| 32 | const monitor = ( |
| 33 | samples: Array<EventLoopMonitor.EventLoopReadings>, |
| 34 | stallLog: EventLoopMonitor.StallLogOptions = { filePath: "\0", maxBytes: 1 }, |
| 35 | ) => |
| 36 | Layer.build( |
| 37 | EventLoopMonitor.layerWith( |
| 38 | Effect.succeed(Effect.sync(() => samples.shift() ?? quiet)), |
| 39 | stallLog, |
| 40 | ), |
| 41 | ); |
| 42 | |
| 43 | describe("EventLoopMonitor", () => { |
| 44 | it.effect("records a warning span only for samples that saw a stall", () => |
| 45 | Effect.gen(function* () { |
| 46 | const spans: Array<Tracer.NativeSpan> = []; |
| 47 | const tracer = Tracer.make({ |
| 48 | span: (options) => { |
| 49 | const span = new Tracer.NativeSpan(options); |
| 50 | spans.push(span); |
| 51 | return span; |
| 52 | }, |
| 53 | }); |
| 54 | yield* Effect.gen(function* () { |
| 55 | // The default stall log path cannot open, and the span must still be recorded. |
| 56 | yield* monitor([stalled, quiet, stalled]); |
| 57 | yield* TestClock.adjust("60 seconds"); |
| 58 | assert.lengthOf(spans, 0); |
| 59 | yield* TestClock.adjust("30 seconds"); |
| 60 | }).pipe(Effect.scoped, Effect.withTracer(tracer)); |
| 61 | |
| 62 | assert.deepStrictEqual( |
| 63 | spans.map((span) => span.name), |
| 64 | ["server.eventLoop.stall"], |
| 65 | ); |
| 66 | const [span] = spans; |
| 67 | assert.deepStrictEqual(Object.fromEntries(span!.attributes), { |
| 68 | delayMaxMs: 4_950, |
| 69 | utilization: 0.18, |
| 70 | cpuUserMs: 310, |
| 71 | cpuSystemMs: 96, |
| 72 | majorPageFaults: 8_412, |
| 73 | minorPageFaults: 20_031, |
| 74 | involuntaryContextSwitches: 57, |
| 75 | rssMb: 1536, |
| 76 | }); |
| 77 | assert.deepStrictEqual( |
| 78 | span!.events.map(([name, , attributes]) => [name, attributes["effect.logLevel"]]), |
| 79 | [["event loop stalled for 4950 ms", "WARN"]], |
| 80 | ); |
| 81 | }), |
| 82 | ); |
| 83 | |
| 84 | it.effect("keeps each stall as one line in a capped stall log", () => |
| 85 | Effect.gen(function* () { |
| 86 | const fs = yield* FileSystem.FileSystem; |
| 87 | const directory = yield* fs.makeTempDirectoryScoped({ prefix: "t3-event-loop-" }); |
| 88 | const filePath = `${directory}/event-loop-stalls.ndjson`; |
| 89 | // Room for one line, so each new stall rotates the last one into the backup. |
| 90 | yield* monitor([stalled, stalled, stalled, { ...stalled, delayMaxNs: ms(7_000) }], { |
| 91 | filePath, |
| 92 | maxBytes: 300, |
| 93 | }); |
| 94 | yield* TestClock.adjust("120 seconds"); |
| 95 | |
| 96 | const lines = (yield* fs.readFileString(filePath)).trim().split("\n"); |
| 97 | assert.deepStrictEqual( |
| 98 | lines.map((line) => JSON.parse(line)), |
| 99 | [ |
| 100 | { |
| 101 | time: "1970-01-01T00:02:00.000Z", |
| 102 | delayMaxMs: 6_000, |
| 103 | utilization: 0.18, |
| 104 | cpuUserMs: 310, |
| 105 | cpuSystemMs: 96, |
| 106 | majorPageFaults: 8_412, |
| 107 | minorPageFaults: 20_031, |
| 108 | involuntaryContextSwitches: 57, |
| 109 | rssMb: 1536, |
| 110 | }, |
| 111 | ], |
| 112 | ); |
| 113 | assert.lengthOf((yield* fs.readFileString(`${filePath}.1`)).trim().split("\n"), 1); |
| 114 | assert.isFalse(yield* fs.exists(`${filePath}.2`)); |
| 115 | }).pipe(Effect.scoped, Effect.provide(NodeServices.layer)), |
| 116 | ); |
| 117 | |
| 118 | it.effect("warns when a stall cannot be written to the stall log", () => { |
| 119 | const warnings: Array<unknown> = []; |
| 120 | const logger = Logger.make(({ message }) => { |
| 121 | warnings.push(Array.isArray(message) ? message[0] : message); |
| 122 | }); |
| 123 | return Effect.gen(function* () { |
| 124 | const fs = yield* FileSystem.FileSystem; |
| 125 | // A directory at the log path opens, but every append to it fails. The cap |
| 126 | // is large so the sink never rotates the directory out of the way. |
| 127 | const filePath = yield* fs.makeTempDirectoryScoped({ prefix: "t3-event-loop-" }); |
| 128 | yield* monitor([stalled, stalled], { filePath, maxBytes: 1024 * 1024 }); |
| 129 | yield* TestClock.adjust("60 seconds"); |
| 130 | assert.deepStrictEqual(warnings, [ |
| 131 | "event loop stalled for 4950 ms", |
| 132 | "event loop stall log write failed", |
| 133 | ]); |
| 134 | }).pipe( |
| 135 | Effect.scoped, |
| 136 | Effect.provide( |
| 137 | Layer.merge(NodeServices.layer, Logger.layer([logger], { mergeWithExisting: false })), |
| 138 | ), |
| 139 | ); |
| 140 | }); |
| 141 | |
| 142 | it("ignores delay the loop spent idle, such as a system sleep", () => { |
| 143 | // Waking from sleep reads as a long gap, but the loop was idle in poll for it. |
| 144 | const asleep: EventLoopMonitor.EventLoopReadings = { |
| 145 | ...stalled, |
| 146 | delayMaxNs: ms(600_000), |
| 147 | activeMs: 900, |
| 148 | }; |
| 149 | assert.isUndefined(EventLoopMonitor.stallMs(asleep)); |
| 150 | assert.strictEqual(EventLoopMonitor.stallMs({ ...asleep, activeMs: 600_000 }), 599_000); |
| 151 | }); |
| 152 | }); |