Skip to content
199 linesCodeBlameRaw

Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.

Merge remote-tracking branch 'origin/main' into workspace-chat1// The runner's Durable Object error report, from answers shaped as
2// Cloudflare gives them, without the network.
3
4import assert from "node:assert/strict";
5import { test } from "node:test";
6
7import {
8 DO_QUERIES,
9 KEYS,
10 SERVICE_KEYS,
11 bucketsOf,
12 byStatus,
13 classify,
14 doQuery,
15 format,
16 parseArgs,
17 parseCalculations,
18 parseDoGroups,
19 parseEvents,
20 telemetryQuestions,
21 windowOf,
22} from "./runner-errors.mjs";
23
24const answer = (rows) => ({ data: { viewer: { accounts: [{ rows }] } } });
25const byNamespace = DO_QUERIES.find((query) => query.key === "by_namespace").variants[1];
26
27test("the command line has defaults and bounds", () => {
28 assert.deepEqual(parseArgs([], {}), { help: false, json: false, keys: false, days: 7, samples: 10, script: "g1t-runner", account: "1e6f2cffa3f445920836e8ebe446bb58" });
29 const options = parseArgs(["--days", "90", "--samples", "3", "--script", "other", "--json", "--keys"], { CLOUDFLARE_ACCOUNT_ID: "acct" });
30 assert.equal(options.days, 31);
31 assert.equal(options.samples, 3);
32 assert.equal(options.script, "other");
33 assert.equal(options.account, "acct");
34 assert.ok(options.json && options.keys);
35 assert.equal(parseArgs(["--days", "soon"], {}).days, 7);
36});
37
38test("the window is the last N days", () => {
39 assert.deepEqual(windowOf(new Date("2026-10-08T12:00:00Z"), 7), { start: "2026-10-01T12:00:00.000Z", end: "2026-10-08T12:00:00.000Z" });
40});
41
42test("the query is for one script, with the variant's sums and dimensions", () => {
43 const query = doQuery(byNamespace);
44 assert.match(query, /durableObjectsInvocationsAdaptiveGroups\(limit: 10000, filter: \{ datetime_geq: \$start, datetime_leq: \$end, scriptName: \$script \}\)/);
45 assert.match(query, /sum \{ requests errors \}/);
46 assert.match(query, /dimensions \{ namespaceId status \}/);
47 assert.match(query, /\$script: String!/);
48});
49
50test("rows are named, summed per status and sorted by requests", () => {
51 const body = answer([
52 { sum: { requests: 900, errors: 300 }, dimensions: { namespaceId: "ns1", status: "scriptThrewException" } },
53 { sum: { requests: 1000, errors: 0 }, dimensions: { namespaceId: "ns1", status: "success" } },
54 { sum: { requests: 40, errors: 20 }, dimensions: { namespaceId: "ns2", status: "internalError" } },
55 { sum: { requests: 10, errors: 0 }, dimensions: { namespaceId: "ns2", status: "success" } },
56 ]);
57 const parsed = parseDoGroups(body, byNamespace, { ns1: "g1t-runner_AttemptSandbox" });
58 assert.deepEqual(parsed.dims, ["namespace_id", "status"]);
59 assert.deepEqual(parsed.rows[0], { namespace_id: "g1t-runner_AttemptSandbox", status: "success", requests: 1000, errors: 0 });
60 assert.equal(parsed.rows.at(-1).namespace_id, "ns2");
61 assert.deepEqual(parsed.totals, { requests: 1950, errors: 320 });
62 assert.deepEqual(byStatus(parsed), [
63 { status: "success", requests: 1010, errors: 0 },
64 { status: "scriptThrewException", requests: 900, errors: 300 },
65 { status: "internalError", requests: 40, errors: 20 },
66 ]);
67});
68
69test("Cloudflare's refusal becomes the error, so the next variant is tried", () => {
70 assert.throws(() => parseDoGroups({ errors: [{ message: "unknown field wallTime" }] }, byNamespace), /unknown field wallTime/);
71 assert.throws(() => parseDoGroups({ data: {} }, byNamespace), /no rows/);
72});
73
74test("the telemetry questions filter on the script and on failures", () => {
75 const questions = telemetryQuestions({ script: "g1t-runner", from: 1, to: 2, samples: 5 });
76 assert.deepEqual(questions.map((q) => q.key), ["failed_invocations", "exceptions", "error_logs", "samples"]);
77 const failed = questions[0].body;
78 assert.equal(failed.view, "calculations");
79 assert.deepEqual(failed.timeframe, { from: 1, to: 2 });
80 assert.deepEqual(failed.parameters.filters, [
81 { key: SERVICE_KEYS[0], operation: "eq", type: "string", value: "g1t-runner" },
82 { key: KEYS.outcome, operation: "neq", type: "string", value: "ok" },
83 ]);
84 assert.deepEqual(failed.parameters.groupBys.map((g) => g.value), [KEYS.entrypoint, KEYS.eventType, KEYS.outcome]);
85 assert.equal(questions[3].body.view, "events");
86 assert.equal(questions[3].body.limit, 5);
87 const other = telemetryQuestions({ script: "g1t-runner", from: 1, to: 2, samples: 5, serviceKey: "$metadata.service" });
88 assert.equal(other[1].body.parameters.filters[0].key, "$metadata.service");
89});
90
91test("a calculations answer becomes rows with counts, largest first", () => {
92 const body = {
93 success: true,
94 result: {
95 calculations: [
96 {
97 aggregates: [
98 { groups: [{ key: KEYS.entrypoint, value: "AttemptSandbox" }, { key: KEYS.eventType, value: "rpc" }], value: 12 },
99 { groups: [{ key: KEYS.entrypoint, value: "AttemptSandbox" }, { key: KEYS.eventType, value: "alarm" }], value: 540 },
100 { groups: [{ key: KEYS.entrypoint, value: "" }, { key: KEYS.eventType, value: "alarm" }], count: 3 },
101 ],
102 },
103 ],
104 },
105 };
106 assert.deepEqual(parseCalculations(body), [
107 { [KEYS.entrypoint]: "AttemptSandbox", [KEYS.eventType]: "alarm", count: 540 },
108 { [KEYS.entrypoint]: "AttemptSandbox", [KEYS.eventType]: "rpc", count: 12 },
109 { [KEYS.entrypoint]: "(none)", [KEYS.eventType]: "alarm", count: 3 },
110 ]);
111 assert.throws(() => parseCalculations({ success: false, errors: [{ message: "Unauthorized" }] }), /Unauthorized/);
112 assert.throws(() => parseCalculations({ success: true, result: {} }), /no calculations/);
113});
114
115test("an events answer becomes plain records, the object id cut short", () => {
116 const body = {
117 success: true,
118 result: {
119 events: {
120 events: [
121 {
122 timestamp: Date.UTC(2026, 9, 7, 3, 4, 5),
123 $workers: { entrypoint: "AttemptSandbox", eventType: "alarm", outcome: "exception", durableObjectId: "0123456789abcdef0123" },
124 $metadata: { error: "Durable Object reset because its code was updated." },
125 },
126 { $workers: { eventType: "rpc", outcome: "canceled" }, source: { message: "sandbox not started" } },
127 ],
128 },
129 },
130 };
131 assert.deepEqual(parseEvents(body), [
132 { at: "2026-10-07T03:04:05.000Z", entrypoint: "AttemptSandbox", event_type: "alarm", outcome: "exception", object: "0123456789ab", error: "Durable Object reset because its code was updated.", message: null },
133 { at: null, entrypoint: null, event_type: "rpc", outcome: "canceled", object: null, error: null, message: "sandbox not started" },
134 ]);
135});
136
137test("messages fall into buckets, expected or not", () => {
138 assert.equal(classify("Durable Object reset because its code was updated.").bucket, "deploy_reset");
139 assert.equal(classify("Durable Object reset because its code was updated.").expected, true);
140 assert.equal(classify("sandbox stop not reported checks report_checks failed with status 500").bucket, "stop_not_reported");
141 assert.equal(classify("sandbox alarm failed abc retry 2 storage overloaded").bucket, "alarm_failed");
142 assert.equal(classify("there is no container instance that can be provided to this durable object").bucket, "no_capacity");
143 assert.equal(classify("sandbox not started agent g1t could not read this project's guardrails: down").bucket, "not_started");
144 assert.equal(classify("container exited with unexpected exit code: 137").bucket, "container_exited");
145 assert.equal(classify("Network connection lost.").bucket, "connection_lost");
146 assert.equal(classify("", "canceled").bucket, "caller_gone");
147 assert.equal(classify("something new").bucket, "other");
148 assert.equal(classify("something new").expected, false);
149});
150
151test("buckets sum the counts of their messages", () => {
152 const rows = [
153 { [KEYS.error]: "Durable Object reset because its code was updated.", count: 200 },
154 { [KEYS.error]: "Network connection lost.", count: 7 },
155 { [KEYS.error]: "Durable Object reset because its code was updated (2).", count: 50 },
156 ];
157 assert.deepEqual(
158 bucketsOf(rows, KEYS.error).map((b) => [b.bucket, b.count]),
159 [
160 ["deploy_reset", 250],
161 ["connection_lost", 7],
162 ],
163 );
164});
165
166test("the text report shows rates, statuses, buckets and what could not be read", () => {
167 const parsed = parseDoGroups(
168 answer([
169 { sum: { requests: 100, errors: 31 }, dimensions: { namespaceId: "ns1", status: "scriptThrewException" } },
170 { sum: { requests: 200, errors: 0 }, dimensions: { namespaceId: "ns1", status: "success" } },
171 ]),
172 byNamespace,
173 { ns1: "g1t-runner_AttemptSandbox" },
174 );
175 const rows = [{ [KEYS.error]: "Durable Object reset because its code was updated.", count: 31 }];
176 const text = format({
177 account: "acct",
178 script: "g1t-runner",
179 start: "2026-10-01T00:00:00.000Z",
180 end: "2026-10-08T00:00:00.000Z",
181 analytics: [
182 { key: "by_namespace", label: "By namespace", ...parsed, by_status: byStatus(parsed) },
183 { key: "by_day", label: "By day", error: "unknown field date" },
184 ],
185 telemetry: [
186 { key: "exceptions", label: "Exceptions", rows, buckets: bucketsOf(rows, KEYS.error) },
187 { key: "samples", label: "Samples", rows: [{ at: "2026-10-07T00:00:00.000Z", entrypoint: "AttemptSandbox", event_type: "alarm", outcome: "exception", object: "abc", error: "boom", message: null }] },
188 ],
189 keys: null,
190 notes: ["Telemetry is filtered on $workers.scriptName."],
191 });
192 assert.match(text, /g1t-runner_AttemptSandbox\s+scriptThrewException\s+100\s+31\s+31\.0%/);
193 assert.match(text, /by status:/);
194 assert.match(text, /could not read: unknown field date/);
195 assert.match(text, /31\s+deploy_reset \(expected\)/);
196 assert.match(text, /AttemptSandbox alarm exception abc/);
197 assert.match(text, /\n {4}boom/);
198 assert.doesNotMatch(text, /Bearer|authorization/i);
199});

This file's history is long; its oldest lines are credited to the oldest commit read.