feat: prepare 2.6.1-rc5 shared databases and build SSE

This commit is contained in:
2026-10-05 11:09:52 +00:00 Unverified
parent f76165603e
commit 828bf3a328
25 changed files with 2557 additions and 196 deletions
+113 -9
View File
@@ -1,9 +1,10 @@
import type { KubernetesObject } from "@kubernetes/client-node";
import { randomUUID } from "node:crypto";
import type { ComposeSpecification } from "../schema/docker.d";
import { DatabaseReconciliationError } from "../lib/database";
import { DATABASE_RECONCILE_PHASES, DatabaseReconciliationError } from "../lib/database";
import type { Operation as PublicOperation } from "../shared/api";
import type { BuildRequest, Sha256Digest } from "../shared/build-protocol";
import { KUBER_VERSION, KUBER_VERSION_HEADER } from "../shared/version";
import {
createToken,
hashToken,
@@ -28,6 +29,7 @@ import {
BuildValidationError,
type BuildController,
} from "./build-controller";
import { BuildEventStreamHub } from "./build-event-stream";
import { KubernetesLogError, type LogService } from "./log-service";
import {
ExecService,
@@ -142,6 +144,11 @@ export type UnknownFailureLog = {
export type RequestErrorLog = Omit<UnknownFailureLog, "event"> & {
event: string;
operationId?: string;
phase?: string;
errorClass?: string;
providerStatus?: number;
providerCode?: string;
topFrame?: string;
};
export interface AppLogger {
@@ -200,6 +207,62 @@ class HttpError extends Error {
}
}
const DATABASE_PHASES = new Set<string>(DATABASE_RECONCILE_PHASES);
const DATABASE_ERROR_CLASSES = new Set([
"Error", "TypeError", "RangeError", "SyntaxError", "AbortError",
"ApiException", "ResponseError", "FetchError", "TimeoutError",
]);
const DATABASE_ERROR_CODES = new Set([
"ECONNREFUSED", "ECONNRESET", "ETIMEDOUT", "EHOSTUNREACH",
"ENOTFOUND", "EAI_AGAIN", "ABORT_ERR", "UND_ERR_CONNECT_TIMEOUT",
]);
/** Never serialize an unknown error: its message, stack and provider fields may contain secrets. */
function databaseFailureDiagnostics(error: unknown): Record<string, unknown> {
let current = error;
let phase = "operation execution";
const seen = new Set<unknown>();
for (
let depth = 0;
depth < 5 && current && typeof current === "object" && !seen.has(current);
depth++
) {
seen.add(current);
if (
current instanceof DatabaseReconciliationError &&
DATABASE_PHASES.has(current.phase)
) {
phase = current.phase;
}
const cause = (current as { cause?: unknown }).cause;
if (!cause || typeof cause !== "object" || seen.has(cause)) break;
current = cause;
}
const source = current && typeof current === "object"
? current as Record<string, unknown> : {};
const errorClass = current instanceof Error && DATABASE_ERROR_CLASSES.has(current.name)
? current.name
: "UnknownError";
const status = [source.statusCode, source.status, source.code].find(
(value) => typeof value === "number" && Number.isInteger(value) &&
value >= 400 && value < 600,
);
const errorCode = typeof source.code === "string" && DATABASE_ERROR_CODES.has(source.code)
? source.code
: undefined;
// Only the first frame and a fixed set of our source files; never log raw stacks or paths.
const firstFrame = current instanceof Error ? current.stack?.split("\n")[1] : undefined;
const frame = firstFrame &&
/(?:^|\/)\b((?:lib\/database|server\/management|server\/app)\.ts):(\d+):(\d+)\b/.exec(firstFrame);
return {
phase,
errorClass,
...(status !== undefined && { providerStatus: status }),
...(errorCode && { providerCode: errorCode }),
...(frame && { topFrame: `${frame[1]}:${frame[2]}:${frame[3]}` }),
};
}
function isRecord(value: unknown): value is Record<string, unknown> {
return typeof value === "object" && value !== null && !Array.isArray(value);
}
@@ -282,6 +345,9 @@ export function createApp(
const now = options.now ?? Date.now;
const makeRequestId = options.requestId ?? randomUUID;
const logger = options.logger ?? defaultAppLogger;
const buildEventStreams = options.builds
? new BuildEventStreamHub(options.builds)
: undefined;
const bodyLimit = options.jsonBodyLimit ?? DEFAULT_JSON_LIMIT;
const allowedOrigins = new Set(options.allowedOrigins ?? []);
const hasOriginConfiguration = options.allowedOrigins !== undefined;
@@ -1150,25 +1216,43 @@ export function createApp(
});
} catch (error) {
if (!operationStarted) throw error;
const message = error instanceof Error ? error.message : String(error);
const leaseLost =
leaseOwnershipLost ||
(error instanceof HttpError && error.code === "WORKSPACE_LEASE_LOST");
const databaseFailure =
action === "databases.reconcile" && !leaseLost
? error instanceof DatabaseReconciliationError
? error
: new DatabaseReconciliationError("operation execution", error)
: undefined;
if (databaseFailure) {
logger.error({
event: "operation.database_reconcile.failed",
requestId: requestIds.get(request) ?? makeRequestId(),
method: "POST",
pathname: "/api/v2/workspaces/:workspaceId/databases",
operationId: operation.metadata.name,
status: 500,
code: "DATABASE_RECONCILE_FAILED",
errorName: "DatabaseReconciliationError",
message: "Database reconciliation failed",
...databaseFailureDiagnostics(databaseFailure),
});
}
const message =
databaseFailure?.message ??
(error instanceof Error ? error.message : String(error));
const failed = await transitionOperationToFailure(
operation.metadata.name,
{
code: leaseLost
? "WORKSPACE_LEASE_LOST"
: action === "databases.reconcile" &&
error instanceof DatabaseReconciliationError
: databaseFailure
? "DATABASE_RECONCILE_FAILED"
: "OPERATION_FAILED",
message: leaseLost
? "Workspace operation lease ownership was lost"
: action === "databases.reconcile" &&
!(error instanceof DatabaseReconciliationError)
? "Database reconciliation failed"
: message,
: message,
},
);
const failure = failed.status.error;
@@ -1470,13 +1554,31 @@ export function createApp(
);
if (!action && request.method === "GET")
return response(await builds.getBuildStatus(id));
if (action === "events" && request.method === "GET")
if (action === "events" && request.method === "GET") {
if (request.headers.get("accept")?.split(",").some((value) =>
value.trim().split(";")[0]?.trim().toLowerCase() === "text/event-stream",
)) {
const cursor = request.headers.get("last-event-id") ?? url.searchParams.get("after");
if (
cursor !== null &&
(!/^(0|[1-9]\d*)$/.test(cursor) ||
!Number.isSafeInteger(Number(cursor)))
)
throw new HttpError(
400,
"Invalid cursor",
"INVALID_QUERY",
"Event cursor must be a non-negative safe integer",
);
return buildEventStreams!.open(id, request);
}
return response(
await builds.getBuildEvents(
id,
nonNegativeInteger(url.searchParams.get("after"), "after") ?? 0,
),
);
}
if (action === "reconcile" && request.method === "POST")
return response(await builds.reconcileBuild(id));
if (action === "cancel" && request.method === "POST")
@@ -2553,6 +2655,7 @@ export function createApp(
result = problem(normalized, requestId);
}
result.headers.set("x-request-id", requestId);
result.headers.set(KUBER_VERSION_HEADER, KUBER_VERSION);
if (origin && originAllowed) {
result.headers.set("access-control-allow-origin", origin);
result.headers.set("vary", "Origin");
@@ -2689,6 +2792,7 @@ export function execProblem(error: unknown, requestId = ""): Response {
},
);
result.headers.set("content-type", "application/problem+json");
result.headers.set(KUBER_VERSION_HEADER, KUBER_VERSION);
return result;
}
+255
View File
@@ -0,0 +1,255 @@
import type { BuildController } from "./build-controller";
import type { BuildStatus } from "../shared/build-protocol";
const terminal = (status: BuildStatus) =>
status.state === "succeeded" || status.state === "failed";
function cursor(value: string | null): number {
if (value === null) return 0;
if (!/^(0|[1-9]\d*)$/.test(value) || !Number.isSafeInteger(Number(value)))
throw new RangeError("Event cursor must be a non-negative safe integer");
return Number(value);
}
type Subscriber = {
sequence: number;
status?: string;
lastActivity: number;
controller: ReadableStreamDefaultController<Uint8Array>;
stop: () => void;
};
type Group = {
members: Set<Subscriber>;
abort: AbortController;
nextReconcileAt: number;
reconciling: boolean;
};
/** One hub per app/controller instance; call only after the route's capability and workspace checks.
* The authenticated GET /api/v2/builds/:id/events route should branch on
* Accept: text/event-stream, then return hub.open(id, request); retain JSON otherwise.
* A hub coalesces reconciliation and persisted event reads per physical build on this
* replica. BuildController's lease coordinates reconciliation across replicas; its
* persisted event store supplies logs even when a different replica owns the job.
*/
export class BuildEventStreamHub {
private readonly builds = new Map<string, Group>();
constructor(
private readonly controller: Pick<
BuildController,
"getBuildStatus" | "getBuildEvents" | "reconcileBuild"
>,
private readonly options: {
pollMs?: number;
heartbeatMs?: number;
reconcileMs?: number;
} = {},
) {}
async open(id: string, request: Request): Promise<Response> {
const header = request.headers.get("last-event-id");
const query = new URL(request.url).searchParams.get("after");
const after = cursor(header ?? query);
// Validate existence before committing response headers.
await this.controller.getBuildStatus(id);
let subscriber: Subscriber;
const stream = new ReadableStream<Uint8Array>(
{
start: (controller) => {
let group: Group;
const stop = () => {
request.signal.removeEventListener("abort", stop);
group.members.delete(subscriber);
if (group.members.size === 0 && this.builds.get(id) === group) {
this.builds.delete(id);
group.abort.abort();
}
try {
controller.close();
} catch {
/* Already closed or cancelled. */
}
};
subscriber = {
sequence: after,
lastActivity: Date.now(),
controller,
stop,
};
const existing = this.builds.get(id);
if (existing) {
group = existing;
group.members.add(subscriber);
} else {
group = {
members: new Set([subscriber]),
abort: new AbortController(),
nextReconcileAt: 0,
reconciling: false,
};
this.builds.set(id, group);
void this.run(id, group);
}
request.signal.addEventListener("abort", stop, { once: true });
if (request.signal.aborted) stop();
},
cancel: () => subscriber.stop(),
},
{ highWaterMark: 16 },
);
return new Response(stream, {
headers: {
"content-type": "text/event-stream; charset=utf-8",
"cache-control": "no-cache, no-transform",
connection: "keep-alive",
"x-accel-buffering": "no",
},
});
}
private async run(id: string, group: Group): Promise<void> {
const encoder = new TextEncoder();
const pollMs = this.options.pollMs ?? 1_000;
const reconcileMs = this.options.reconcileMs ?? 7_000;
const { members } = group;
// Bun's default HTTP idle timeout is 10 seconds; keep bytes flowing well
// inside it even while Kubernetes reconciliation makes no progress.
const heartbeatMs = this.options.heartbeatMs ?? 5_000;
const heartbeat = setInterval(() => {
if (group.abort.signal.aborted) return;
for (const member of members) {
if (Date.now() - member.lastActivity < heartbeatMs) continue;
if ((member.controller.desiredSize ?? 0) <= 0) member.stop();
else {
member.controller.enqueue(encoder.encode(": heartbeat\n\n"));
member.lastActivity = Date.now();
}
}
}, heartbeatMs);
heartbeat.unref?.();
try {
while (!group.abort.signal.aborted) {
try {
const initial = await this.controller.getBuildStatus(id);
if (group.abort.signal.aborted) break;
// Reconciliation is shared per build and never blocks persisted progress.
// The controller's lease still coordinates attempts across replicas.
if (
!terminal(initial) &&
!group.reconciling &&
Date.now() >= group.nextReconcileAt
) {
group.reconciling = true;
group.nextReconcileAt = Date.now() + reconcileMs;
const signal = group.abort.signal;
let onAbort = () => {};
const aborted = new Promise<void>((resolve) => {
onAbort = resolve;
signal.addEventListener("abort", onAbort, { once: true });
});
const reconciliation = this.controller
.reconcileBuild(id, { signal })
.catch(() => {
// Another replica may hold the lease; keep observing its progress.
});
// Drop this group's references even if an external API ignores abort
// and leaves the reconciliation promise unresolved indefinitely.
void Promise.race([reconciliation, aborted]).finally(() => {
signal.removeEventListener("abort", onAbort);
group.reconciling = false;
});
}
const status = await this.controller.getBuildStatus(id);
if (group.abort.signal.aborted) break;
const oldest = Math.min(
...[...members].map((member) => member.sequence),
);
const events = await this.controller.getBuildEvents(id, oldest);
if (group.abort.signal.aborted) break;
const firstLogSequence = events.find(
(event) => event.type === "log",
)?.sequence;
for (const member of members) {
const send = (frame: string) => {
// A slow consumer reconnects from its last delivered sequence instead
// of retaining an unbounded in-memory backlog on this replica.
if ((member.controller.desiredSize ?? 0) <= 0) {
member.stop();
return false;
}
member.controller.enqueue(encoder.encode(frame));
member.lastActivity = Date.now();
return true;
};
if (
firstLogSequence !== undefined &&
firstLogSequence > member.sequence + 1
) {
const gap = {
type: "gap",
after: member.sequence,
before: firstLogSequence,
missing: firstLogSequence - member.sequence - 1,
message: `Build log history was trimmed; ${firstLogSequence - member.sequence - 1} log event(s) before sequence ${firstLogSequence} are unavailable.`,
};
if (
!send(
`id: ${firstLogSequence - 1}\nevent: gap\ndata: ${JSON.stringify(gap)}\n\n`,
)
)
continue;
// Advance over the unavailable range so reconnects do not repeat
// the marker. The cursor now refers to the last unavailable
// sequence; delivered log records retain their normal IDs.
member.sequence = firstLogSequence - 1;
}
for (const event of events) {
if (event.type !== "log" || event.sequence <= member.sequence)
continue;
if (
!send(
`id: ${event.sequence}\nevent: log\ndata: ${JSON.stringify(event)}\n\n`,
)
)
break;
member.sequence = event.sequence;
}
if (!members.has(member)) continue;
const fingerprint = JSON.stringify(status);
if (member.status !== fingerprint) {
if (
!send(
`event: status\ndata: ${JSON.stringify({ type: "status", status })}\n\n`,
)
)
continue;
member.status = fingerprint;
}
if (terminal(status)) member.stop();
}
} catch {
// A transient API/store failure closes streams so clients reconnect with
// their last sequence; no silent terminal state is fabricated.
for (const member of members) member.stop();
}
if (group.abort.signal.aborted) break;
await new Promise<void>((resolve) => {
const stop = () => {
clearTimeout(timer);
resolve();
};
const timer = setTimeout(() => {
group.abort.signal.removeEventListener("abort", stop);
resolve();
}, pollMs);
group.abort.signal.addEventListener("abort", stop, { once: true });
});
}
} finally {
clearInterval(heartbeat);
group.abort.abort();
}
}
}
+60 -14
View File
@@ -8,7 +8,9 @@ import {
sortResources,
} from "../lib/apply";
import {
DATABASE_CLUSTER,
DATABASE_NAMESPACE,
DatabaseReconciliationError,
getComposePostgresClaims,
getRoleCredentials,
listManagedDatabaseResources,
@@ -494,8 +496,15 @@ export function createManagementService(dependencies: ManagementDependencies) {
);
if (full) {
// Database CRs may have consumers in other projects. Labels are not a
// reference count, including when this workspace created the CR.
const databases = await dependencies.listDatabaseResources(workspace.project);
retainedResources.push(...databases.filter(
(resource) => resource.kind === "Database" &&
resource.metadata?.labels?.[WORKSPACE_UID_LABEL] === workspace.uid,
));
deleteResources.push(
...(await dependencies.listDatabaseResources(workspace.project)),
...databases.filter((resource) => resource.kind !== "Database"),
...(await dependencies.listStorageResources(workspace.project)),
);
if (!safety.namespaceUid) {
@@ -640,19 +649,56 @@ export function createManagementService(dependencies: ManagementDependencies) {
compose: ComposeSpecification,
execution?: OperationExecution,
) {
throwIfExecutionAborted(execution);
await assertSafe(workspace, true);
const environment = await dependencies.reconcileDatabases(
workspace.project,
compose,
execution,
);
await ownExternalResources(
workspace,
DATABASE_NAMESPACE,
await dependencies.listDatabaseResources(workspace.project),
execution,
);
try {
throwIfExecutionAborted(execution);
await assertSafe(workspace, true);
} catch (error) {
throw new DatabaseReconciliationError("namespace precheck", error);
}
let environment: Record<string, Record<string, string>>;
try {
environment = await dependencies.reconcileDatabases(
workspace.project,
compose,
execution,
);
} catch (error) {
if (error instanceof DatabaseReconciliationError) throw error;
throw new DatabaseReconciliationError("database dependency", error);
}
let resources: KubernetesObject[];
try {
resources = await dependencies.listDatabaseResources(workspace.project);
} catch (error) {
throw new DatabaseReconciliationError("database resource listing", error);
}
try {
const claims = getComposePostgresClaims(compose);
const claimed = new Set(claims.map((claim) => claim.database));
for (const resource of resources) {
// Only a declared database may be shared. Never adopt its foreign
// workspace UID or apply its metadata through this field manager.
const existingOwner = resource.metadata?.labels?.[WORKSPACE_UID_LABEL];
if (resource.kind === "Database" && existingOwner &&
existingOwner !== workspace.uid &&
claimed.has(resource.metadata?.name ?? "")) {
if (resource.metadata?.namespace !== DATABASE_NAMESPACE) {
throw new Error("Database is outside expected namespace");
}
const claim = claims.find((item) => item.database === resource.metadata?.name);
const spec = (resource as KubernetesObject & {
spec?: { owner?: string; cluster?: { name?: string } };
}).spec;
if (spec?.owner !== claim?.username || spec?.cluster?.name !== DATABASE_CLUSTER) {
throw new Error("Database does not match the declared claim");
}
continue;
}
await ownExternalResources(workspace, DATABASE_NAMESPACE, [resource], execution);
}
} catch (error) {
throw new DatabaseReconciliationError("database resource ownership", error);
}
return environment;
},