| author |  Julius Marminge <julius0216@outlook.com> 2026-09-05 01:47:36-07:00 |
| committer |  GitHub <noreply@github.com> 2026-09-05 01:47:36-07:00 |
| log | 45f5a5ffb3257d510f249b14c36a514beff2868a |
| tree | 2de38440e26c06a44a000d2b24879e207c7cb926 |
| parent | 4631000f5a7666c88402ce11a9ebb8cdef7a7dad |
| signature | Signed by PGP key B5690EEEBB952194 |
refactor(server): simplify native telemetry error internals (#10057)
2 files changed, 3 insertions(+), 49 deletions(-)
apps
server
src
resourceTelemetry
apps/server/src/resourceTelemetry/NativeTelemetryClient.test.ts-42
| ... | ... | @@ -1,6 +1,5 @@ |
| 1 | 1 | import type { HostPowerSnapshot } from "@t3tools/contracts"; |
| 2 | 2 | import { describe, expect, it } from "@effect/vitest"; |
| 3 | import * as Cause from "effect/Cause"; | |
| 4 | 3 | import * as DateTime from "effect/DateTime"; |
| 5 | 4 | import * as Deferred from "effect/Deferred"; |
| 6 | 5 | import * as Effect from "effect/Effect"; |
| ... | ... | @@ -9,12 +8,9 @@ import * as Ref from "effect/Ref"; |
| 9 | 8 | import * as Semaphore from "effect/Semaphore"; |
| 10 | 9 | |
| 11 | 10 | import { |
| 12 | NativeTelemetryRequestTimedOut, | |
| 13 | NativeTelemetryStreamClosed, | |
| 14 | 11 | canCommandNativeTelemetrySidecar, |
| 15 | 12 | canRequestNativeTelemetryRetry, |
| 16 | 13 | commitCollectionControlUpdate, |
| 17 | nativeTelemetrySupervisorFailureMessage, | |
| 18 | 14 | retainRecentNativeTelemetryFailures, |
| 19 | 15 | resolveNativeSampleIntervalMs, |
| 20 | 16 | synchronizeCollectionControlOnStart, |
| ... | ... | @@ -82,44 +78,6 @@ describe("canCommandNativeTelemetrySidecar", () => { |
| 82 | 78 | }); |
| 83 | 79 | }); |
| 84 | 80 | |
| 85 | describe("NativeTelemetryRequestTimedOut", () => { | |
| 86 | it("models history and sample request deadlines without a fabricated cause", () => { | |
| 87 | const historyTimeout = new NativeTelemetryRequestTimedOut({ | |
| 88 | operation: "readHistory", | |
| 89 | timeoutMs: 15_000, | |
| 90 | }); | |
| 91 | const sampleTimeout = new NativeTelemetryRequestTimedOut({ | |
| 92 | operation: "sampleNow", | |
| 93 | timeoutMs: 5_000, | |
| 94 | }); | |
| 95 | ||
| 96 | expect(historyTimeout.message).toBe( | |
| 97 | "Resource monitor 'readHistory' request timed out after 15000ms.", | |
| 98 | ); | |
| 99 | expect(sampleTimeout.message).toBe( | |
| 100 | "Resource monitor 'sampleNow' request timed out after 5000ms.", | |
| 101 | ); | |
| 102 | expect("cause" in historyTimeout).toBe(false); | |
| 103 | expect("cause" in sampleTimeout).toBe(false); | |
| 104 | }); | |
| 105 | }); | |
| 106 | ||
| 107 | describe("native telemetry supervisor failures", () => { | |
| 108 | it("distinguishes a closed event stream from a process exit", () => { | |
| 109 | expect(new NativeTelemetryStreamClosed().message).toBe( | |
| 110 | "Resource monitor event stream closed unexpectedly.", | |
| 111 | ); | |
| 112 | }); | |
| 113 | ||
| 114 | it("keeps defect details out of the caller-visible health message", () => { | |
| 115 | const secret = "credential=do-not-expose"; | |
| 116 | const message = nativeTelemetrySupervisorFailureMessage(Cause.die(new Error(secret))); | |
| 117 | ||
| 118 | expect(message).toBe("Resource monitor supervisor stopped unexpectedly."); | |
| 119 | expect(message).not.toContain(secret); | |
| 120 | }); | |
| 121 | }); | |
| 122 | ||
| 123 | 81 | describe("retainRecentNativeTelemetryFailures", () => { |
| 124 | 82 | it("expires old failures so an isolated crash restarts from the initial backoff", () => { |
| 125 | 83 | expect(retainRecentNativeTelemetryFailures([0, 30_000], 90_001)).toEqual([]); |
apps/server/src/resourceTelemetry/NativeTelemetryClient.ts+3-7
| ... | ... | @@ -73,7 +73,7 @@ export class NativeTelemetryHandshakeTimedOut extends Schema.TaggedErrorClass<Na |
| 73 | 73 | } |
| 74 | 74 | } |
| 75 | 75 | |
| 76 | export class NativeTelemetryRequestTimedOut extends Schema.TaggedErrorClass<NativeTelemetryRequestTimedOut>()( | |
| 76 | class NativeTelemetryRequestTimedOut extends Schema.TaggedErrorClass<NativeTelemetryRequestTimedOut>()( | |
| 77 | 77 | "NativeTelemetryRequestTimedOut", |
| 78 | 78 | { |
| 79 | 79 | operation: Schema.Literals(["readHistory", "sampleNow"]), |
| ... | ... | @@ -131,7 +131,7 @@ export class NativeTelemetryExited extends Schema.TaggedErrorClass<NativeTelemet |
| 131 | 131 | } |
| 132 | 132 | } |
| 133 | 133 | |
| 134 | export class NativeTelemetryStreamClosed extends Schema.TaggedErrorClass<NativeTelemetryStreamClosed>()( | |
| 134 | class NativeTelemetryStreamClosed extends Schema.TaggedErrorClass<NativeTelemetryStreamClosed>()( | |
| 135 | 135 | "NativeTelemetryStreamClosed", |
| 136 | 136 | {}, |
| 137 | 137 | ) { |
| ... | ... | @@ -340,10 +340,6 @@ function errorMessage(error: NativeTelemetryClientError): string { |
| 340 | 340 | return error.message; |
| 341 | 341 | } |
| 342 | 342 | |
| 343 | export function nativeTelemetrySupervisorFailureMessage(_cause: Cause.Cause<unknown>): string { | |
| 344 | return "Resource monitor supervisor stopped unexpectedly."; | |
| 345 | } | |
| 346 | ||
| 347 | 343 | export function canRequestNativeTelemetryRetry( |
| 348 | 344 | status: ResourceTelemetrySourceStatus, |
| 349 | 345 | hasHandle: boolean, |
| ... | ... | @@ -734,7 +730,7 @@ export const make = Effect.fn("resourceTelemetry.nativeTelemetryClient.make")(fu |
| 734 | 730 | ...current, |
| 735 | 731 | status: "unavailable" as const, |
| 736 | 732 | hello: Option.none(), |
| 737 | lastError: Option.some(nativeTelemetrySupervisorFailureMessage(cause)), | |
| 733 | lastError: Option.some("Resource monitor supervisor stopped unexpectedly."), | |
| 738 | 734 | })).pipe( |
| 739 | 735 | Effect.andThen(publishHealth), |
| 740 | 736 | Effect.andThen( |