Files
kuber/command/up.ts
T

580 lines
17 KiB
TypeScript

import { defineCommand } from "citty";
import { Listr } from "listr2";
import { randomUUID } from "node:crypto";
import type { ComposeSpecification } from "../schema/docker.d";
import type { KuberResource } from "../types";
import {
apiRequest,
KuberApiError,
type ApiRequestInit,
type ApiRequestOptions,
} from "../lib/api";
import {
buildServices,
getRepoRoot,
resolveBuildImages,
type ApiRequester,
} from "../lib/build";
import { composeToKubernetes, type KubernetesResource } from "../lib/convert";
import { ctx } from "../lib/context";
import { getComposePostgresClaims } from "../lib/database";
import { getComposeS3Claims } from "../lib/storage";
import { enumerateWorkspace, type WorkspaceSnapshot } from "../lib/workspace";
import {
requireLocalTrust,
resolveTrustIdentity,
trustHeaders,
} from "../lib/trust";
type Workspace = {
metadata: { name: string; uid: string; resourceVersion: string };
};
type ResourceIdentity = {
apiVersion: string;
kind: string;
name: string;
namespace?: string;
uid: string;
workspaceUid: string;
};
type ResourcePlan = {
desired: KubernetesResource[];
stale: ResourceIdentity[];
};
type OperationStatus = {
state?: unknown;
error?: { code?: unknown; message?: unknown };
};
type OperationResponse = {
operationId?: unknown;
operation?: { status?: OperationStatus };
};
export type OperationResumeOptions = {
now?: () => number;
sleep?: (milliseconds: number) => Promise<void>;
};
type UpContext = {
compose?: ComposeSpecification;
snapshot?: WorkspaceSnapshot;
buildImages?: Record<string, string>;
serviceEnv?: Record<string, Record<string, string>>;
resources?: KubernetesResource[];
plan?: ResourcePlan;
};
export const WORKSPACE_ADOPTION_METHOD = "POST";
const OPERATION_RESUME_INITIAL_BACKOFF_MS = 250;
const OPERATION_RESUME_MAX_BACKOFF_MS = 5_000;
const OPERATION_RESUME_GRACE_MS = 60_000;
const RECOVERABLE_API_ERROR_CODES = new Set([
"HTTP_502",
"HTTP_503",
"HTTP_504",
]);
export function workspaceAdoptionRoute(project: string): string {
return `/workspaces/${encodeURIComponent(project)}/adopt`;
}
export function mergeServiceEnv(
current: Record<string, Record<string, string>> | undefined,
next: Record<string, Record<string, string>>,
): Record<string, Record<string, string>> {
const merged = { ...current };
for (const [service, environment] of Object.entries(next))
merged[service] = { ...merged[service], ...environment };
return merged;
}
export function getDeploymentNames(resources: KubernetesResource[]): string[] {
return resources
.filter((resource) => resource.kind === "Deployment")
.map((resource) => resource.metadata?.name)
.filter((name): name is string => Boolean(name));
}
function workspaceInput(
compose: ComposeSpecification,
snapshot: WorkspaceSnapshot,
) {
return {
source: {
uri: `cas://${snapshot.digest}`,
digest: snapshot.digest,
},
config: { compose },
};
}
export async function ensureWorkspace(
project: string,
compose: ComposeSpecification,
snapshot: WorkspaceSnapshot,
request: ApiRequester = apiRequest,
): Promise<Workspace> {
const path = `/workspaces/${encodeURIComponent(project)}`;
const input = workspaceInput(compose, snapshot);
let current: Workspace;
try {
current = await request<Workspace>(path);
} catch (error) {
if (!(error instanceof KuberApiError) || error.status !== 404) throw error;
return request<Workspace>("/workspaces", {
method: "POST",
json: { id: project, ...input },
});
}
return request<Workspace>(path, {
method: "PUT",
headers: { "if-match": `"${current.metadata.resourceVersion}"` },
json: input,
});
}
function operationEnvironment(
response: Record<string, unknown>,
): Record<string, Record<string, string>> {
return Object.fromEntries(
Object.entries(response).filter(
([key, value]) =>
key !== "operation" &&
key !== "operationId" &&
value !== null &&
typeof value === "object" &&
!Array.isArray(value),
),
) as Record<string, Record<string, string>>;
}
function adoptionHint(project: string, error: unknown): Error {
const message = error instanceof Error ? error.message : String(error);
if (!/namespace .* (?:external|different-workspace)/i.test(message))
return error instanceof Error ? error : new Error(message);
return new Error(
`${message}. Safe adoption requires ${WORKSPACE_ADOPTION_METHOD} ${workspaceAdoptionRoute(project)} with namespace UID and ownership preconditions.`,
{ cause: error },
);
}
async function managementRequest<T>(
project: string,
request: ApiRequester,
path: string,
init: ApiRequestInit,
): Promise<T> {
try {
return await request<T>(path, init);
} catch (error) {
throw adoptionHint(project, error);
}
}
function operationIdFromResponse(response: unknown): string | undefined {
if (!response || typeof response !== "object") return;
const operationId = (response as OperationResponse).operationId;
return typeof operationId === "string" && operationId
? operationId
: undefined;
}
function operationStatusFromResponse(
response: unknown,
): OperationStatus | undefined {
if (!response || typeof response !== "object") return;
const status = (response as OperationResponse).operation?.status;
return status && typeof status === "object" ? status : undefined;
}
function operationFailure(
operationId: string,
status: OperationStatus,
): KuberApiError {
const error = status.error;
const cancelled = status.state === "cancelled";
const message =
typeof error?.message === "string"
? error.message
: cancelled
? "The operation was cancelled"
: "The operation failed";
const code =
typeof error?.code === "string"
? error.code
: cancelled
? "OPERATION_CANCELLED"
: "OPERATION_FAILED";
return new KuberApiError(message, cancelled ? 409 : 500, {
title: cancelled ? "Operation cancelled" : "Operation failed",
status: cancelled ? 409 : 500,
code,
operationId,
});
}
function isRecoverableConnectionInterruption(error: unknown): boolean {
if (error instanceof KuberApiError)
return RECOVERABLE_API_ERROR_CODES.has(error.code);
if (error instanceof DOMException && error.name === "AbortError")
return false;
if (error instanceof Error && error.name === "AbortError") return false;
return (
error instanceof TypeError ||
(error instanceof Error && error.name === "TimeoutError")
);
}
function isInterruptedOperation(status: OperationStatus): boolean {
return (
status.state === "failed" && status.error?.code === "OPERATION_INTERRUPTED"
);
}
function operationResumeDeadline(
rolloutTimeoutMs: number,
now: number,
): number {
return now + Math.max(rolloutTimeoutMs, 0) + OPERATION_RESUME_GRACE_MS;
}
async function resumeManagedOperation(
project: string,
request: ApiRequester,
path: string,
init: ApiRequestInit,
rolloutTimeoutMs: number,
options: OperationResumeOptions,
): Promise<void> {
const now = options.now ?? Date.now;
const sleep = options.sleep ?? ((milliseconds) => Bun.sleep(milliseconds));
const deadline = operationResumeDeadline(rolloutTimeoutMs, now());
const headers = new Headers(init.headers);
headers.set("idempotency-key", randomUUID());
let operationInit = { ...init, headers };
let operationId: string | undefined;
let backoffMs = OPERATION_RESUME_INITIAL_BACKOFF_MS;
let restartRetryPending = false;
for (;;) {
if (restartRetryPending && now() >= deadline)
throw new Error("Timed out while reconnecting to resume the operation");
try {
let status: OperationStatus | undefined;
if (operationId) {
const operation = await managementRequest<{ status: OperationStatus }>(
project,
request,
`/operations/${encodeURIComponent(operationId)}`,
{},
);
status = operation.status;
} else {
restartRetryPending = false;
const response = await managementRequest<OperationResponse>(
project,
request,
path,
operationInit,
);
operationId = operationIdFromResponse(response);
status = operationStatusFromResponse(response);
}
if (!operationId || !status || status.state === "succeeded") return;
if (isInterruptedOperation(status)) {
if (now() >= deadline) throw operationFailure(operationId, status);
operationId = undefined;
restartRetryPending = true;
headers.set("idempotency-key", randomUUID());
operationInit = { ...init, headers };
} else if (status.state === "failed" || status.state === "cancelled")
throw operationFailure(operationId, status);
} catch (error) {
if (
error instanceof KuberApiError &&
error.code === "OPERATION_INTERRUPTED"
) {
if (now() >= deadline) throw adoptionHint(project, error);
operationId = undefined;
restartRetryPending = true;
headers.set("idempotency-key", randomUUID());
operationInit = { ...init, headers };
} else {
if (!isRecoverableConnectionInterruption(error))
throw adoptionHint(project, error);
if (now() >= deadline) throw adoptionHint(project, error);
}
}
const remainingMs = deadline - now();
if (remainingMs <= 0)
throw new Error("Timed out while reconnecting to resume the operation");
await sleep(Math.min(backoffMs, remainingMs));
backoffMs = Math.min(backoffMs * 2, OPERATION_RESUME_MAX_BACKOFF_MS);
}
}
export async function reconcileResources(
project: string,
resources: KubernetesResource[],
rolloutTimeoutMs: number,
postApply:
| ((resources: KubernetesResource[]) => void | Promise<void>)
| undefined,
request: ApiRequester = apiRequest,
resumeOptions: OperationResumeOptions = {},
): Promise<ResourcePlan> {
const workspacePath = `/workspaces/${encodeURIComponent(project)}`;
const plan = await managementRequest<ResourcePlan>(
project,
request,
`${workspacePath}/resources/plan`,
{ method: "POST", json: { resources } },
);
await resumeManagedOperation(
project,
request,
`${workspacePath}/resources/apply`,
{
method: "POST",
json: { resources: plan.desired },
},
rolloutTimeoutMs,
resumeOptions,
);
await postApply?.(plan.desired);
const deployments = getDeploymentNames(plan.desired);
if (deployments.length > 0) {
await resumeManagedOperation(
project,
request,
`${workspacePath}/resources/wait`,
{
method: "POST",
json: { deployments, timeoutMs: rolloutTimeoutMs },
},
rolloutTimeoutMs,
resumeOptions,
);
}
if (plan.stale.length > 0) {
await resumeManagedOperation(
project,
request,
`${workspacePath}/resources/delete`,
{
method: "POST",
json: { resources: plan.stale },
},
rolloutTimeoutMs,
resumeOptions,
);
}
return plan;
}
export async function runUp(
build: boolean,
request: ApiRequester = apiRequest,
) {
const { project, compose, cwd, config, hookContext: getHookContext } = ctx();
const trusted = await requireLocalTrust(
await resolveTrustIdentity(project, cwd),
);
const baseRequest = request;
request = async <T>(
path: string,
init: ApiRequestInit = {},
options?: ApiRequestOptions,
) => {
const headers = new Headers(init.headers);
for (const [key, value] of Object.entries(trustHeaders(trusted)))
headers.set(key, value);
return baseRequest<T>(path, { ...init, headers }, options);
};
const workspacePath = `/workspaces/${encodeURIComponent(project)}`;
const taskCtx = await new Listr<UpContext>(
[
{
title: "Read compose",
task: async (taskCtx, task) => {
taskCtx.compose = await compose();
if (build)
await config.preBuild?.(taskCtx.compose, await getHookContext());
task.output = `${Object.keys(taskCtx.compose.services ?? {}).length} services`;
},
},
{
title: "Snapshot workspace",
task: async (taskCtx, task) => {
taskCtx.snapshot = await enumerateWorkspace(await getRepoRoot(cwd));
task.output = `${taskCtx.snapshot.manifest.files.length} files`;
},
},
{
title: "Build images",
rendererOptions: { outputBar: 10, persistentOutput: true },
enabled: async () =>
build &&
Object.values((await compose()).services ?? {}).some(
(service) => service.build,
),
task: async (taskCtx, task) => {
const result = await buildServices(
project,
taskCtx.compose!,
cwd,
{
progress: (message) => {
task.output = message;
},
stream: task.stdout(),
},
{ ...config, request, snapshot: taskCtx.snapshot },
);
taskCtx.buildImages = result.images;
await config.postBuild?.(result, await getHookContext());
task.output = `Built ${result.built.length} image${result.built.length === 1 ? "" : "s"}`;
},
},
{
title: "Resolve images",
enabled: async () =>
!build &&
Object.values((await compose()).services ?? {}).some(
(service) => service.build,
),
task: async (taskCtx, task) => {
taskCtx.buildImages = await resolveBuildImages(
project,
taskCtx.compose!,
{
...config,
request,
},
);
task.output = `${Object.keys(taskCtx.buildImages).length} images`;
},
},
{
title: "Update workspace",
task: async (taskCtx, task) => {
const workspace = await ensureWorkspace(
project,
taskCtx.compose!,
taskCtx.snapshot!,
request,
);
const adopted = await request<{ resourcesAdopted: number }>(
workspaceAdoptionRoute(project),
{
method: WORKSPACE_ADOPTION_METHOD,
json: { workspaceUid: workspace.metadata.uid },
},
);
task.output = adopted.resourcesAdopted
? `Adopted ${adopted.resourcesAdopted} existing resources`
: "Workspace ready";
},
},
{
title: "Reconcile databases",
enabled: async () =>
getComposePostgresClaims(await compose()).length > 0,
task: async (taskCtx, task) => {
const response = await managementRequest<Record<string, unknown>>(
project,
request,
`${workspacePath}/databases`,
{ method: "POST", json: { compose: taskCtx.compose } },
);
taskCtx.serviceEnv = mergeServiceEnv(
taskCtx.serviceEnv,
operationEnvironment(response),
);
task.output = `${Object.keys(taskCtx.serviceEnv).length} services`;
},
},
{
title: "Reconcile S3 storage",
enabled: async () => getComposeS3Claims(await compose()).length > 0,
task: async (taskCtx, task) => {
const response = await managementRequest<Record<string, unknown>>(
project,
request,
`${workspacePath}/storage`,
{ method: "POST", json: { compose: taskCtx.compose } },
);
taskCtx.serviceEnv = mergeServiceEnv(
taskCtx.serviceEnv,
operationEnvironment(response),
);
task.output = `${Object.keys(taskCtx.serviceEnv).length} services`;
},
},
{
title: "Render manifests",
task: async (taskCtx, task) => {
taskCtx.resources = await composeToKubernetes(
project,
taskCtx.compose!,
cwd,
taskCtx.serviceEnv,
taskCtx.buildImages,
);
await config.postRender?.(
taskCtx.resources as KuberResource[],
await getHookContext(),
);
task.output = `${taskCtx.resources.length} resources`;
},
},
{
title: "Reconcile resources",
task: async (taskCtx, task) => {
taskCtx.plan = await reconcileResources(
project,
taskCtx.resources!,
config.rolloutTimeoutMs,
config.postApply
? async (resources) =>
config.postApply?.(
resources as KuberResource[],
await getHookContext(),
)
: undefined,
request,
);
task.output = `${taskCtx.plan.desired.length} applied, ${taskCtx.plan.stale.length} stale deleted`;
},
},
],
{ rendererOptions: { collapseErrors: false } },
).run();
return taskCtx;
}
export const up = defineCommand({
meta: { name: "up", description: "Create and start deployments" },
args: {
build: {
type: "boolean",
default: true,
alias: "b",
description: "Build images before starting deployments",
negativeDescription: "Don't build an image, even if it's policy",
},
},
async run({ args }) {
await runUp(args.build);
},
});