Skip to main content
Network failures, rate limits, and transient server errors are facts of life in distributed systems. This guide covers patterns for handling them gracefully when integrating with the Signa API.

Transient vs. Permanent Failures

Before retrying, determine whether the failure is recoverable.
Status CodeTypeAction
429TransientRate limited. Retry after the Retry-After header value.
500TransientInternal server error. Retry with exponential backoff.
502TransientBad gateway. Retry with backoff (typically resolves within seconds).
503TransientService temporarily unavailable. Retry with longer backoff.
504TransientGateway timeout. Retry with backoff.
400PermanentValidation error. Fix the request payload before resending.
401PermanentInvalid or expired API key. Check your credentials.
403PermanentInsufficient scopes. Update your API key permissions.
404PermanentResource not found. Verify the ID is correct.
409PermanentConflict (e.g., duplicate create). Inspect the error body.
410PermanentEntity was merged. Follow the merged_into in the response.
422PermanentSemantic error. The request is well-formed but cannot be processed.
Only retry on transient failures (4xx rate limits and 5xx server errors). Retrying permanent failures wastes your rate limit budget and will never succeed.

Exponential Backoff with Jitter

The standard retry strategy for transient errors. Each retry waits longer than the previous one, with random jitter to avoid thundering-herd problems when many clients retry simultaneously. Algorithm:
delay = min(max_delay, base_delay * 2^attempt) + random(0, jitter)
interface RetryOptions {
  maxRetries?: number;
  baseDelayMs?: number;
  maxDelayMs?: number;
}

async function fetchWithRetry(
  url: string,
  options: RequestInit,
  { maxRetries = 3, baseDelayMs = 1000, maxDelayMs = 60_000 }: RetryOptions = {}
): Promise<Response> {
  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    const response = await fetch(url, options);

    // Do not retry permanent failures
    if (response.status >= 400 && response.status < 500 && response.status !== 429) {
      return response;
    }

    // Success -- return immediately
    if (response.ok) {
      return response;
    }

    // All retries exhausted
    if (attempt === maxRetries) {
      return response;
    }

    // Calculate delay
    let delayMs: number;

    if (response.status === 429) {
      // Prefer the server's Retry-After value
      const retryAfter = response.headers.get('Retry-After');
      delayMs = retryAfter ? parseInt(retryAfter, 10) * 1000 : baseDelayMs * 2 ** attempt;
    } else {
      delayMs = baseDelayMs * 2 ** attempt;
    }

    // Cap at max delay, add jitter
    delayMs = Math.min(delayMs, maxDelayMs);
    delayMs += Math.random() * baseDelayMs;

    await new Promise((resolve) => setTimeout(resolve, delayMs));
  }

  // Unreachable, but satisfies TypeScript
  throw new Error('Retry loop exited unexpectedly');
}

Retry Backoff Schedule

With the default settings (base_delay=1s, max_retries=3), the schedule looks like this:
AttemptBase DelayWith Jitter (approx.)Cumulative Wait
11 s1.0 — 2.0 s~1.5 s
22 s2.0 — 3.0 s~4 s
34 s4.0 — 5.0 s~8.5 s
For 429 responses, the Retry-After header overrides the calculated base delay. Always respect this value.
The Signa TypeScript SDK (@signa-so/sdk) has built-in retry logic with these defaults. If you are using the SDK, you get this behavior automatically.

Bulk Operation Retry

When using the batch endpoint, some items in a batch may succeed while others fail. Each item in the response has a status field (success or error) and an optional error object, so you can retry only the failed items:
import { Signa } from '@signa-so/sdk';

const signa = new Signa({ api_key: 'sig_YOUR_KEY' });

async function fetchBatchWithRetry(ids: string[], maxRetries = 3): Promise<any[]> {
  let pending = ids;
  const results: any[] = [];

  for (let attempt = 0; attempt <= maxRetries && pending.length > 0; attempt++) {
    const response = await signa.trademarks.batch({ ids: pending });

    const retryIds: string[] = [];

    for (const item of response.data) {
      if (item.status === 'success') {
        results.push(item.data);
      } else if (item.error?.type === 'rate_limited' || item.error?.type === 'internal_error') {
        retryIds.push(item.id);
      }
      // Permanent errors are skipped (not retried)
    }

    pending = retryIds;

    if (pending.length > 0 && attempt < maxRetries) {
      const delay = 1000 * 2 ** attempt + Math.random() * 1000;
      await new Promise((resolve) => setTimeout(resolve, delay));
    }
  }

  return results;
}

Circuit Breaker Pattern

For high-throughput integrations, wrap your API calls in a circuit breaker to stop sending requests when the API is consistently failing. This protects both your application and the API from cascading failures. The circuit has three states:
  • Closed (normal): Requests flow through. Failures are counted.
  • Open (tripped): All requests fail immediately without contacting the API.
  • Half-open (probing): A single test request is sent. If it succeeds, the circuit closes; if it fails, it re-opens.
class CircuitBreaker {
  private state: 'closed' | 'open' | 'half-open' = 'closed';
  private failureCount = 0;
  private lastFailureTime = 0;

  constructor(
    private readonly failureThreshold: number = 5,
    private readonly resetTimeoutMs: number = 30_000
  ) {}

  async execute<T>(fn: () => Promise<T>): Promise<T> {
    if (this.state === 'open') {
      // Check if enough time has passed to try again
      if (Date.now() - this.lastFailureTime >= this.resetTimeoutMs) {
        this.state = 'half-open';
      } else {
        throw new Error('Circuit breaker is open -- request blocked');
      }
    }

    try {
      const result = await fn();
      this.onSuccess();
      return result;
    } catch (error) {
      this.onFailure();
      throw error;
    }
  }

  private onSuccess(): void {
    this.failureCount = 0;
    this.state = 'closed';
  }

  private onFailure(): void {
    this.failureCount++;
    this.lastFailureTime = Date.now();

    if (this.failureCount >= this.failureThreshold) {
      this.state = 'open';
    }
  }
}

// Usage
const breaker = new CircuitBreaker(5, 30_000);

async function getTrademarkSafe(id: string) {
  return breaker.execute(() =>
    fetchWithRetry(`https://api.signa.so/v1/trademarks/${id}`, {
      headers: { Authorization: 'Bearer sig_YOUR_KEY' },
    })
  );
}
A circuit breaker should wrap your retry logic, not replace it. The retry function handles transient blips; the circuit breaker prevents sustained outages from overwhelming your application.

Request Timeouts

Always set explicit timeouts on API calls. A reasonable default for Signa endpoints:
Endpoint TypeRecommended Timeout
Single resource (GET /v1/trademarks/:id)10 s
List / search (GET or POST /v1/trademarks)15 s
Batch (POST /v1/trademarks/batch)30 s
Image similarity (POST /v1/trademarks/image-search)30 s
const controller = new AbortController();
const timeout = setTimeout(() => controller.abort(), 10_000);

try {
  const response = await fetch('https://api.signa.so/v1/trademarks/tm_abc123', {
    headers: { Authorization: 'Bearer sig_YOUR_KEY' },
    signal: controller.signal,
  });
} finally {
  clearTimeout(timeout);
}

Idempotent Requests

Mutation endpoints (PATCH, DELETE, and non-exempt POSTs — see Exempt Endpoints) require an Idempotency-Key header. This guarantees that if a request is retried — due to a network timeout, a dropped connection, or an ambiguous failure — the operation only executes once.
curl -X PATCH https://api.signa.so/v1/organization/api-keys/key_abc123 \
  -H "Authorization: Bearer $SIGNA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: rename-prod-key-2026-04-14" \
  -d '{"name": "Production key (renamed)"}'

How It Works

  1. On the first request with a given key, the API processes the request normally and caches the response.
  2. If the same key is sent again with the same request body, the API returns the cached response without re-executing the operation. Replays carry an Idempotent-Replayed: true response header and preserve the original request_id (in the body and on x-request-id), so you can tell a cached replay from a fresh execution.
  3. If the same key is sent with a different body, the API returns 409 conflict — each key is bound to a specific request body.
  4. If the same key is sent while the first request is still in flight, the API returns 409 idempotency_processing — wait for the original to complete, then retry with the same key to get its cached result.
Cached responses are stored for 24 hours, after which the key can be reused.
Only successful (2xx) responses are cached. Client errors (4xx) and server errors (5xx) are never cached — if a request fails, you can safely retry with the same idempotency key without being pinned to the error.

Key Format

  • 1-255 characters
  • Alphanumeric, dashes, and underscores only ([a-zA-Z0-9_-])
  • Must be unique per operation — use UUIDs, request identifiers, or a deterministic string derived from the operation (e.g., create-api-key-{name}-{timestamp})

Exempt Endpoints

Read-shaped POST endpoints are exempt by design from the Idempotency-Key requirement — they exist as POSTs only because their query bodies are too large or complex for a URL, and they create nothing. Sending a key is harmless (the value’s format is still validated — a malformed key returns 400 even here), but the middleware will not enforce or replay it.
EndpointWhy exempt
POST /v1/trademarksRead-only search, inherently safe to retry
POST /v1/trademarks/batchRead-only batch lookup, inherently safe to retry
POST /v1/suggest (and other suggest endpoints)Read-only, inherently safe to retry
POST /v1/watches/previewDry-run match count — creates nothing
POST /v1/alerts/lookupBulk read by IDs (polling pattern) — creates nothing
POST /v1/organization/api-keys and POST /v1/organization/api-keys/{id}/rotate are not exempt, even though they return one-time secrets. Idempotency replay is exactly what you want there: a retry with the same key returns the same secret instead of minting a second credential. The cached secret is only ever served to the caller that supplied the original Idempotency-Key.
Every other mutating endpoint (PATCH, DELETE, and any non-exempt POST) requires Idempotency-Key and will return 400 validation_error if it is missing.

Example: Safe Retry Pattern

The example below demonstrates a retry loop against PATCH /v1/organization/api-keys/{id} (renaming a key), which is enforced by the idempotency middleware.
import { randomUUID } from 'crypto';

async function renameApiKeySafe(keyId: string, name: string, maxRetries = 2): Promise<any> {
  const idempotencyKey = randomUUID();

  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    try {
      const response = await fetch(`https://api.signa.so/v1/organization/api-keys/${keyId}`, {
        method: 'PATCH',
        headers: {
          'Authorization': `Bearer ${process.env.SIGNA_API_KEY}`,
          'Content-Type': 'application/json',
          'Idempotency-Key': idempotencyKey,
        },
        body: JSON.stringify({ name }),
      });

      if (response.ok) return response.json();
      if (response.status < 500) return response.json(); // permanent error
    } catch {
      if (attempt === maxRetries) throw new Error('Request failed after retries');
    }

    await new Promise(r => setTimeout(r, 1000 * 2 ** attempt));
  }
}
The Signa TypeScript SDK sets the Idempotency-Key header automatically on all mutation requests. If you are using the SDK, you get idempotent retries without any extra code.

Decision Tree

Use this to determine the right strategy for any failure:
Request failed
  |
  |--> Status 400/401/403/404/410/422?
  |      --> Permanent failure. Do not retry. Log and handle.
  |
  |--> Status 409 (idempotency conflict)?
  |      --> Same key, different body. Use a new Idempotency-Key.
  |
  |--> Status 429?
  |      --> Read Retry-After header.
  |      --> Wait and retry with backoff.
  |
  |--> Status 500/502/503/504?
  |      --> Retry with exponential backoff + jitter.
  |      --> If 3+ consecutive 5xx: trip circuit breaker.
  |
  |--> Network error / timeout?
         --> Retry with backoff.
         --> If persistent: trip circuit breaker.