postgresai 0.16.0-dev.0 → 0.16.0-dev.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bin/postgres-ai.ts +71 -6
- package/dist/bin/postgres-ai.js +51 -12
- package/lib/aas-onboard.ts +41 -7
- package/package.json +1 -1
- package/test/aas-onboard.test.ts +84 -0
- package/test/monitoring.test.ts +54 -3
package/bin/postgres-ai.ts
CHANGED
|
@@ -2443,9 +2443,44 @@ interface MonitoringRegistration {
|
|
|
2443
2443
|
*
|
|
2444
2444
|
* Never throws — registration is best-effort; returns null on failure.
|
|
2445
2445
|
*/
|
|
2446
|
+
|
|
2447
|
+
/**
|
|
2448
|
+
* Classify how `mon local-install` should register the monitoring instance,
|
|
2449
|
+
* given the raw `--project` value and the resolved instance id.
|
|
2450
|
+
*
|
|
2451
|
+
* - With an instance id: ADOPT the provisioned instance. No project name is
|
|
2452
|
+
* required (the platform returns the real project); a provided name is
|
|
2453
|
+
* normalized and carried through for messaging/fallback.
|
|
2454
|
+
* - No instance id and no project name: ERROR. The hardcoded
|
|
2455
|
+
* "postgres-ai-monitoring" default was removed, and a nameless legacy
|
|
2456
|
+
* self-registration is rejected by v1.monitoring_instance_register (PT400).
|
|
2457
|
+
* - No instance id but a project name: legacy SELF-REGISTER.
|
|
2458
|
+
*
|
|
2459
|
+
* Pure (no I/O) so the decision is unit-testable independently of the large
|
|
2460
|
+
* install command. `projectName` is the trimmed value, or undefined when empty.
|
|
2461
|
+
*/
|
|
2462
|
+
type MonRegistrationPlan =
|
|
2463
|
+
| { kind: "adopt"; projectName: string | undefined }
|
|
2464
|
+
| { kind: "self-register"; projectName: string }
|
|
2465
|
+
| { kind: "error-missing-project"; projectName: undefined };
|
|
2466
|
+
|
|
2467
|
+
function planMonitoringRegistration(args: {
|
|
2468
|
+
project?: string;
|
|
2469
|
+
instanceId?: string;
|
|
2470
|
+
}): MonRegistrationPlan {
|
|
2471
|
+
const projectName = args.project?.trim() || undefined;
|
|
2472
|
+
if (args.instanceId) {
|
|
2473
|
+
return { kind: "adopt", projectName };
|
|
2474
|
+
}
|
|
2475
|
+
if (!projectName) {
|
|
2476
|
+
return { kind: "error-missing-project", projectName: undefined };
|
|
2477
|
+
}
|
|
2478
|
+
return { kind: "self-register", projectName };
|
|
2479
|
+
}
|
|
2480
|
+
|
|
2446
2481
|
async function registerMonitoringInstance(
|
|
2447
2482
|
apiKey: string,
|
|
2448
|
-
projectName: string,
|
|
2483
|
+
projectName: string | undefined,
|
|
2449
2484
|
opts?: { apiBaseUrl?: string; debug?: boolean; instanceId?: string; retries?: number; retryDelayMs?: number }
|
|
2450
2485
|
): Promise<MonitoringRegistration | null> {
|
|
2451
2486
|
const { apiBaseUrl } = resolveBaseUrls(opts);
|
|
@@ -2457,16 +2492,25 @@ async function registerMonitoringInstance(
|
|
|
2457
2492
|
// moment to recover; skipped before the first attempt. Tests pass 0.
|
|
2458
2493
|
const retryDelayMs = opts?.retryDelayMs ?? 400;
|
|
2459
2494
|
|
|
2495
|
+
// Omit project_name entirely when empty: the adopt path (instance_id present)
|
|
2496
|
+
// relies on the platform returning the real project, and a nameless legacy
|
|
2497
|
+
// self-registration is rejected by v1.monitoring_instance_register (PT400).
|
|
2498
|
+
const hasProjectName = !!(projectName && projectName.trim());
|
|
2499
|
+
|
|
2460
2500
|
if (debug) {
|
|
2461
2501
|
console.error(`\nDebug: Registering monitoring instance...`);
|
|
2462
2502
|
console.error(`Debug: POST ${url}`);
|
|
2463
|
-
console.error(
|
|
2503
|
+
console.error(
|
|
2504
|
+
`Debug: ${hasProjectName ? `project_name=${projectName}` : "project_name=(omitted)"}${instanceId ? ` instance_id=${instanceId}` : ""}`
|
|
2505
|
+
);
|
|
2464
2506
|
}
|
|
2465
2507
|
|
|
2466
2508
|
const requestBody: Record<string, string> = {
|
|
2467
2509
|
api_token: apiKey,
|
|
2468
|
-
project_name: projectName,
|
|
2469
2510
|
};
|
|
2511
|
+
if (hasProjectName) {
|
|
2512
|
+
requestBody.project_name = projectName as string;
|
|
2513
|
+
}
|
|
2470
2514
|
if (instanceId) {
|
|
2471
2515
|
requestBody.instance_id = instanceId;
|
|
2472
2516
|
}
|
|
@@ -3137,8 +3181,11 @@ mon
|
|
|
3137
3181
|
// and persisted; the legacy self-registration stays fire-and-forget
|
|
3138
3182
|
// (issue platform-all#311).
|
|
3139
3183
|
if (apiKey && !opts.demo) {
|
|
3140
|
-
const projectName = opts.project || "postgres-ai-monitoring";
|
|
3141
3184
|
const instanceId = opts.instanceId || process.env.PGAI_INSTANCE_ID;
|
|
3185
|
+
const plan = planMonitoringRegistration({ project: opts.project, instanceId });
|
|
3186
|
+
const projectName = plan.projectName;
|
|
3187
|
+
// `instanceId` truthy ⟺ plan.kind === "adopt"; branch on it directly so
|
|
3188
|
+
// TypeScript narrows instanceId to a defined string in the adopt path.
|
|
3142
3189
|
if (instanceId) {
|
|
3143
3190
|
const reg = await registerMonitoringInstance(apiKey, projectName, {
|
|
3144
3191
|
apiBaseUrl: globalOpts.apiBaseUrl,
|
|
@@ -3160,11 +3207,17 @@ mon
|
|
|
3160
3207
|
// Request succeeded but carried no usable project field — don't claim
|
|
3161
3208
|
// adoption, but don't report a hard failure either (no re-run needed).
|
|
3162
3209
|
console.error(
|
|
3163
|
-
`⚠ Adopted provisioned instance ${instanceId} but the platform returned no project
|
|
3210
|
+
`⚠ Adopted provisioned instance ${instanceId} but the platform returned no project` +
|
|
3211
|
+
(projectName
|
|
3212
|
+
? ` — reports will use project '${projectName}'`
|
|
3213
|
+
: ` — reports will have no project until 'postgresai mon local-install' is re-run with --project <name>`)
|
|
3164
3214
|
);
|
|
3165
3215
|
} else {
|
|
3166
3216
|
console.error(
|
|
3167
|
-
`⚠ Could not adopt provisioned instance ${instanceId}
|
|
3217
|
+
`⚠ Could not adopt provisioned instance ${instanceId}` +
|
|
3218
|
+
(projectName
|
|
3219
|
+
? ` — reports will use project '${projectName}' until 'postgresai mon local-install' is re-run`
|
|
3220
|
+
: ` — reports will have no project until 'postgresai mon local-install' is re-run with --project <name>`)
|
|
3168
3221
|
);
|
|
3169
3222
|
}
|
|
3170
3223
|
|
|
@@ -3188,6 +3241,17 @@ mon
|
|
|
3188
3241
|
`⚠ AAS auto-collection not registered (${aas.reason}); it can be enabled later by re-running 'postgresai mon local-install'\n`
|
|
3189
3242
|
);
|
|
3190
3243
|
}
|
|
3244
|
+
} else if (plan.kind === "error-missing-project") {
|
|
3245
|
+
// Legacy self-registration (no --instance-id) now requires a project
|
|
3246
|
+
// name: the hardcoded "postgres-ai-monitoring" default was removed, and
|
|
3247
|
+
// v1.monitoring_instance_register raises PT400 for a nameless legacy
|
|
3248
|
+
// registration. Console-provisioned installs should adopt with
|
|
3249
|
+
// --instance-id instead.
|
|
3250
|
+
console.error(
|
|
3251
|
+
"✗ A project name is required for self-registration (the 'postgres-ai-monitoring' default was removed). " +
|
|
3252
|
+
"Re-run with --project <name>, or adopt a console-provisioned instance with --instance-id <uuid>."
|
|
3253
|
+
);
|
|
3254
|
+
process.exitCode = 1;
|
|
3191
3255
|
} else {
|
|
3192
3256
|
void registerMonitoringInstance(apiKey, projectName, {
|
|
3193
3257
|
apiBaseUrl: globalOpts.apiBaseUrl,
|
|
@@ -5488,3 +5552,4 @@ if (import.meta.main) {
|
|
|
5488
5552
|
// same functions used by the `mon` commands).
|
|
5489
5553
|
export { refreshBundledComposeIfStale, readDeployedTag, isValidComposeYaml };
|
|
5490
5554
|
export { registerMonitoringInstance, resolveAdoptedProject, type MonitoringRegistration };
|
|
5555
|
+
export { planMonitoringRegistration, type MonRegistrationPlan };
|
package/dist/bin/postgres-ai.js
CHANGED
|
@@ -13425,7 +13425,7 @@ var {
|
|
|
13425
13425
|
// package.json
|
|
13426
13426
|
var package_default = {
|
|
13427
13427
|
name: "postgresai",
|
|
13428
|
-
version: "0.16.0-dev.
|
|
13428
|
+
version: "0.16.0-dev.2",
|
|
13429
13429
|
description: "postgres_ai CLI",
|
|
13430
13430
|
license: "Apache-2.0",
|
|
13431
13431
|
private: false,
|
|
@@ -16256,7 +16256,7 @@ var Result = import_lib.default.Result;
|
|
|
16256
16256
|
var TypeOverrides = import_lib.default.TypeOverrides;
|
|
16257
16257
|
var defaults = import_lib.default.defaults;
|
|
16258
16258
|
// package.json
|
|
16259
|
-
var version = "0.16.0-dev.
|
|
16259
|
+
var version = "0.16.0-dev.2";
|
|
16260
16260
|
var package_default2 = {
|
|
16261
16261
|
name: "postgresai",
|
|
16262
16262
|
version,
|
|
@@ -28498,7 +28498,7 @@ function grafanaBaseUrl() {
|
|
|
28498
28498
|
return (process.env.PGAI_GRAFANA_LOCAL_URL || "http://localhost:3000").replace(/\/+$/, "");
|
|
28499
28499
|
}
|
|
28500
28500
|
function grafanaAdminUser() {
|
|
28501
|
-
return process.env.GF_SECURITY_ADMIN_USER || "
|
|
28501
|
+
return process.env.GF_SECURITY_ADMIN_USER || "monitor";
|
|
28502
28502
|
}
|
|
28503
28503
|
function parseVcpus(raw) {
|
|
28504
28504
|
if (raw === undefined || raw === null || raw === "")
|
|
@@ -28575,9 +28575,14 @@ async function resolveDatasourceId(adminPassword, debug = false) {
|
|
|
28575
28575
|
return null;
|
|
28576
28576
|
const list = await res.json().catch(() => []);
|
|
28577
28577
|
const prom = list.filter((d) => d.type === "prometheus");
|
|
28578
|
-
if (prom.length
|
|
28578
|
+
if (prom.length > 1) {
|
|
28579
28579
|
if (debug)
|
|
28580
|
-
console.error(`Debug: AAS:
|
|
28580
|
+
console.error(`Debug: AAS: ${prom.length} prometheus datasources (ambiguous); not retrying`);
|
|
28581
|
+
return "ambiguous";
|
|
28582
|
+
}
|
|
28583
|
+
if (prom.length === 0) {
|
|
28584
|
+
if (debug)
|
|
28585
|
+
console.error(`Debug: AAS: no prometheus datasource resolvable yet`);
|
|
28581
28586
|
return null;
|
|
28582
28587
|
}
|
|
28583
28588
|
return typeof prom[0].id === "number" ? prom[0].id : null;
|
|
@@ -28593,7 +28598,23 @@ async function registerAasCollection(apiKey, instanceId, opts) {
|
|
|
28593
28598
|
const labels = resolveAasLabels(opts.instancesPath);
|
|
28594
28599
|
if (!labels)
|
|
28595
28600
|
return { ok: false, reason: "could not determine a single (cluster, node_name) target" };
|
|
28596
|
-
const
|
|
28601
|
+
const maxAttempts = opts.datasourceMaxAttempts ?? 20;
|
|
28602
|
+
const retryDelayMs = opts.datasourceRetryDelayMs ?? 3000;
|
|
28603
|
+
let datasourceId = null;
|
|
28604
|
+
for (let attempt = 1;attempt <= maxAttempts; attempt++) {
|
|
28605
|
+
const resolved = await resolveDatasourceId(opts.grafanaPassword, debug);
|
|
28606
|
+
if (typeof resolved === "number") {
|
|
28607
|
+
datasourceId = resolved;
|
|
28608
|
+
break;
|
|
28609
|
+
}
|
|
28610
|
+
if (resolved === "ambiguous")
|
|
28611
|
+
break;
|
|
28612
|
+
if (attempt < maxAttempts) {
|
|
28613
|
+
if (debug)
|
|
28614
|
+
console.error(`Debug: AAS: datasource not resolvable yet (attempt ${attempt}/${maxAttempts}); waiting for Grafana…`);
|
|
28615
|
+
await new Promise((resolve4) => setTimeout(resolve4, retryDelayMs));
|
|
28616
|
+
}
|
|
28617
|
+
}
|
|
28597
28618
|
if (datasourceId == null)
|
|
28598
28619
|
return { ok: false, reason: "could not resolve the Prometheus datasource id" };
|
|
28599
28620
|
const saToken = await mintAasServiceAccountToken(opts.grafanaPassword, debug);
|
|
@@ -35638,6 +35659,16 @@ function checkRunningContainers() {
|
|
|
35638
35659
|
return { running: false, containers: [] };
|
|
35639
35660
|
}
|
|
35640
35661
|
}
|
|
35662
|
+
function planMonitoringRegistration(args) {
|
|
35663
|
+
const projectName = args.project?.trim() || undefined;
|
|
35664
|
+
if (args.instanceId) {
|
|
35665
|
+
return { kind: "adopt", projectName };
|
|
35666
|
+
}
|
|
35667
|
+
if (!projectName) {
|
|
35668
|
+
return { kind: "error-missing-project", projectName: undefined };
|
|
35669
|
+
}
|
|
35670
|
+
return { kind: "self-register", projectName };
|
|
35671
|
+
}
|
|
35641
35672
|
async function registerMonitoringInstance(apiKey, projectName, opts) {
|
|
35642
35673
|
const { apiBaseUrl } = resolveBaseUrls2(opts);
|
|
35643
35674
|
const url = `${apiBaseUrl}/rpc/monitoring_instance_register`;
|
|
@@ -35645,16 +35676,19 @@ async function registerMonitoringInstance(apiKey, projectName, opts) {
|
|
|
35645
35676
|
const instanceId = opts?.instanceId;
|
|
35646
35677
|
const retries = opts?.retries ?? (instanceId ? 1 : 0);
|
|
35647
35678
|
const retryDelayMs = opts?.retryDelayMs ?? 400;
|
|
35679
|
+
const hasProjectName = !!(projectName && projectName.trim());
|
|
35648
35680
|
if (debug) {
|
|
35649
35681
|
console.error(`
|
|
35650
35682
|
Debug: Registering monitoring instance...`);
|
|
35651
35683
|
console.error(`Debug: POST ${url}`);
|
|
35652
|
-
console.error(`Debug: project_name=${projectName}${instanceId ? ` instance_id=${instanceId}` : ""}`);
|
|
35684
|
+
console.error(`Debug: ${hasProjectName ? `project_name=${projectName}` : "project_name=(omitted)"}${instanceId ? ` instance_id=${instanceId}` : ""}`);
|
|
35653
35685
|
}
|
|
35654
35686
|
const requestBody = {
|
|
35655
|
-
api_token: apiKey
|
|
35656
|
-
project_name: projectName
|
|
35687
|
+
api_token: apiKey
|
|
35657
35688
|
};
|
|
35689
|
+
if (hasProjectName) {
|
|
35690
|
+
requestBody.project_name = projectName;
|
|
35691
|
+
}
|
|
35658
35692
|
if (instanceId) {
|
|
35659
35693
|
requestBody.instance_id = instanceId;
|
|
35660
35694
|
}
|
|
@@ -36197,8 +36231,9 @@ Searched: ${demoCandidates.join(", ")}
|
|
|
36197
36231
|
console.log(`\u2713 Services started
|
|
36198
36232
|
`);
|
|
36199
36233
|
if (apiKey && !opts.demo) {
|
|
36200
|
-
const projectName = opts.project || "postgres-ai-monitoring";
|
|
36201
36234
|
const instanceId = opts.instanceId || process.env.PGAI_INSTANCE_ID;
|
|
36235
|
+
const plan = planMonitoringRegistration({ project: opts.project, instanceId });
|
|
36236
|
+
const projectName = plan.projectName;
|
|
36202
36237
|
if (instanceId) {
|
|
36203
36238
|
const reg = await registerMonitoringInstance(apiKey, projectName, {
|
|
36204
36239
|
apiBaseUrl: globalOpts.apiBaseUrl,
|
|
@@ -36214,9 +36249,9 @@ Searched: ${demoCandidates.join(", ")}
|
|
|
36214
36249
|
console.log(`\u2713 ${verb} monitoring instance (project: ${adoptedProject})
|
|
36215
36250
|
`);
|
|
36216
36251
|
} else if (reg) {
|
|
36217
|
-
console.error(`\u26A0 Adopted provisioned instance ${instanceId} but the platform returned no project \u2014 reports will use project '${projectName}'`);
|
|
36252
|
+
console.error(`\u26A0 Adopted provisioned instance ${instanceId} but the platform returned no project` + (projectName ? ` \u2014 reports will use project '${projectName}'` : ` \u2014 reports will have no project until 'postgresai mon local-install' is re-run with --project <name>`));
|
|
36218
36253
|
} else {
|
|
36219
|
-
console.error(`\u26A0 Could not adopt provisioned instance ${instanceId} \u2014 reports will use project '${projectName}' until 'postgresai mon local-install' is re-run`);
|
|
36254
|
+
console.error(`\u26A0 Could not adopt provisioned instance ${instanceId}` + (projectName ? ` \u2014 reports will use project '${projectName}' until 'postgresai mon local-install' is re-run` : ` \u2014 reports will have no project until 'postgresai mon local-install' is re-run with --project <name>`));
|
|
36220
36255
|
}
|
|
36221
36256
|
const aas = await registerAasCollection(apiKey, instanceId, {
|
|
36222
36257
|
grafanaPassword,
|
|
@@ -36232,6 +36267,9 @@ Searched: ${demoCandidates.join(", ")}
|
|
|
36232
36267
|
console.error(`\u26A0 AAS auto-collection not registered (${aas.reason}); it can be enabled later by re-running 'postgresai mon local-install'
|
|
36233
36268
|
`);
|
|
36234
36269
|
}
|
|
36270
|
+
} else if (plan.kind === "error-missing-project") {
|
|
36271
|
+
console.error("\u2717 A project name is required for self-registration (the 'postgres-ai-monitoring' default was removed). " + "Re-run with --project <name>, or adopt a console-provisioned instance with --instance-id <uuid>.");
|
|
36272
|
+
process.exitCode = 1;
|
|
36235
36273
|
} else {
|
|
36236
36274
|
registerMonitoringInstance(apiKey, projectName, {
|
|
36237
36275
|
apiBaseUrl: globalOpts.apiBaseUrl,
|
|
@@ -37997,5 +38035,6 @@ export {
|
|
|
37997
38035
|
registerMonitoringInstance,
|
|
37998
38036
|
refreshBundledComposeIfStale,
|
|
37999
38037
|
readDeployedTag,
|
|
38038
|
+
planMonitoringRegistration,
|
|
38000
38039
|
isValidComposeYaml
|
|
38001
38040
|
};
|
package/lib/aas-onboard.ts
CHANGED
|
@@ -28,7 +28,11 @@ function grafanaBaseUrl(): string {
|
|
|
28
28
|
}
|
|
29
29
|
|
|
30
30
|
function grafanaAdminUser(): string {
|
|
31
|
-
|
|
31
|
+
// The monitoring stack's compose hardcodes the Grafana admin user to
|
|
32
|
+
// "monitor" (GF_SECURITY_ADMIN_USER: monitor), so default to that rather than
|
|
33
|
+
// Grafana's stock "admin" — otherwise AAS arming logs in as the wrong user
|
|
34
|
+
// and every datasource lookup 401s. An explicit env override still wins.
|
|
35
|
+
return process.env.GF_SECURITY_ADMIN_USER || "monitor";
|
|
32
36
|
}
|
|
33
37
|
|
|
34
38
|
/** Parse a vcpus input (flag/env) to a non-negative integer; 0 = "unknown" fallback. */
|
|
@@ -130,17 +134,26 @@ export async function mintAasServiceAccountToken(
|
|
|
130
134
|
* Resolve the single Prometheus-typed datasource's numeric id on the local
|
|
131
135
|
* Grafana. The monitoring stack's VictoriaMetrics datasource is type
|
|
132
136
|
* "prometheus" (VM speaks PromQL), and the stack registers exactly one such
|
|
133
|
-
* datasource — the same one the collector queries.
|
|
134
|
-
*
|
|
137
|
+
* datasource — the same one the collector queries. 0 / API-not-ready → null
|
|
138
|
+
* (a provisioning transient — the readiness loop retries); >1 → "ambiguous"
|
|
139
|
+
* (a permanent misconfiguration — the loop stops at once), matching
|
|
140
|
+
* v1.aas_onboard's >1 skip.
|
|
135
141
|
*/
|
|
136
|
-
export async function resolveDatasourceId(adminPassword: string, debug = false): Promise<number | null> {
|
|
142
|
+
export async function resolveDatasourceId(adminPassword: string, debug = false): Promise<number | "ambiguous" | null> {
|
|
137
143
|
try {
|
|
138
144
|
const res = await grafanaApi("GET", "/api/datasources", adminPassword);
|
|
139
145
|
if (!res.ok) return null;
|
|
140
146
|
const list = (await res.json().catch(() => [])) as Array<{ id?: unknown; type?: unknown }>;
|
|
141
147
|
const prom = list.filter((d) => d.type === "prometheus");
|
|
142
|
-
if (prom.length
|
|
143
|
-
|
|
148
|
+
if (prom.length > 1) {
|
|
149
|
+
// >1 is a permanent misconfiguration, not a provisioning transient: the
|
|
150
|
+
// datasource count only grows as Grafana provisions, so retrying can never
|
|
151
|
+
// resolve it. Signal a definitive skip so the readiness loop bails at once.
|
|
152
|
+
if (debug) console.error(`Debug: AAS: ${prom.length} prometheus datasources (ambiguous); not retrying`);
|
|
153
|
+
return "ambiguous";
|
|
154
|
+
}
|
|
155
|
+
if (prom.length === 0) {
|
|
156
|
+
if (debug) console.error(`Debug: AAS: no prometheus datasource resolvable yet`);
|
|
144
157
|
return null;
|
|
145
158
|
}
|
|
146
159
|
return typeof prom[0].id === "number" ? prom[0].id : null;
|
|
@@ -169,6 +182,10 @@ export async function registerAasCollection(
|
|
|
169
182
|
apiBaseUrl?: string;
|
|
170
183
|
debug?: boolean;
|
|
171
184
|
fetchImpl?: typeof fetch;
|
|
185
|
+
// Grafana-readiness polling for the datasource lookup (Grafana has just
|
|
186
|
+
// been started by `compose up`). Defaults: 20 attempts × 3s.
|
|
187
|
+
datasourceMaxAttempts?: number;
|
|
188
|
+
datasourceRetryDelayMs?: number;
|
|
172
189
|
}
|
|
173
190
|
): Promise<AasRegisterResult> {
|
|
174
191
|
const debug = !!opts.debug;
|
|
@@ -178,7 +195,24 @@ export async function registerAasCollection(
|
|
|
178
195
|
const labels = resolveAasLabels(opts.instancesPath);
|
|
179
196
|
if (!labels) return { ok: false, reason: "could not determine a single (cluster, node_name) target" };
|
|
180
197
|
|
|
181
|
-
|
|
198
|
+
// Grafana was just started by `compose up`; it needs time to create its
|
|
199
|
+
// admin user, provision datasources, and serve its API. Querying too early
|
|
200
|
+
// makes the datasource lookup fail transiently, so poll until it resolves
|
|
201
|
+
// (best-effort, capped — the install never blocks on this).
|
|
202
|
+
const maxAttempts = opts.datasourceMaxAttempts ?? 20;
|
|
203
|
+
const retryDelayMs = opts.datasourceRetryDelayMs ?? 3000;
|
|
204
|
+
let datasourceId: number | null = null;
|
|
205
|
+
for (let attempt = 1; attempt <= maxAttempts; attempt++) {
|
|
206
|
+
const resolved = await resolveDatasourceId(opts.grafanaPassword, debug);
|
|
207
|
+
if (typeof resolved === "number") { datasourceId = resolved; break; }
|
|
208
|
+
// "ambiguous" (>1 prometheus datasource) is permanent — retrying can't fix
|
|
209
|
+
// it, so stop polling immediately instead of waiting out the whole budget.
|
|
210
|
+
if (resolved === "ambiguous") break;
|
|
211
|
+
if (attempt < maxAttempts) {
|
|
212
|
+
if (debug) console.error(`Debug: AAS: datasource not resolvable yet (attempt ${attempt}/${maxAttempts}); waiting for Grafana…`);
|
|
213
|
+
await new Promise((resolve) => setTimeout(resolve, retryDelayMs));
|
|
214
|
+
}
|
|
215
|
+
}
|
|
182
216
|
if (datasourceId == null) return { ok: false, reason: "could not resolve the Prometheus datasource id" };
|
|
183
217
|
|
|
184
218
|
const saToken = await mintAasServiceAccountToken(opts.grafanaPassword, debug);
|
package/package.json
CHANGED
package/test/aas-onboard.test.ts
CHANGED
|
@@ -198,6 +198,8 @@ describe("registerAasCollection", () => {
|
|
|
198
198
|
installFetch({ prometheusCount: n });
|
|
199
199
|
const r = await registerAasCollection("apikey-1", "inst-123", {
|
|
200
200
|
grafanaPassword: "pw", instancesPath, vcpus: 8, apiBaseUrl: "https://api.test",
|
|
201
|
+
// 0/>1 is a definitive skip; cap the readiness retry so the test stays fast.
|
|
202
|
+
datasourceMaxAttempts: 2, datasourceRetryDelayMs: 0,
|
|
201
203
|
});
|
|
202
204
|
expect(r.ok).toBe(false);
|
|
203
205
|
expect(r.reason).toContain("datasource");
|
|
@@ -205,6 +207,88 @@ describe("registerAasCollection", () => {
|
|
|
205
207
|
}
|
|
206
208
|
});
|
|
207
209
|
|
|
210
|
+
test("polls the datasource until Grafana is ready, then registers", async () => {
|
|
211
|
+
// Grafana isn't ready on the first probes (no prometheus datasource yet),
|
|
212
|
+
// then it provisions — the readiness retry must keep going and then succeed.
|
|
213
|
+
let dsProbes = 0;
|
|
214
|
+
calls = [];
|
|
215
|
+
fetchSpy = spyOn(globalThis, "fetch").mockImplementation((async (input: unknown, init?: { method?: string; body?: string }) => {
|
|
216
|
+
const url = String(input);
|
|
217
|
+
const method = (init?.method || "GET").toUpperCase();
|
|
218
|
+
calls.push({ url, method, body: init?.body });
|
|
219
|
+
if (url.includes("/api/serviceaccounts/search")) return res(true, 200, { serviceAccounts: [] });
|
|
220
|
+
if (url.match(/\/tokens$/) && method === "POST") return res(true, 200, { key: "glsa_mock" });
|
|
221
|
+
if (url.endsWith("/api/serviceaccounts") && method === "POST") return res(true, 201, { id: 42 });
|
|
222
|
+
if (url.includes("/api/datasources")) {
|
|
223
|
+
dsProbes++;
|
|
224
|
+
return dsProbes < 3
|
|
225
|
+
? res(true, 200, [{ id: 3, type: "loki" }]) // not ready yet
|
|
226
|
+
: res(true, 200, [{ id: 8, type: "prometheus" }, { id: 3, type: "loki" }]);
|
|
227
|
+
}
|
|
228
|
+
if (url.includes("/rpc/monitoring_instance_aas_register")) return res(true, 200, {});
|
|
229
|
+
return res(false, 404, {});
|
|
230
|
+
}) as unknown as typeof fetch);
|
|
231
|
+
|
|
232
|
+
const r = await registerAasCollection("apikey-1", "inst-123", {
|
|
233
|
+
grafanaPassword: "pw", instancesPath, vcpus: 8, apiBaseUrl: "https://api.test",
|
|
234
|
+
datasourceMaxAttempts: 6, datasourceRetryDelayMs: 0,
|
|
235
|
+
});
|
|
236
|
+
expect(r.ok).toBe(true);
|
|
237
|
+
expect(dsProbes).toBeGreaterThanOrEqual(3); // kept polling past the not-ready probes
|
|
238
|
+
const rpc = calls.find((c) => c.url.includes("/rpc/monitoring_instance_aas_register"));
|
|
239
|
+
expect(rpc).toBeDefined();
|
|
240
|
+
expect(JSON.parse(rpc!.body!).datasource_id).toBe(8);
|
|
241
|
+
});
|
|
242
|
+
|
|
243
|
+
test(">1 prometheus datasource is a definitive skip: one probe, no retry", async () => {
|
|
244
|
+
// The >1 case is permanent (the datasource count only grows), so the
|
|
245
|
+
// readiness loop must bail after a single probe, not burn its whole budget.
|
|
246
|
+
let dsProbes = 0;
|
|
247
|
+
calls = [];
|
|
248
|
+
fetchSpy = spyOn(globalThis, "fetch").mockImplementation((async (input: unknown, init?: { method?: string; body?: string }) => {
|
|
249
|
+
const url = String(input);
|
|
250
|
+
const method = (init?.method || "GET").toUpperCase();
|
|
251
|
+
calls.push({ url, method, body: init?.body });
|
|
252
|
+
if (url.includes("/api/datasources")) {
|
|
253
|
+
dsProbes++;
|
|
254
|
+
return res(true, 200, [{ id: 8, type: "prometheus" }, { id: 9, type: "prometheus" }, { id: 3, type: "loki" }]);
|
|
255
|
+
}
|
|
256
|
+
return res(false, 404, {});
|
|
257
|
+
}) as unknown as typeof fetch);
|
|
258
|
+
|
|
259
|
+
const r = await registerAasCollection("apikey-1", "inst-123", {
|
|
260
|
+
grafanaPassword: "pw", instancesPath, vcpus: 8, apiBaseUrl: "https://api.test",
|
|
261
|
+
datasourceMaxAttempts: 5, datasourceRetryDelayMs: 0,
|
|
262
|
+
});
|
|
263
|
+
expect(r.ok).toBe(false);
|
|
264
|
+
expect(r.reason).toContain("datasource");
|
|
265
|
+
expect(dsProbes).toBe(1); // bailed after one probe; did NOT retry 5x
|
|
266
|
+
expect(calls.some((c) => c.url.includes("/rpc/monitoring_instance_aas_register"))).toBe(false);
|
|
267
|
+
});
|
|
268
|
+
|
|
269
|
+
test("never-ready datasource: polls exactly maxAttempts times, then ok:false", async () => {
|
|
270
|
+
// Bounds the readiness loop: a never-appearing datasource must probe exactly
|
|
271
|
+
// maxAttempts times (N probes, N-1 sleeps) and then give up — not loop forever.
|
|
272
|
+
let dsProbes = 0;
|
|
273
|
+
calls = [];
|
|
274
|
+
fetchSpy = spyOn(globalThis, "fetch").mockImplementation((async (input: unknown, init?: { method?: string; body?: string }) => {
|
|
275
|
+
const url = String(input);
|
|
276
|
+
const method = (init?.method || "GET").toUpperCase();
|
|
277
|
+
calls.push({ url, method, body: init?.body });
|
|
278
|
+
if (url.includes("/api/datasources")) { dsProbes++; return res(true, 200, [{ id: 3, type: "loki" }]); } // never a prometheus
|
|
279
|
+
return res(false, 404, {});
|
|
280
|
+
}) as unknown as typeof fetch);
|
|
281
|
+
|
|
282
|
+
const r = await registerAasCollection("apikey-1", "inst-123", {
|
|
283
|
+
grafanaPassword: "pw", instancesPath, vcpus: 8, apiBaseUrl: "https://api.test",
|
|
284
|
+
datasourceMaxAttempts: 3, datasourceRetryDelayMs: 0,
|
|
285
|
+
});
|
|
286
|
+
expect(r.ok).toBe(false);
|
|
287
|
+
expect(r.reason).toContain("datasource");
|
|
288
|
+
expect(dsProbes).toBe(3); // bounded: exactly maxAttempts probes
|
|
289
|
+
expect(calls.some((c) => c.url.includes("/rpc/monitoring_instance_aas_register"))).toBe(false);
|
|
290
|
+
});
|
|
291
|
+
|
|
208
292
|
test("mint returning no key → ok:false, no RPC call", async () => {
|
|
209
293
|
installFetch({ mintKey: null });
|
|
210
294
|
const r = await registerAasCollection("apikey-1", "inst-123", {
|
package/test/monitoring.test.ts
CHANGED
|
@@ -19,6 +19,7 @@ import {
|
|
|
19
19
|
import {
|
|
20
20
|
registerMonitoringInstance,
|
|
21
21
|
resolveAdoptedProject,
|
|
22
|
+
planMonitoringRegistration,
|
|
22
23
|
} from "../bin/postgres-ai";
|
|
23
24
|
|
|
24
25
|
/**
|
|
@@ -241,12 +242,13 @@ describe("registerMonitoringInstance", () => {
|
|
|
241
242
|
});
|
|
242
243
|
|
|
243
244
|
// Issue platform-all#311: console-provisioned installs pass instance_id so
|
|
244
|
-
// the platform adopts the provisioned instance
|
|
245
|
-
//
|
|
245
|
+
// the platform adopts the provisioned instance and returns its real project
|
|
246
|
+
// — the CLI sends NO project_name on the adopt path (the hardcoded
|
|
247
|
+
// "postgres-ai-monitoring" default was removed).
|
|
246
248
|
test("includes instance_id in body when adopting a provisioned instance", async () => {
|
|
247
249
|
const instanceId = "019eb300-3f2a-7a75-b54d-4f10572b25b8";
|
|
248
250
|
|
|
249
|
-
await registerMonitoringInstance("key",
|
|
251
|
+
await registerMonitoringInstance("key", undefined, opts({ instanceId }));
|
|
250
252
|
|
|
251
253
|
const body = JSON.parse(fetchCalls[0].options.body as string);
|
|
252
254
|
expect(body.instance_id).toBe(instanceId);
|
|
@@ -255,6 +257,22 @@ describe("registerMonitoringInstance", () => {
|
|
|
255
257
|
expect(headers["instance-id"]).toBeUndefined();
|
|
256
258
|
});
|
|
257
259
|
|
|
260
|
+
test("omits project_name from the body when undefined (adopt path)", async () => {
|
|
261
|
+
await registerMonitoringInstance("key", undefined, opts({ instanceId: "i" }));
|
|
262
|
+
|
|
263
|
+
const body = JSON.parse(fetchCalls[0].options.body as string);
|
|
264
|
+
// The adopt path sends no name; the platform returns the real project.
|
|
265
|
+
expect("project_name" in body).toBe(false);
|
|
266
|
+
expect(body.api_token).toBe("key");
|
|
267
|
+
});
|
|
268
|
+
|
|
269
|
+
test("omits project_name from the body when empty/whitespace", async () => {
|
|
270
|
+
await registerMonitoringInstance("key", " ", opts({ instanceId: "i" }));
|
|
271
|
+
|
|
272
|
+
const body = JSON.parse(fetchCalls[0].options.body as string);
|
|
273
|
+
expect("project_name" in body).toBe(false);
|
|
274
|
+
});
|
|
275
|
+
|
|
258
276
|
test("omits instance_id from the body for legacy self-registration", async () => {
|
|
259
277
|
await registerMonitoringInstance("key", "my-project", opts());
|
|
260
278
|
|
|
@@ -262,6 +280,8 @@ describe("registerMonitoringInstance", () => {
|
|
|
262
280
|
// PostgREST matches the 3-arg function via its default — the key must be
|
|
263
281
|
// ABSENT (not null) so legacy CLIs and the new one hit the same overload.
|
|
264
282
|
expect("instance_id" in body).toBe(false);
|
|
283
|
+
// A real project name still rides in the body for legacy registration.
|
|
284
|
+
expect(body.project_name).toBe("my-project");
|
|
265
285
|
});
|
|
266
286
|
|
|
267
287
|
test("a 200 with {project_id, project_name} returns a populated result", async () => {
|
|
@@ -323,6 +343,37 @@ describe("registerMonitoringInstance", () => {
|
|
|
323
343
|
});
|
|
324
344
|
});
|
|
325
345
|
|
|
346
|
+
describe("planMonitoringRegistration — mon local-install registration decision", () => {
|
|
347
|
+
test("instance id present → adopt, no project name required", () => {
|
|
348
|
+
const plan = planMonitoringRegistration({ instanceId: "i-123" });
|
|
349
|
+
expect(plan.kind).toBe("adopt");
|
|
350
|
+
expect(plan.projectName).toBeUndefined();
|
|
351
|
+
});
|
|
352
|
+
|
|
353
|
+
test("instance id present + project → adopt, carries the trimmed name", () => {
|
|
354
|
+
const plan = planMonitoringRegistration({ instanceId: "i-123", project: " prod-db " });
|
|
355
|
+
expect(plan.kind).toBe("adopt");
|
|
356
|
+
expect(plan.projectName).toBe("prod-db");
|
|
357
|
+
});
|
|
358
|
+
|
|
359
|
+
test("no instance id + no project → error-missing-project (exit 1 path)", () => {
|
|
360
|
+
const plan = planMonitoringRegistration({});
|
|
361
|
+
expect(plan.kind).toBe("error-missing-project");
|
|
362
|
+
expect(plan.projectName).toBeUndefined();
|
|
363
|
+
});
|
|
364
|
+
|
|
365
|
+
test("no instance id + empty/whitespace project → error-missing-project", () => {
|
|
366
|
+
expect(planMonitoringRegistration({ project: "" }).kind).toBe("error-missing-project");
|
|
367
|
+
expect(planMonitoringRegistration({ project: " " }).kind).toBe("error-missing-project");
|
|
368
|
+
});
|
|
369
|
+
|
|
370
|
+
test("no instance id + real project → legacy self-register with the trimmed name", () => {
|
|
371
|
+
const plan = planMonitoringRegistration({ project: " my-project " });
|
|
372
|
+
expect(plan.kind).toBe("self-register");
|
|
373
|
+
expect(plan.projectName).toBe("my-project");
|
|
374
|
+
});
|
|
375
|
+
});
|
|
376
|
+
|
|
326
377
|
describe("resolveAdoptedProject — what gets persisted to .pgwatch-config", () => {
|
|
327
378
|
test("prefers the numeric project_id over the name (survives renames)", () => {
|
|
328
379
|
expect(resolveAdoptedProject({ projectId: 42, projectName: "prod-db" })).toBe("42");
|