cairnq 0.1.0 → 0.3.0

package/src/store/base.ts

@@ -1,6 +1,68 @@
-import type { Task } from "../models.js";
+import { newId } from "../ids.js";
+import {
+  AlreadyExists,
+  errorEnvelope,
+  LostLease,
+  ProtocolVersionMismatch,
+  SerializationError,
+} from "../errors.js";
+import { rowToTask, STATUSES, type Task, type TaskStatus } from "../models.js";
 
-export type Conflict = "reuse" | "reject" | "replace";
+const rejectMangled = function (this: unknown, _key: string, v: unknown): unknown {
+  if (typeof v === "number" && !Number.isFinite(v)) {
+    throw new SerializationError(`non-finite number ${v} is not JSON-serializable`);
+  }
+  // In an array these become the literal `null` (in an object they are merely
+  // omitted, the JS idiom for "absent") — the twin SDK would read back a null
+  // the caller never wrote.
+  if (Array.isArray(this) && (v === undefined || typeof v === "function" || typeof v === "symbol")) {
+    throw new SerializationError(`${typeof v} inside an array is not JSON-serializable`);
+  }
+  return v;
+};
+
+/** Encode a value for a protocol JSON column, raising SerializationError on
+ * anything JSON cannot represent. Refuses what JSON.stringify would silently
+ * mangle into `null`: NaN/Infinity anywhere, undefined/function/symbol inside an
+ * array, and a top-level undefined that disappears entirely — either way the
+ * twin SDK reads back something other than what the caller meant (the Python
+ * SDK rejects the same values, via allow_nan=False). */
+export function dumpJson(value: unknown): string {
+  let text: string | undefined;
+  try {
+    text = JSON.stringify(value);
+  } catch (err) {
+    // BigInt or a circular structure.
+    throw new SerializationError(err instanceof Error ? err.message : String(err));
+  }
+  if (text === undefined) {
+    throw new SerializationError(`value of type ${typeof value} is not JSON-serializable`);
+  }
+  // Every mangled value reaches the output as the literal `null`, so a
+  // null-free result needs no strict pass — this keeps the replacer (which
+  // forfeits V8's native stringifier) off the hot path.
+  if (text.includes("null")) JSON.stringify(value, rejectMangled);
+  return text;
+}
+
+const SUPPORTED_PROTOCOL_MAJOR = 1;
+
+/** Refuse to run against a store whose protocol major this SDK does not speak.
+ * The supported major is a protocol fact, not a dialect one — every backend
+ * checks it here so the constant can't fork per store. */
+export function checkProtocolVersion(version: number): void {
+  if (version !== SUPPORTED_PROTOCOL_MAJOR) {
+    throw new ProtocolVersionMismatch(
+      `storage protocol_version=${version}, SDK supports ${SUPPORTED_PROTOCOL_MAJOR}`,
+    );
+  }
+}
+
+// CONFLICTS is the canonical declaration; the type derives from it so the
+// runtime guard in submit() and the type can't drift apart (same pattern as
+// STATUSES/TaskStatus in models.ts).
+const CONFLICTS = ["reuse", "reject", "replace"] as const;
+export type Conflict = (typeof CONFLICTS)[number];
 
 export interface SubmitInput {
   name: string;
@@ -18,7 +80,7 @@ export interface SubmitInput {
 }
 
 export interface ListInput {
-  status?: string | null;
+  status?: TaskStatus | null;
   queue?: string | null;
   name?: string | null;
   rootId?: string | null;
@@ -27,41 +89,395 @@ export interface ListInput {
   offset?: number;
 }
 
-/** The storage seam. SQLiteStore is the only MVP implementation. */
-export interface TaskStore {
-  connect(): Promise<void>;
-  close(): Promise<void>;
-  protocolVersion(): Promise<number>;
-
-  submit(input: SubmitInput): Promise<Task>;
-  get(taskId: string): Promise<Task | null>;
-  getByKey(key: string): Promise<Task | null>;
-  list(input?: ListInput): Promise<Task[]>;
-  cancel(taskId: string): Promise<Task | null>;
-  cancelByKey(key: string): Promise<Task | null>;
-  retry(taskId: string, opts?: { resetAttempt?: boolean }): Promise<Task | null>;
-  retryByKey(key: string, opts?: { resetAttempt?: boolean }): Promise<Task | null>;
-
-  claim(input: {
+export interface PurgeInput {
+  olderThanMs?: number;
+  limit?: number;
+}
+
+export type Params = Record<string, unknown>;
+/** Runs one named protocol statement and returns its rows. */
+export type Fetch = (name: string, params: Params) => Promise<any[]>;
+
+export const LEASE_EXPIRED_ERROR_JSON = dumpJson(
+  errorEnvelope({
+    type: "LeaseExpired",
+    code: "lease_expired",
+    message: "task lease expired and max attempts reached",
+    retryable: false,
+  }),
+);
+
+/** Strips SQL line comments, so a `:name` in a header comment isn't a parameter. */
+export const COMMENT = /--[^\n]*/g;
+/** A `:name` placeholder. The lookbehind spares Postgres `::type` casts. */
+export const NAMED = /(?<!:):(\w+)/g;
+
+// Statement text is loaded once at construction and never varies, so the parse is
+// memoized on it: every dialect's binding path runs on each query, and re-scanning
+// the SQL each time would put a regex sweep on the worker's poll loop.
+const paramCache = new Map<string, readonly string[]>();
+
+/**
+ * The parameter names a statement binds, in first-appearance order.
+ *
+ * Callers pass a superset of parameters and each dialect takes what its own SQL
+ * asks for — that is what lets one call site serve both dialects even though e.g.
+ * SQLite binds `:lease_until_ms` where Postgres binds `:lease_ms`. This is the one
+ * place that decides what counts as a parameter; both dialects' binding goes
+ * through it.
+ */
+export function statementParams(sql: string): readonly string[] {
+  let names = paramCache.get(sql);
+  if (!names) {
+    const seen = new Set<string>();
+    for (const m of sql.replace(COMMENT, "").matchAll(NAMED)) seen.add(m[1]);
+    names = [...seen];
+    paramCache.set(sql, names);
+  }
+  return names;
+}
+
+/**
+ * The storage seam.
+ *
+ * A backend supplies three things: how to run one protocol statement, how to run
+ * several inside a transaction, and how its dialect binds parameters. Everything
+ * above that — the submit conflict branches, the *_by_key lookups, the
+ * recover-then-claim sequence, the ownership-checked writes — lives here once,
+ * because those are protocol decisions rather than storage decisions. Keeping
+ * them in one place is what stops SQLite and Postgres from drifting apart in
+ * behavior; the shared SQL already stops them from drifting in wording.
+ */
+export abstract class TaskStore {
+  // ------------------------------------------------------------ dialect seam
+  abstract connect(): Promise<void>;
+  abstract close(): Promise<void>;
+  abstract protocolVersion(): Promise<number>;
+
+  /** Run one protocol statement outside a transaction, connecting if needed. */
+  protected abstract fetch(name: string, params: Params): Promise<any[]>;
+  /**
+   * Run several statements atomically; `fn` receives a Fetch bound to the txn.
+   *
+   * A backend may invoke `fn` more than once, retrying the transaction after a
+   * transient failure (SQLite does, on write-lock contention). So `fn` must be
+   * replayable: derive nothing inside it that the caller cannot derive twice —
+   * build ids and payloads before opening the transaction, not within it.
+   */
+  protected abstract tx<T>(fn: (fetch: Fetch) => Promise<T>): Promise<T>;
+
+  /**
+   * Whether it is worth opening the claim transaction at all. SQLite gates its
+   * single write lock behind a read-only probe; Postgres readers don't block
+   * writers, so it just says yes.
+   */
+  protected async hasClaimableWork(_params: Params): Promise<boolean> {
+    return true;
+  }
+
+  // ------------------------------------------------------------ wake channel
+  // Wake-or-timeout contract (PROTOCOL.md "Push wakeups"): resolve when the
+  // watched event may have happened, or after timeoutMs at the latest. The
+  // default is a plain sleep — polling IS the wake mechanism; a dialect with a
+  // push channel (PostgresStore, LISTEN/NOTIFY) resolves earlier.
+
+  /** Resolves when a task may have become claimable on one of `queues`. The
+   * timer is unref'd: the worker races this against its own stop-aware, ref'd
+   * sleep, so it must neither hold the process open nor need clearing. */
+  claimWake(_queues: string[], timeoutMs: number): Promise<void> {
+    return new Promise((resolve) => setTimeout(resolve, timeoutMs).unref?.());
+  }
+
+  /** Resolves when `taskId` may have gone terminal. Plain ref'd sleep —
+   * pollWait awaits it directly, so it is what keeps the process alive. */
+  taskDoneWake(_taskId: string, timeoutMs: number): Promise<void> {
+    return new Promise((resolve) => setTimeout(resolve, timeoutMs));
+  }
+
+  // --------------------------------------------------------------- internals
+  /**
+   * An ownership-checked worker write (heartbeat/progress/succeed/complete/fail).
+   * Each statement's WHERE pins worker_id + a live lease, so 0 rows back means
+   * the lease was lost — every such write reports it the same way.
+   */
+  private async ownedWrite(name: string, taskId: string, params: Params): Promise<Task> {
+    const rows = await this.fetch(name, params);
+    if (!rows.length) throw new LostLease(taskId);
+    return rowToTask(rows[0]);
+  }
+
+  private static one(rows: any[]): Task | null {
+    return rows.length ? rowToTask(rows[0]) : null;
+  }
+
+  // ------------------------------------------------------------- client side
+  async submit(input: SubmitInput): Promise<Task> {
+    const id = newId("task");
+    const ins: Params = {
+      id,
+      name: input.name,
+      queue: input.queue ?? "default",
+      payload: dumpJson(input.payload ?? {}),
+      metadata: dumpJson(input.metadata ?? {}),
+      max_attempts: input.maxAttempts ?? 3,
+      priority: input.priority ?? 0,
+      delay_ms: input.runAtDelayMs ?? 0,
+      parent_id: input.parentId ?? null,
+      root_id: input.rootId ?? id,
+      correlation_id: input.correlationId ?? null,
+    };
+    const key = input.key ?? null;
+    const conflict = input.conflict ?? "reuse";
+    // Validate up front: untyped callers otherwise only hit the strategy branch
+    // on the second submit of a key, deep inside the transaction.
+    if (!CONFLICTS.includes(conflict)) {
+      throw new Error(`unknown conflict strategy: ${conflict}`);
+    }
+    // maxAttempts < 1 would still run once (claim increments before the check),
+    // a silently different meaning than the number says; a negative delay is
+    // always a mistake. Both fail loudly instead. Only supplied values are
+    // checked — the defaults live in the params object alone.
+    if (input.maxAttempts != null && input.maxAttempts < 1) {
+      throw new Error(`maxAttempts must be >= 1, got ${input.maxAttempts}`);
+    }
+    if (input.runAtDelayMs != null && input.runAtDelayMs < 0) {
+      throw new Error(`runAtDelayMs must be >= 0, got ${input.runAtDelayMs}`);
+    }
+    if (key === null) return rowToTask((await this.fetch("insert_task", ins))[0]);
+
+    // A key makes submit a read-then-write, so it has to be one transaction —
+    // opened by taking the key's lock, because on Postgres the transaction alone
+    // is not enough: concurrent same-key submits must not both see "no existing
+    // task" (see lock_key.sql; on SQLite it is a no-op).
+    return this.tx(async (fetch) => {
+      await fetch("lock_key", { key });
+      const existing = (await fetch("get_key", { key })) as { task_id: string }[];
+      if (existing.length) {
+        // Read the task itself before branching: a concurrent purge (which
+        // takes no key lock) may have deleted it — cascading the key row away —
+        // between our statements' snapshots. A vanished task means the key is
+        // free after all, whatever the strategy.
+        const current = (await fetch("get", { id: existing[0].task_id }))[0];
+        if (current) {
+          if (conflict === "reuse") return rowToTask(current);
+          if (conflict === "reject") throw new AlreadyExists(key);
+          // "replace": cancel the recorded task, then repoint the key below.
+          await fetch("cancel", { id: existing[0].task_id });
+        }
+      }
+      const row = (await fetch("insert_task", ins))[0];
+      await fetch("upsert_key", { key, task_id: id });
+      return rowToTask(row);
+    });
+  }
+
+  async get(taskId: string): Promise<Task | null> {
+    return TaskStore.one(await this.fetch("get", { id: taskId }));
+  }
+
+  async getByKey(key: string): Promise<Task | null> {
+    return TaskStore.one(await this.fetch("get_by_key", { key }));
+  }
+
+  async list(input: ListInput = {}): Promise<Task[]> {
+    // Validate up front, like submit's conflict guard: a typo'd status otherwise
+    // matches nothing and returns [] indistinguishably from "no such tasks".
+    if (input.status != null && !STATUSES.includes(input.status)) {
+      throw new Error(`unknown status filter: ${input.status}`);
+    }
+    if ((input.limit != null && input.limit < 0) || (input.offset != null && input.offset < 0)) {
+      throw new Error(`limit/offset must be >= 0, got limit=${input.limit} offset=${input.offset}`);
+    }
+    const rows = await this.fetch("list", {
+      status: input.status ?? null,
+      queue: input.queue ?? null,
+      name: input.name ?? null,
+      root_id: input.rootId ?? null,
+      correlation_id: input.correlationId ?? null,
+      limit: input.limit ?? 100,
+      offset: input.offset ?? 0,
+    });
+    return rows.map(rowToTask);
+  }
+
+  async cancel(taskId: string): Promise<Task | null> {
+    return TaskStore.one(await this.fetch("cancel", { id: taskId }));
+  }
+
+  async retry(taskId: string, opts: { resetAttempt?: boolean } = {}): Promise<Task | null> {
+    return TaskStore.one(
+      await this.fetch("retry", { id: taskId, reset_attempt: opts.resetAttempt ?? false }),
+    );
+  }
+
+  async cancelByKey(key: string): Promise<Task | null> {
+    return this.byKey("cancel", key, {});
+  }
+
+  async retryByKey(key: string, opts: { resetAttempt?: boolean } = {}): Promise<Task | null> {
+    return this.byKey("retry", key, { reset_attempt: opts.resetAttempt ?? false });
+  }
+
+  /**
+   * Resolve a key to the task it currently points at, then act on that task —
+   * under the key's lock, so a concurrent `replace` can't repoint the key
+   * between the lookup and the write (the transaction alone is not enough on
+   * Postgres; see lock_key.sql).
+   */
+  private async byKey(name: string, key: string, params: Params): Promise<Task | null> {
+    return this.tx(async (fetch) => {
+      await fetch("lock_key", { key });
+      const existing = (await fetch("get_key", { key })) as { task_id: string }[];
+      if (!existing.length) return null;
+      return TaskStore.one(await fetch(name, { id: existing[0].task_id, ...params }));
+    });
+  }
+
+  /**
+   * Delete terminal tasks that completed more than `olderThanMs` ago and return
+   * their ids. Nothing else removes rows, so a long-lived database needs this
+   * called periodically. Bounded by `limit` to keep each sweep a short write;
+   * call it in a loop until it returns fewer than `limit`.
+   */
+  async purge(input: PurgeInput = {}): Promise<string[]> {
+    if (input.olderThanMs != null && input.olderThanMs < 0) {
+      throw new Error(`olderThanMs must be >= 0, got ${input.olderThanMs}`);
+    }
+    if (input.limit != null && input.limit < 1) {
+      throw new Error(`limit must be >= 1, got ${input.limit}`);
+    }
+    const rows = await this.fetch("purge", {
+      older_than_ms: input.olderThanMs ?? 0,
+      limit: input.limit ?? 1_000,
+    });
+    return rows.map((r) => r.id as string);
+  }
+
+  /**
+   * Task counts per queue, keyed by status and zero-filled across all statuses —
+   * `(await stats()).default.queued` is the backlog of a queue. A queue appears
+   * only while it has rows; terminal tasks keep counting until `purge` removes
+   * them.
+   */
+  async stats(): Promise<Record<string, Record<TaskStatus, number>>> {
+    const out: Record<string, Record<TaskStatus, number>> = {};
+    for (const row of await this.fetch("stats", {})) {
+      const per = (out[row.queue] ??= Object.fromEntries(
+        STATUSES.map((s) => [s, 0]),
+      ) as Record<TaskStatus, number>);
+      per[row.status as TaskStatus] = Number(row.count);
+    }
+    return out;
+  }
+
+  // ------------------------------------------------------------- worker side
+  /**
+   * Take up to `limit` claimable tasks. `names` restricts the claim to task names
+   * this caller can actually run — a worker passes its registered handlers.
+   * Queues alone do not partition work, so without it a worker claims a task it
+   * cannot run and fails it permanently. Undefined means no filter; an empty
+   * array claims nothing.
+   */
+  async claim(input: {
     queues: string[];
     workerId: string;
     leaseMs?: number;
     limit?: number;
-  }): Promise<Task[]>;
-  heartbeat(input: { taskId: string; workerId: string; leaseMs?: number }): Promise<Task>;
-  progress(input: {
+    names?: string[];
+  }): Promise<Task[]> {
+    // One queue is the common case and gets its own statement: a list-valued queue
+    // filter cannot be read in claim order, so the planner sorts every claimable
+    // row to take LIMIT of them, and claim's cost grows with the queued backlog
+    // while it holds the claim transaction. See claim_one_queue.sql.
+    const oneQueue = input.queues.length === 1;
+    const params: Params = {
+      queues: input.queues,
+      queue: oneQueue ? input.queues[0] : null,
+      names: input.names ?? null,
+      worker_id: input.workerId,
+      lease_ms: input.leaseMs ?? 30_000,
+      limit: input.limit ?? 1,
+      lease_expired_error: LEASE_EXPIRED_ERROR_JSON,
+    };
+    if (!(await this.hasClaimableWork(params))) return [];
+    // Recovery must share the claim's transaction: a lease reclaimed here has to
+    // be visible to the claim that follows, and to nobody in between.
+    return this.tx(async (fetch) => {
+      await fetch("recover_leases", params);
+      return (await fetch(oneQueue ? "claim_one_queue" : "claim", params)).map(rowToTask);
+    });
+  }
+
+  async heartbeat(input: { taskId: string; workerId: string; leaseMs?: number }): Promise<Task> {
+    return this.ownedWrite("heartbeat", input.taskId, {
+      id: input.taskId,
+      worker_id: input.workerId,
+      lease_ms: input.leaseMs ?? 30_000,
+    });
+  }
+
+  async progress(input: {
     taskId: string;
     workerId: string;
     progress: number | null;
     message: string | null;
-  }): Promise<Task>;
-  succeed(input: { taskId: string; workerId: string; result: unknown }): Promise<Task>;
-  complete(input: { taskId: string; workerId: string; result: unknown }): Promise<Task>;
-  fail(input: {
+  }): Promise<Task> {
+    return this.ownedWrite("progress", input.taskId, {
+      id: input.taskId,
+      worker_id: input.workerId,
+      progress: input.progress,
+      message: input.message,
+    });
+  }
+
+  async succeed(input: { taskId: string; workerId: string; result: unknown }): Promise<Task> {
+    return this.ownedWrite("succeed", input.taskId, {
+      id: input.taskId,
+      worker_id: input.workerId,
+      result: input.result == null ? null : dumpJson(input.result),
+      message: null,
+    });
+  }
+
+  async complete(input: { taskId: string; workerId: string; result: unknown }): Promise<Task> {
+    return this.ownedWrite("complete", input.taskId, {
+      id: input.taskId,
+      worker_id: input.workerId,
+      result: input.result == null ? null : dumpJson(input.result),
+    });
+  }
+
+  async fail(input: {
     taskId: string;
     workerId: string;
     error: unknown;
     retryable?: boolean;
     delayMs?: number;
-  }): Promise<Task>;
+  }): Promise<Task> {
+    let error: string;
+    try {
+      error = dumpJson(input.error ?? {});
+    } catch (err) {
+      if (!(err instanceof SerializationError)) throw err;
+      // A failure record must never itself fail to serialize (a TaskError
+      // carrying exotic details would otherwise strand the task until lease
+      // expiry). Strip the envelope to its string fields and record that.
+      const e = (input.error ?? {}) as { type?: unknown; code?: unknown; message?: unknown };
+      error = dumpJson(
+        errorEnvelope({
+          type: String(e.type ?? "TaskError"),
+          code: String(e.code ?? "task_error"),
+          message: String(e.message ?? ""),
+          retryable: input.retryable !== false,
+        }),
+      );
+    }
+    return this.ownedWrite("fail", input.taskId, {
+      id: input.taskId,
+      worker_id: input.workerId,
+      error,
+      retryable: input.retryable !== false,
+      delay_ms: input.delayMs ?? 0,
+    });
+  }
 }