Filtering events

Everything here runs in the browser, before an event is queued or sent. Filtered events cost you no bandwidth and no storage.

The order things run

capture → ignoreErrors → allowHosts / denyHosts → sampleRate → redaction → beforeSend → queue

beforeSend runs last, so it sees the event exactly as it would be sent — already redacted, already classified.

Drop by message

CrxTrace.init({
  dsn: "...",
  ignoreErrors: [
    "ResizeObserver loop limit exceeded",
    /Extension context invalidated/,
  ],
});

Strings match as substrings; regular expressions are tested against the message. Use this for known-benign noise, not for hiding real bugs — a dropped event is gone, with no record that it happened.

Filter by host

For content scripts:

CrxTrace.init({
  dsn: "...",
  allowHosts: ["app.example.com", /\.mycompany\.com$/],
  denyHosts: [/bank/, "internal.corp"],
});

allowHosts — if non-empty, only these hosts are reported
denyHosts — always wins, even over allowHosts

Both accept strings (matched as substrings) and regular expressions. Neither affects service worker or extension page events, which have no host.

Sample high-volume events

CrxTrace.init({ dsn: "...", sampleRate: 0.25 });   // send 25%

The decision is made per event, client-side. Useful once an extension is large enough that a single bad release would otherwise flood your endpoint.

beforeSend — the escape hatch

Return the event to send it, a modified event to change it, or null to drop it:

CrxTrace.init({
  dsn: "...",
  beforeSend: (event) => {
    // Never report anything from this host
    if (event.host === "internal.example.com") return null;

    // Offline network errors aren't actionable
    if (event.mv3Category === "network" && !navigator.onLine) return null;

    // Strip something noisy you attached yourself
    delete event.contexts.extra?.rawResponse;

    // Add routing context
    event.tags.tier = isPaidUser ? "paid" : "free";

    return event;
  },
});

Keep it cheap and synchronous — it runs on every event, sometimes during a worker’s dying seconds. It cannot be async.

If beforeSend throws, the event is sent unmodified rather than lost.

Grouping control

Set fingerprint on a capture to force events together or apart:

CrxTrace.captureException(error, {
  fingerprint: ["sync-failure", provider],
});

Events sharing a fingerprint group into one issue regardless of stack. Useful when one logical failure surfaces through several code paths.

Turning off auto-instrumentation

Breadcrumbs from console.error/warn, fetch, and runtime.sendMessage are on by default:

CrxTrace.init({ dsn: "...", autoBreadcrumbs: false });

You can still call addBreadcrumb() manually. To trim rather than disable, cap how many are retained:

CrxTrace.init({ dsn: "...", maxBreadcrumbs: 10 });

What not to filter

Two things are handled for you and shouldn’t be re-implemented:

Host page errors are already excluded — only events with a frame in your bundle are reported. See Content scripts.
Routine worker recycling is already not reported as a fault. See Service worker deaths.

Configuration — every option
Privacy and redaction