Calls

The library has three ways to call a server method. They differ in waiting, return value, and behavior when the consumer ends.

invokeCopy Anchor Link

hubInvoke returns a function. The function waits for a connected hub, calls the method, and resolves with the typed result. It rejects with the original error. This sample loads the history of a room:

<script lang="ts">
  import { onMount } from "svelte";
  import { hubInvoke } from "./client";
  import type { ChatMessage } from "./contract";

  let { roomId }: { roomId: string } = $props();
  let history = $state<ChatMessage[]>([]);
  let error = $state<string>();
  const getHistory = hubInvoke("/hubs/rooms", "GetHistory");

  async function load() {
    try {
      history = await getHistory(roomId);
    } catch (e) {
      if (e instanceof DOMException && e.name === "AbortError") return;
      error = e instanceof Error ? e.message : String(e);
    }
  }

  onMount(load);
</script>

{#if error}
  <p role="alert">{error}</p>
{:else}
  <ul>
    {#each history as message}
      <li>{message.text}</li>
    {/each}
  </ul>
{/if}

sendCopy Anchor Link

hubSend returns a function that does not wait. It resolves true when it sends the call. It resolves false when the hub is not connected, or when send rejects. It never throws. This composer shows a notice when the hub drops the message:

<script lang="ts">
  import { hubSend } from "./client";

  let { roomId }: { roomId: string } = $props();
  let text = $state("");
  let dropped = $state(false);
  const send = hubSend("/hubs/rooms", "SendMessage");

  async function submit(event: SubmitEvent) {
    event.preventDefault();
    const sent = await send(roomId, text);
    dropped = !sent;
    if (sent) text = "";
  }
</script>

<form onsubmit={submit}>
  <input bind:value={text} />
  <button>Send</button>
  {#if dropped}
    <span role="alert">Not connected. The message was dropped.</span>
  {/if}
</form>

teardownCopy Anchor Link

hubTeardown returns a function that you can call in a cleanup. It does not depend on its consumer. It waits up to timeout (default 10 000 ms) for the hub, sends the call, and holds a lazy hub open until it is done. It resolves false when the hub does not connect in time, when the session is disabled, when the hub failed, or when send rejects. It never throws:

<script lang="ts">
  import { onDestroy } from "svelte";
  import { hubTeardown } from "./client";

  let { roomId }: { roomId: string } = $props();
  const leaveRoom = hubTeardown("/hubs/rooms", "LeaveRoom");

  onDestroy(() => {
    void leaveRoom(roomId);
  });
</script>

A teardown cannot land when the session stops at the same time. The call then resolves false.

InvokeOptionsCopy Anchor Link

An invoke takes these options. The configuration reference lists the types and defaults:

• timeout limits the wait for a connected hub, for each attempt. It does not limit a call that is already sent, because SignalR cannot cancel a sent call.
• retries sets the number of retries after a retriable failure.
• backoff sets the delays between attempts.
• isRetriable overrides the default rule below. An undefined answer keeps the default rule.
• keepAliveOnUnmount keeps a pending wait and the retry delays when the consumer ends.

A teardown takes only timeout.

Default retry ruleCopy Anchor Link

The library never retries an error named SignalRDisabledError or SignalRDisconnectedError, because waiting longer cannot help. For other errors, it follows the connection state:

• If the hub is not connected, the library retries.
• If the hub is connected, the library retries only a SignalR TimeoutError, an AbortError, or an HttpError with the status 0, 408, 429, or 5xx.
• A business error that the server hub throws is not retried.

A retry can run the call twice, for example when the connection drops after the server received the call. Use retries only for idempotent methods.

Abort when the consumer endsCopy Anchor Link

When a consumer ends, its calls abort. A pending wait and a retry delay reject with an AbortError and this message: The consumer unmounted, so the call was aborted. A call that is already sent finishes. Set keepAliveOnUnmount: true to prevent the abort.

A fire-and-forget call needs a .catch that ignores the abort. The samples in these docs use this helper, which logs every other error:

// An unmount aborts a waiting call with an AbortError. That is expected, so only other errors are logged.
export function ignoreAbort(error: unknown) {
  if (error instanceof DOMException && error.name === "AbortError") return;
  console.error(error);
}

Attach it to a call that you do not await, for example joinRoom(roomId).catch(ignoreAbort).

Disabled and failed hubsCopy Anchor Link

The errors in this section are Error objects with a name. Test the name, not a class:

if (error instanceof Error && error.name === "SignalRDisabledError") {
  // enabled is false, or baseUrl is empty
}

When enabled is false or baseUrl is empty, a call that waits for the connection rejects at once with an error named SignalRDisabledError. The library does not queue the call. This also applies while your app loads the token and keeps enabled at false. Make the call after enabled is true. An invoke that was already sent fails with the close error of SignalR.

When a hub is "disconnected" after a non-retriable error, a call rejects at once with an error named SignalRDisconnectedError.

When the hub does not connect within timeout, the call rejects with an error named SignalRTimeoutError. See Troubleshooting for the messages.