The tricky thing about third-party API failures is that they look like your bug. A customer hits your checkout, gets an error, and blames your site. Your logs show a 500 from Stripe's end, but by the time you dig through the stack trace and realize it's not your code, fifteen minutes have passed. Stripe's own status page still says "All Systems Operational." This happens more often than the uptime numbers suggest.
You can't control Stripe's uptime, or Twilio's, or SendGrid's. But you can control how fast you find out. Hit their health endpoints on a schedule, check the response, alert the moment something breaks. Cloud Cron handles the scheduling and the alerting. This guide covers the setup: which endpoints to check, how often, what to look for in the response, and what to do when a dependency goes dark.
Why third-party API monitoring matters more than you think#
Your application is only as reliable as its weakest dependency. A SaaS product with 99.99% uptime relies on five external APIs. If each of those runs at 99.9%, combined availability drops to around 99.5%. That's over 43 hours of potential downtime per year, even if your own code never breaks.
The real problem isn't downtime itself. It's detection time. AWS has had multi-hour S3 outages that cascaded through thousands of apps. Twilio incidents have taken 30-40 minutes to appear on their status page. OpenAI's public status history shows roughly one incident every 2.5 days. By the time the provider acknowledges the issue, your users have already hit errors.
Most teams monitor their own endpoints with uptime checks but treat external APIs as somebody else's problem. They are somebody else's problem to fix. They're your problem to detect. Industry estimates put the annual cost of downtime in the hundreds of billions, and a significant chunk comes from cascading failures in dependencies nobody was watching.
Health check endpoints worth monitoring#
Most major APIs expose a dedicated health or status endpoint. Lightweight, unauthenticated, designed to return quickly.
Payment processors
Stripe doesn't publish an official health endpoint, but https://api.stripe.com/v1/charges with an invalid key returns a 401 with a structured JSON error in under 200ms when healthy. If it returns a 500 or times out, something is genuinely wrong. For authenticated checks, GET /v1/balance with your API key returns your account balance and confirms the API is processing requests end-to-end.
PayPal exposes https://api-m.paypal.com/v1/oauth2/token. A 401 with valid JSON on bad credentials means the API is up. Response times above 2 seconds on this endpoint historically precede broader issues.
Communication APIs
Twilio: https://api.twilio.com/2010-04-01 returns an XML response with available API versions. No auth needed. When this endpoint degrades, SMS delivery follows within minutes.
SendGrid: https://api.sendgrid.com/v3/scopes with your API key confirms both connectivity and auth. Their v3 API returns clean JSON errors on failure. We noticed during a past SendGrid incident that response times climbed for about 8 minutes before the first delivery actually failed.
AI and ML services
OpenAI: https://api.openai.com/v1/models with your API key lists available models. Response time on this endpoint correlates closely with inference latency. If the models list takes 3 seconds, your completions will be slow too.
Cloud storage
AWS S3: A HEAD request to your bucket (https://your-bucket.s3.amazonaws.com) returns a 200 or 403 depending on permissions, both confirming S3 is reachable. A timeout or 503 means trouble. For region-specific checks, hit the regional endpoint directly.
Setting up Cloud Cron for API health checks#
Each external API you depend on gets its own Cloud Cron task. One task per endpoint, not one task that checks everything. You want independent failure detection so a Stripe outage doesn't mask a simultaneous SendGrid problem.
Step 1: Create the Cloud Cron task
In your WatchCron dashboard, create a new Cloud Cron task with these settings:
- URL: The health endpoint, e.g.
https://api.twilio.com/2010-04-01 - Method: GET (or HEAD for S3-style checks)
- Schedule:
*/1 * * * *for critical payment APIs,*/5 * * * *for less critical services - Headers: Add
Authorization: Bearer sk-...if the endpoint requires authentication - Expected status: Set the success codes, 200 for most or 401 if you're intentionally hitting without auth
[Screenshot: Cloud Cron task creation form with Twilio health endpoint, GET method, */1 * * * * schedule, and 200 expected status]
For endpoints where you need to verify the response body beyond just a status code, enable keyword validation. Point it at a string you know appears in healthy responses, like "object": "list" for OpenAI's models endpoint or "Versions" for Twilio's root. The Cloud Cron feature page covers the full configuration options.
Step 2: Set response time thresholds
API degradation usually shows up as slow responses before it turns into errors. A Stripe health check that normally returns in 180ms jumping to 1,200ms is a warning sign even if the status code is still 200.
Configure response time alerts on each task. Set the threshold at roughly 3x the endpoint's normal response time. Our Twilio check normally returns in about 90ms, so we set the threshold at 300ms. This catches degradation during the window between "working fine" and "completely down," the window where you can still reroute traffic or warn users.
Step 3: Configure alerts that actually wake people up
A Slack notification about a Stripe outage at 2 AM helps nobody if nobody checks Slack at 2 AM. For payment and communication APIs, set up escalation through multiple alert channels:
- First alert: Slack or Telegram for the on-call channel
- If unacknowledged after 5 minutes: SMS on Pro plans
- If still unacknowledged: PagerDuty or OpsGenie for full incident escalation
Match alert urgency to business impact. A slow OpenAI response? Slack message. A down Stripe API at peak hours needs a phone call. Nobody has ever complained about being woken up too early for a payment outage. They complain about finding out too late.
Check intervals: how often is often enough?#
Does every API need the same check frequency? No. Over-monitoring wastes your Cloud Cron task slots. Under-monitoring defeats the purpose.
| Interval | Use case | Examples |
|---|---|---|
| Every 1 minute | Payment and auth APIs where downtime means lost revenue | Stripe, PayPal, Auth0, Okta |
| Every 5 minutes | Communication and storage APIs that queue messages | Twilio, SendGrid, S3, GCS |
| Every 15 minutes | Analytics, non-critical integrations called a few times daily | Maps, weather, reporting |
At the every-minute tier, you detect issues within 60 seconds, faster than most providers update their own status pages. Five-minute detection still beats the typical 15-30 minute provider acknowledgment. We run this exact setup for our own product: 3 payment checks every minute, 4 communication checks every five minutes, 3 non-critical checks at fifteen-minute intervals. Ten Cloud Cron tasks total, fitting comfortably in the Pro plan's allowance with room for actual scheduled jobs.
What do healthy and unhealthy results look like?#
After a few hours of checks, your WatchCron dashboard shows response time charts for each API endpoint.
Baseline shift: Stripe's health check averaged 160ms last week, now it averages 280ms. Something changed on their end. It might not be an outage yet. Customers feel it before dashboards show it.
Intermittent failures: One failed check out of 60 in an hour is probably a network hiccup. Three failures in a row is a pattern. Cloud Cron retries failed requests with increasing delays before marking a check as failed, which filters out most transient noise.
Timeout without error: The endpoint doesn't return an error. It just doesn't return at all within 60 seconds. This is the most dangerous failure mode because it ties up your application's HTTP connections. We saw this with a mapping API dependency last year: the endpoint hung for 45 seconds per request, and our queue workers backed up within minutes. If your Cloud Cron task consistently times out for a specific API, your app is probably experiencing the same thing. Our timeout guide covers handling strategies.
When an API goes down: the response playbook#
Detection is half the battle. The other half? What happens in the five minutes after your phone buzzes at 3 AM.
- Confirm it's them, not you. Check the provider's status page. Check DownDetector for community reports. Run a quick curl from a different network. If your Cloud Cron detected it and it's confirmed external, you know it's real.
- Activate your fallback. If you have one. Dual payment processors, like Stripe primary and PayPal fallback, or dual email providers let you reroute in minutes. Without a fallback, move to step 3.
- Communicate. Update your status page with a note like "Payment processing delayed due to third-party provider issue." Your users would rather see honesty than silence. WatchCron's incident management ties the alert directly to a status page update, so the workflow is: alert fires, create incident, status page reflects it automatically.
- Queue, don't drop. If the API processes transactions like payments, messages, or webhooks, queue the requests locally and replay them when the API recovers. Losing a charge is worse than delaying it by 30 minutes.
Building a complete dependency monitoring setup#
API health checks tell you when a provider is down. But what about the ripple effects? If Stripe goes down, your /checkout page probably returns errors too. If SendGrid is unreachable, your transactional emails stop. A complete setup watches the dependencies and your own surfaces that touch them.
- Cloud Cron tasks hit health endpoints on schedule and check status codes and response times
- Uptime monitors watch your own endpoints that depend on these APIs
- SSL monitors track certificate expiration on API domains you call. An expired cert on your payment provider's endpoint breaks your integration just as surely as a server crash
- Port monitors verify self-hosted dependencies like internal APIs and databases are accepting connections
- Blocklist monitors check your sending domain reputation, which matters if you rely on SendGrid or Twilio for email and SMS
When the Stripe health check fails at the same time your checkout uptime drops, having both in one dashboard tells you the cause without switching between three different tools.
Cloud Cron API monitoring: limitations to know#
Cloud Cron's 60-second timeout means you're testing reachability and basic response, not running end-to-end transaction tests. If you need to verify that Stripe actually processes a $0.50 test charge, you'll need a server-side script with a longer timeout and a heartbeat ping confirming completion.
The minimum check interval is one minute. If you need sub-minute API monitoring, every 15 or 30 seconds, dedicated API monitoring tools like Checkly at $24 a month or synthetic monitoring platforms handle that. For most production use cases, 60-second detection is fast enough. The bottleneck is usually your team's response time, not detection time.
Health endpoints aren't perfect proxies for full API functionality. Stripe's balance endpoint might return 200 while their webhooks are delayed. We learned this firsthand during a late-2025 incident when Stripe's charge endpoint worked fine but webhook delivery lagged by 12 minutes. You're monitoring a signal, not the whole system. Combine it with heartbeat monitoring on your actual business-critical workflows for defense in depth.
Frequently asked#
Most major APIs expose a dedicated health or status endpoint. Stripe responds to GET /v1/balance with your API key. Twilio returns API versions at https://api.twilio.com/2010-04-01 without authentication. OpenAI lists available models at GET /v1/models. Even hitting these endpoints without valid credentials returns a structured 401 error, which still confirms the API is responding. Create a Cloud Cron task for each endpoint you depend on.
Every minute for payment and authentication APIs where downtime means lost revenue. Every 5 minutes for communication and storage APIs that queue messages and tolerate brief blips. Every 15 minutes for analytics and non-critical integrations. A typical SaaS setup uses 10 Cloud Cron tasks across these three tiers.
Yes. Configure response time alerts at roughly 3x the endpoint's normal response time. If Stripe normally responds in 180ms and suddenly takes 1,200ms, Cloud Cron flags it even though the status code is still 200. Degradation typically shows up as slow responses 5-15 minutes before the API starts returning errors, giving you time to activate fallbacks or warn users.
Cloud Cron checks reachability and basic response health at 1-minute intervals within a 60-second timeout. Dedicated tools like Checkly at $24 a month offer sub-minute checks, multi-step API tests, and synthetic transaction monitoring. The trade-off: dedicated tools go deeper on API testing, while Cloud Cron integrates health checks into the same dashboard as your uptime, SSL, and cron job monitoring. For most teams, the combined view is more practical than a separate tool.
Four steps. First, confirm it's the provider and not your network by checking their status page and DownDetector. Second, activate your fallback if you have dual providers. Third, update your status page with an honest note about the third-party issue. Fourth, queue requests locally instead of dropping them and replay when the API recovers. WatchCron's incident management ties the alert to your status page automatically.
Uptime monitoring watches your own endpoints. API health checks watch the external services your application depends on. Both can fail independently: your server might be healthy while Stripe is down, or Stripe might be fine while your server crashes. Running both gives you the full picture. Cloud Cron handles the external API checks, while WatchCron's uptime monitors cover your own surfaces.