chore: simplify honeycomb alerts (#26142)
This commit is contained in:
parent
293bb422fa
commit
f8aa4a3be0
15 changed files with 185 additions and 408 deletions
|
|
@ -1,5 +1,6 @@
|
|||
import { domain } from "./stage"
|
||||
import { EMAILOCTOPUS_API_KEY } from "./app"
|
||||
import { SECRET } from "./secret"
|
||||
|
||||
////////////////
|
||||
// DATABASE
|
||||
|
|
@ -221,8 +222,6 @@ const AUTH_API_URL = new sst.Linkable("AUTH_API_URL", {
|
|||
const STRIPE_WEBHOOK_SECRET = new sst.Linkable("STRIPE_WEBHOOK_SECRET", {
|
||||
properties: { value: stripeWebhook.secret },
|
||||
})
|
||||
const INCIDENT_WEBHOOK_SIGNING_SECRET = new sst.Secret("INCIDENT_WEBHOOK_SIGNING_SECRET")
|
||||
const DISCORD_INCIDENT_WEBHOOK_URL = new sst.Secret("DISCORD_INCIDENT_WEBHOOK_URL")
|
||||
|
||||
const gatewayKv = new sst.cloudflare.Kv("GatewayKv")
|
||||
|
||||
|
|
@ -233,6 +232,7 @@ const gatewayKv = new sst.cloudflare.Kv("GatewayKv")
|
|||
const bucket = new sst.cloudflare.Bucket("ZenData")
|
||||
const bucketNew = new sst.cloudflare.Bucket("ZenDataNew")
|
||||
|
||||
const DISCORD_INCIDENT_WEBHOOK_URL = new sst.Secret("DISCORD_INCIDENT_WEBHOOK_URL")
|
||||
const AWS_SES_ACCESS_KEY_ID = new sst.Secret("AWS_SES_ACCESS_KEY_ID")
|
||||
const AWS_SES_SECRET_ACCESS_KEY = new sst.Secret("AWS_SES_SECRET_ACCESS_KEY")
|
||||
|
||||
|
|
@ -254,8 +254,8 @@ new sst.cloudflare.x.SolidStart("Console", {
|
|||
database,
|
||||
AUTH_API_URL,
|
||||
STRIPE_WEBHOOK_SECRET,
|
||||
INCIDENT_WEBHOOK_SIGNING_SECRET,
|
||||
DISCORD_INCIDENT_WEBHOOK_URL,
|
||||
SECRET.HoneycombWebhookSecret,
|
||||
STRIPE_SECRET_KEY,
|
||||
EMAILOCTOPUS_API_KEY,
|
||||
AWS_SES_ACCESS_KEY_ID,
|
||||
|
|
|
|||
|
|
@ -1,318 +1,91 @@
|
|||
const displayName = (s: string) =>
|
||||
s
|
||||
.split("-")
|
||||
.map((w) => w.charAt(0).toUpperCase() + w.slice(1))
|
||||
.join(" ")
|
||||
.replace(/(?<=\d) (?=\d)/g, ".")
|
||||
import { SECRET } from "./secret"
|
||||
import { domain } from "./stage"
|
||||
|
||||
const resourceName = (s: string) => displayName(s).replace(/[^a-zA-Z0-9]/g, "")
|
||||
|
||||
const varSpec = (label: string, name: string) =>
|
||||
$jsonStringify({
|
||||
content: [
|
||||
{
|
||||
content: [
|
||||
{
|
||||
attrs: {
|
||||
name,
|
||||
label,
|
||||
missing: false,
|
||||
},
|
||||
type: "varSpec",
|
||||
},
|
||||
],
|
||||
type: "paragraph",
|
||||
},
|
||||
],
|
||||
type: "doc",
|
||||
})
|
||||
|
||||
const fields = {
|
||||
model: incident.getAlertAttributeOutput({ name: "Model" }),
|
||||
product: incident.getAlertAttributeOutput({ name: "Product" }),
|
||||
}
|
||||
|
||||
const alertSource = new incident.AlertSource("HoneycombAlertSource", {
|
||||
name: $app.stage === "production" ? "Honeycomb" : `Honeycomb (${$app.stage})`,
|
||||
sourceType: "honeycomb",
|
||||
template: {
|
||||
title: {
|
||||
literal: varSpec("Payload -> Title", "title"),
|
||||
},
|
||||
description: {
|
||||
literal: varSpec("Payload -> Description", "description"),
|
||||
},
|
||||
attributes: [
|
||||
{
|
||||
alertAttributeId: fields.model.id,
|
||||
binding: {
|
||||
value: {
|
||||
reference: 'expressions["model"]',
|
||||
},
|
||||
mergeStrategy: "first_wins",
|
||||
},
|
||||
},
|
||||
{
|
||||
alertAttributeId: fields.product.id,
|
||||
binding: {
|
||||
value: {
|
||||
reference: 'expressions["product"]',
|
||||
},
|
||||
mergeStrategy: "first_wins",
|
||||
},
|
||||
},
|
||||
],
|
||||
expressions: [
|
||||
{
|
||||
label: "Model",
|
||||
operations: [
|
||||
{
|
||||
operationType: "parse",
|
||||
parse: {
|
||||
returns: {
|
||||
array: false,
|
||||
type: fields.model.type,
|
||||
},
|
||||
source: "$['model']",
|
||||
},
|
||||
},
|
||||
],
|
||||
reference: "model",
|
||||
rootReference: "payload",
|
||||
},
|
||||
{
|
||||
label: "Product",
|
||||
operations: [
|
||||
{
|
||||
operationType: "parse",
|
||||
parse: {
|
||||
returns: {
|
||||
array: false,
|
||||
type: fields.product.type,
|
||||
},
|
||||
source: "$['product']",
|
||||
},
|
||||
},
|
||||
],
|
||||
reference: "product",
|
||||
rootReference: "payload",
|
||||
},
|
||||
],
|
||||
},
|
||||
})
|
||||
|
||||
const webhookRecipient = new honeycomb.WebhookRecipient(`IncidentWebhook`, {
|
||||
name: $app.stage === "production" ? "Incident.io" : `Incident.io (${$app.stage})`,
|
||||
url: alertSource.alertEventsUrl,
|
||||
secret: alertSource.secretToken,
|
||||
const webhookRecipient = new honeycomb.WebhookRecipient("DiscordAlerts", {
|
||||
name: $app.stage === "production" ? "Discord Alerts" : `Discord Alerts (${$app.stage})`,
|
||||
url: `https://${domain}/honeycomb/webhook`,
|
||||
secret: SECRET.HoneycombWebhookSecret.result,
|
||||
templates: [
|
||||
{
|
||||
type: "trigger",
|
||||
body: $jsonStringify({
|
||||
title: "{{ .Name }}",
|
||||
description: "{{ .Description }}",
|
||||
status: "{{ .Alert.Status }}",
|
||||
deduplication_key: "{{ .Alert.InstanceID }}",
|
||||
source_url: "{{ .Result.URL }}",
|
||||
model: "{{ .Vars.model }}",
|
||||
product: "{{ .Vars.product }}",
|
||||
}),
|
||||
body: `{
|
||||
"url": {{ .Result.URL | quote }},
|
||||
"type": {{ .Vars.type | quote }},
|
||||
"name": {{ .Name | quote }},
|
||||
"status": {{ .Alert.Status | quote }},
|
||||
"isTest": {{ .Alert.IsTest }},
|
||||
"groups": {{ .Result.GroupsTriggered | toJson }}
|
||||
}`,
|
||||
},
|
||||
],
|
||||
variables: [
|
||||
{
|
||||
name: "model",
|
||||
},
|
||||
{
|
||||
name: "product",
|
||||
name: "type",
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
new incident.AlertRoute("HoneycombAlertRoute", {
|
||||
name: $app.stage === "production" ? "Honeycomb" : `Honeycomb (${$app.stage})`,
|
||||
enabled: true,
|
||||
isPrivate: false,
|
||||
alertSources: [
|
||||
{
|
||||
alertSourceId: alertSource.id,
|
||||
conditionGroups: [
|
||||
{
|
||||
conditions: [
|
||||
{
|
||||
subject: "alert.title",
|
||||
operation: "is_set",
|
||||
paramBindings: [],
|
||||
},
|
||||
],
|
||||
},
|
||||
],
|
||||
},
|
||||
],
|
||||
conditionGroups: [
|
||||
{
|
||||
conditions: [
|
||||
{
|
||||
subject: "alert.title",
|
||||
operation: "is_set",
|
||||
paramBindings: [],
|
||||
},
|
||||
],
|
||||
},
|
||||
],
|
||||
expressions: [],
|
||||
escalationConfig: {
|
||||
autoCancelEscalations: true,
|
||||
escalationTargets: [],
|
||||
},
|
||||
incidentConfig: {
|
||||
autoDeclineEnabled: true,
|
||||
enabled: true,
|
||||
conditionGroups: [],
|
||||
deferTimeSeconds: 0,
|
||||
groupingKeys: [
|
||||
const modelHttpErrorsQuery = (product: "go" | "zen") => {
|
||||
const filters = [
|
||||
{ column: "model", op: "exists" },
|
||||
{ column: "event_type", op: "=", value: "completions" },
|
||||
{ column: "user_agent", op: "contains", value: "opencode" },
|
||||
{ column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
|
||||
]
|
||||
|
||||
return honeycomb.getQuerySpecificationOutput({
|
||||
breakdowns: ["model"],
|
||||
calculatedFields: [
|
||||
{
|
||||
reference: $interpolate`alert.attributes.${fields.model.id}`,
|
||||
},
|
||||
{
|
||||
reference: $interpolate`alert.attributes.${fields.product.id}`,
|
||||
name: "is_failed_http_status",
|
||||
expression: `IF(AND(GTE($status, "400"), NOT(EQUALS($status, "401"))), 1, 0)`,
|
||||
},
|
||||
],
|
||||
groupingWindowSeconds: 3600,
|
||||
},
|
||||
incidentTemplate: {
|
||||
name: {
|
||||
value: {
|
||||
literal: varSpec("Alert -> Title", "alert.title"),
|
||||
},
|
||||
},
|
||||
summary: {
|
||||
value: {
|
||||
literal: varSpec("Alert -> Description", "alert.description"),
|
||||
},
|
||||
},
|
||||
startInTriage: {
|
||||
value: {
|
||||
literal: "true",
|
||||
},
|
||||
},
|
||||
severity: {
|
||||
mergeStrategy: "first-wins",
|
||||
},
|
||||
incidentMode: {
|
||||
value: {
|
||||
literal: $app.stage === "production" ? "standard" : "test",
|
||||
},
|
||||
},
|
||||
},
|
||||
})
|
||||
|
||||
type Product = "go" | "zen"
|
||||
|
||||
type Trigger = (opts: { model: string; product: Product }) => {
|
||||
id: string
|
||||
title: string
|
||||
description: string
|
||||
json: honeycomb.GetQuerySpecificationOutputArgs
|
||||
threshold: { op: ">=" | "<="; value: number }
|
||||
}
|
||||
|
||||
type Model = { id: string; products: Product[]; triggers: Trigger[] }
|
||||
|
||||
const httpErrors: Trigger = ({ model, product }) => ({
|
||||
id: "increased-http-errors",
|
||||
title: `Increased HTTP Errors for ${displayName(model)} on ${displayName(product)}`,
|
||||
description: `Detected increased rate of HTTP errors for ${displayName(model)} on OpenCode ${displayName(product)}`,
|
||||
json: {
|
||||
calculations: [
|
||||
{
|
||||
op: "COUNT",
|
||||
name: "TOTAL",
|
||||
filterCombination: "AND",
|
||||
filters: [
|
||||
{ column: "model", op: "=", value: model },
|
||||
{ column: "event_type", op: "=", value: "completions" },
|
||||
{ column: "user_agent", op: "contains", value: "opencode" },
|
||||
{ column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
|
||||
],
|
||||
},
|
||||
{
|
||||
op: "COUNT",
|
||||
name: "FAILED",
|
||||
filterCombination: "AND",
|
||||
filters: [
|
||||
{ column: "model", op: "=", value: model },
|
||||
{ column: "event_type", op: "=", value: "completions" },
|
||||
{ column: "user_agent", op: "contains", value: "opencode" },
|
||||
{ column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
|
||||
{ column: "status", op: ">=", value: "400" },
|
||||
{ column: "status", op: "!=", value: "401" },
|
||||
],
|
||||
},
|
||||
{ op: "COUNT", name: "TOTAL", filterCombination: "AND", filters },
|
||||
{ op: "SUM", name: "FAILED", column: "is_failed_http_status", filterCombination: "AND", filters },
|
||||
],
|
||||
formulas: [{ name: "ERROR", expression: "$FAILED / $TOTAL" }],
|
||||
formulas: [{ name: "ERROR", expression: "IF(GTE($TOTAL, 2500), DIV($FAILED, $TOTAL), 0)" }],
|
||||
timeRange: 900,
|
||||
},
|
||||
threshold: { op: ">=", value: 0.8 },
|
||||
}).json
|
||||
}
|
||||
|
||||
const description = "Managed by SST (Don't edit in Honeycomb UI)"
|
||||
|
||||
new honeycomb.Trigger("IncreasedModelHttpErrorsGo", {
|
||||
name: "Increased Model HTTP Errors [Go]",
|
||||
description,
|
||||
queryJson: modelHttpErrorsQuery("go"),
|
||||
alertType: "on_change",
|
||||
frequency: 300,
|
||||
thresholds: [{ op: ">=", value: 0.8, exceededLimit: 1 }],
|
||||
recipients: [
|
||||
// {
|
||||
// id: webhookRecipient.id,
|
||||
// notificationDetails: [
|
||||
// {
|
||||
// variables: [{ name: "type", value: "model_http_errors" }],
|
||||
// },
|
||||
// ],
|
||||
// },
|
||||
],
|
||||
})
|
||||
|
||||
const models: Model[] = [
|
||||
{ id: "kimi-k2.6", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "kimi-k2.5", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "deepseek-v4-flash", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "deepseek-v4-pro", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "glm-5.1", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
// { id: "glm-5", products: ["go"], triggers: [httpErrors] },
|
||||
{ id: "qwen3.6-plus", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "qwen3.5-plus", products: ["go"], triggers: [httpErrors] },
|
||||
{ id: "minimax-m2.7", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
// { id: "minimax-m2.5", products: ["go", "zen"], triggers: [httpErrors] },
|
||||
{ id: "mimo-v2.5-pro", products: ["go"], triggers: [httpErrors] },
|
||||
// { id: "mimo-v2.5", products: ["go"], triggers: [httpErrors] },
|
||||
// { id: "mimo-v2-omni", products: ["go"], triggers: [httpErrors] },
|
||||
// { id: "mimo-v2-pro", products: ["go"], triggers: [httpErrors] },
|
||||
{ id: "claude-opus-4-7", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "claude-opus-4-6", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "claude-sonnet-4-6", products: ["zen"], triggers: [httpErrors] },
|
||||
{ id: "gpt-5.5", products: ["zen"], triggers: [httpErrors] },
|
||||
{ id: "big-pickle", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "minimax-m2.5-free", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "hy3-preview-free", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "nemotron-3-super-free", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "trinity-large-preview-free", products: ["zen"], triggers: [httpErrors] },
|
||||
// { id: "ling-2.6-flash-free", products: ["zen"], triggers: [httpErrors] },
|
||||
]
|
||||
|
||||
if ($app.stage !== "production") {
|
||||
models.splice(1)
|
||||
}
|
||||
|
||||
for (const model of models) {
|
||||
for (const product of model.products) {
|
||||
for (const trigger of model.triggers) {
|
||||
const spec = trigger({ model: model.id, product })
|
||||
|
||||
new honeycomb.Trigger(resourceName(`${spec.id}-${product}-${model.id}`), {
|
||||
name: spec.title,
|
||||
description: spec.description,
|
||||
queryJson: honeycomb.getQuerySpecificationOutput(spec.json).json,
|
||||
alertType: "on_change",
|
||||
frequency: 300,
|
||||
thresholds: [{ ...spec.threshold, exceededLimit: 1 }],
|
||||
recipients: [
|
||||
{
|
||||
id: webhookRecipient.id,
|
||||
notificationDetails: [
|
||||
{
|
||||
variables: [
|
||||
{ name: "model", value: model.id },
|
||||
{ name: "product", value: product },
|
||||
],
|
||||
},
|
||||
],
|
||||
},
|
||||
],
|
||||
})
|
||||
}
|
||||
}
|
||||
}
|
||||
new honeycomb.Trigger("IncreasedModelHttpErrorsZen", {
|
||||
name: "Increased Model HTTP Errors [Zen]",
|
||||
description,
|
||||
queryJson: modelHttpErrorsQuery("zen"),
|
||||
alertType: "on_change",
|
||||
frequency: 300,
|
||||
thresholds: [{ op: ">=", value: 0.8, exceededLimit: 1 }],
|
||||
recipients: [
|
||||
// {
|
||||
// id: webhookRecipient.id,
|
||||
// notificationDetails: [
|
||||
// {
|
||||
// variables: [{ name: "type", value: "model_http_errors" }],
|
||||
// },
|
||||
// ],
|
||||
// },
|
||||
],
|
||||
})
|
||||
|
|
|
|||
|
|
@ -1,4 +1,11 @@
|
|||
sst.Linkable.wrap(random.RandomPassword, (resource) => ({
|
||||
properties: {
|
||||
value: resource.result,
|
||||
},
|
||||
}))
|
||||
|
||||
export const SECRET = {
|
||||
R2AccessKey: new sst.Secret("R2AccessKey", "unknown"),
|
||||
R2SecretKey: new sst.Secret("R2SecretKey", "unknown"),
|
||||
HoneycombWebhookSecret: new random.RandomPassword("HoneycombWebhookSecret", { length: 24 }),
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue