mirror of
https://github.com/fosrl/pangolin.git
synced 2026-05-30 20:52:40 +00:00
Add resource degraded
This commit is contained in:
@@ -23,6 +23,7 @@ import {
|
||||
} from "@server/db";
|
||||
import { eq } from "drizzle-orm";
|
||||
import {
|
||||
fireResourceDegradedAlert,
|
||||
fireResourceHealthyAlert,
|
||||
fireResourceUnhealthyAlert
|
||||
} from "./resourceEvents";
|
||||
@@ -217,6 +218,14 @@ async function handleResource(orgId: string, healthCheckTargetId?: number | null
|
||||
undefined,
|
||||
trx
|
||||
);
|
||||
} else if (health === "degraded") {
|
||||
await fireResourceDegradedAlert(
|
||||
orgId,
|
||||
resource.resourceId,
|
||||
resource.name,
|
||||
undefined,
|
||||
trx
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -130,9 +130,9 @@ export async function fireResourceUnhealthyAlert(
|
||||
}
|
||||
|
||||
/**
|
||||
* Fire a `resource_toggle` alert for the given resource.
|
||||
* Fire a `resource_degraded` alert for the given resource.
|
||||
*
|
||||
* Call this when a resource's enabled/disabled status is toggled so that any
|
||||
* Call this after a resource has been detected as degraded so that any
|
||||
* matching `alertRules` can dispatch their email and webhook actions.
|
||||
*
|
||||
* @param orgId - Organisation that owns the resource.
|
||||
@@ -140,7 +140,7 @@ export async function fireResourceUnhealthyAlert(
|
||||
* @param resourceName - Human-readable name shown in notifications (optional).
|
||||
* @param extra - Any additional key/value pairs to include in the payload.
|
||||
*/
|
||||
export async function fireResourceToggleAlert(
|
||||
export async function fireResourceDegradedAlert(
|
||||
orgId: string,
|
||||
resourceId: number,
|
||||
resourceName?: string | null,
|
||||
@@ -148,8 +148,16 @@ export async function fireResourceToggleAlert(
|
||||
trx: Transaction | typeof db = db
|
||||
): Promise<void> {
|
||||
try {
|
||||
await trx.insert(statusHistory).values({
|
||||
entityType: "resource",
|
||||
entityId: resourceId,
|
||||
orgId: orgId,
|
||||
status: "degraded",
|
||||
timestamp: Math.floor(Date.now() / 1000)
|
||||
});
|
||||
|
||||
await processAlerts({
|
||||
eventType: "resource_toggle",
|
||||
eventType: "resource_degraded",
|
||||
orgId,
|
||||
resourceId,
|
||||
data: {
|
||||
@@ -157,9 +165,20 @@ export async function fireResourceToggleAlert(
|
||||
...extra
|
||||
}
|
||||
});
|
||||
await processAlerts({
|
||||
eventType: "resource_toggle",
|
||||
orgId,
|
||||
resourceId,
|
||||
data: {
|
||||
resourceId,
|
||||
status: "degraded",
|
||||
...(resourceName != null ? { resourceName } : {}),
|
||||
...extra
|
||||
}
|
||||
});
|
||||
} catch (err) {
|
||||
logger.error(
|
||||
`fireResourceToggleAlert: unexpected error for resourceId ${resourceId}`,
|
||||
`fireResourceDegradedAlert: unexpected error for resourceId ${resourceId}`,
|
||||
err
|
||||
);
|
||||
}
|
||||
|
||||
@@ -88,6 +88,8 @@ function buildSubject(context: AlertContext): string {
|
||||
return "[Alert] Resource Healthy";
|
||||
case "resource_unhealthy":
|
||||
return "[Alert] Resource Unhealthy";
|
||||
case "resource_degraded":
|
||||
return "[Alert] Resource Degraded";
|
||||
case "resource_toggle":
|
||||
return "[Alert] Resource Status Changed";
|
||||
default: {
|
||||
|
||||
@@ -12,7 +12,10 @@
|
||||
*/
|
||||
|
||||
import logger from "@server/logger";
|
||||
import { AlertContext, WebhookAlertConfig } from "@server/routers/alertRule/types";
|
||||
import {
|
||||
AlertContext,
|
||||
WebhookAlertConfig
|
||||
} from "@server/routers/alertRule/types";
|
||||
|
||||
const REQUEST_TIMEOUT_MS = 15_000;
|
||||
const MAX_RETRIES = 3;
|
||||
@@ -56,7 +59,10 @@ export async function sendAlertWebhook(
|
||||
|
||||
for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
|
||||
const controller = new AbortController();
|
||||
const timeoutHandle = setTimeout(() => controller.abort(), REQUEST_TIMEOUT_MS);
|
||||
const timeoutHandle = setTimeout(
|
||||
() => controller.abort(),
|
||||
REQUEST_TIMEOUT_MS
|
||||
);
|
||||
|
||||
let response: Response;
|
||||
try {
|
||||
@@ -75,7 +81,9 @@ export async function sendAlertWebhook(
|
||||
);
|
||||
} else {
|
||||
const msg = err instanceof Error ? err.message : String(err);
|
||||
lastError = new Error(`Alert webhook: request to "${url}" failed – ${msg}`);
|
||||
lastError = new Error(
|
||||
`Alert webhook: request to "${url}" failed – ${msg}`
|
||||
);
|
||||
}
|
||||
if (attempt < MAX_RETRIES) {
|
||||
const delay = RETRY_BASE_DELAY_MS * 2 ** (attempt - 1);
|
||||
@@ -111,11 +119,18 @@ export async function sendAlertWebhook(
|
||||
continue;
|
||||
}
|
||||
|
||||
logger.debug(`Alert webhook sent successfully to "${url}" for event "${context.eventType}" (attempt ${attempt}/${MAX_RETRIES})`);
|
||||
logger.debug(
|
||||
`Alert webhook sent successfully to "${url}" for event "${context.eventType}" (attempt ${attempt}/${MAX_RETRIES})`
|
||||
);
|
||||
return;
|
||||
}
|
||||
|
||||
throw lastError ?? new Error(`Alert webhook: all ${MAX_RETRIES} attempts failed for "${url}"`);
|
||||
throw (
|
||||
lastError ??
|
||||
new Error(
|
||||
`Alert webhook: all ${MAX_RETRIES} attempts failed for "${url}"`
|
||||
)
|
||||
);
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
@@ -139,6 +154,8 @@ function deriveStatus(
|
||||
case "health_check_unhealthy":
|
||||
case "resource_unhealthy":
|
||||
return "unhealthy";
|
||||
case "resource_degraded":
|
||||
return "degraded";
|
||||
case "health_check_toggle":
|
||||
case "resource_toggle":
|
||||
return String(data.status ?? "unknown");
|
||||
@@ -154,7 +171,9 @@ function deriveStatus(
|
||||
// Header construction (mirrors HttpLogDestination.buildHeaders)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
function buildHeaders(webhookConfig: WebhookAlertConfig): Record<string, string> {
|
||||
function buildHeaders(
|
||||
webhookConfig: WebhookAlertConfig
|
||||
): Record<string, string> {
|
||||
const headers: Record<string, string> = {
|
||||
"Content-Type": "application/json"
|
||||
};
|
||||
|
||||
Reference in New Issue
Block a user