Cloudflare AWS ALB SSL networking

Cloudflare Error 521 (Web server is down) on an AWS origin: ALB listeners, SSL mode and blocked ranges

Fixing Cloudflare error 521 with an AWS origin: a missing 443 listener on the ALB, an SSL/TLS mode that does not match the origin, a process not listening on the right interface, and blocks on Cloudflare IP ranges.

Jerzy Kopaczewski ·
Cloudflare returns error 521, which means your origin actively refused the connection. This is not silence (that is a 522); something on the other end said no. On AWS it nearly always comes down to a mismatch between Cloudflare's SSL/TLS mode and what is actually listening: an ALB with no listener on port 443 while the zone is set to Full, an application that has crashed or listens only on 127.0.0.1, or a firewall rejecting some of Cloudflare's addresses.

If the origin never answers and nothing refuses the connection, you are looking at error 522. If the connection works but the reply is empty or malformed, go to error 520.

Symptoms

Visitors see the Cloudflare page Error 521: Web server is down, and they see it straight away rather than after a long wait. Nothing from Cloudflare shows up in the application logs:

curl -sI https://app.example.com/ | head -4
# HTTP/2 521
# server: cloudflare
# cf-ray: 8a1b2c3d4e5f6a7b-LHR

# Straight to the origin (ALB or EC2), bypassing Cloudflare
curl -svk --connect-timeout 10 \
  --resolve app.example.com:443:$(dig +short my-alb-123456.eu-west-1.elb.amazonaws.com | head -1) \
  https://app.example.com/ 2>&1 | head -15
# "Connection refused" on port 443 confirms a 521

One quick tell: an ALB with no healthy targets returns a 503 of its own. If the response carries server: awselb/2.0, this is not a 521, and the ALB health check runbook is the one you need.

Cause

Cloudflare names two common causes: an origin application that is offline, and Cloudflare requests being blocked. On AWS they take these shapes:

  • No 443 listener on the ALB while the zone is on Full or Full (strict). In those modes Cloudflare connects to the origin over HTTPS on port 443. An ALB with only an HTTP listener on port 80 rejects it.
  • Flexible mode with an origin that listens only on 443 (or the reverse). Flexible connects over HTTP on port 80, so a closed port 80 produces a 521.
  • The application is down, or bound to the wrong interface. The process crashed, a restart failed, or the server listens only on 127.0.0.1 instead of 0.0.0.0. This applies to EC2 instances serving traffic directly, with no ALB in front.
  • A certificate that does not fit the SSL mode. Full modes need an origin that speaks HTTPS, and Full (strict) needs a valid certificate (ACM or a Cloudflare Origin CA certificate).
  • A block on Cloudflare’s IP ranges by fail2ban, ModSecurity or a host WAF. It is often partial, which is why the error looks intermittent.

Fix

Step 1: Find out which SSL/TLS mode the zone uses

The mode decides which port and protocol Cloudflare uses to reach the origin. Check it in the dashboard (SSL/TLS, Overview) or through the API:

curl -s -H "Authorization: Bearer $CF_API_TOKEN" \
  "https://api.cloudflare.com/client/v4/zones/$ZONE_ID/settings/ssl" \
  | python3 -c "import json,sys; print(json.load(sys.stdin)['result']['value'])"
# off | flexible | full | strict
Mode Cloudflare to origin Port the origin must serve
Flexible HTTP 80
Full HTTPS (any certificate) 443
Full (strict) HTTPS (valid or Origin CA certificate) 443

Step 2: Check the ALB has a listener on the right port

ALB_ARN=$(aws elbv2 describe-load-balancers --names my-alb \
  --query 'LoadBalancers[0].LoadBalancerArn' --output text)

aws elbv2 describe-listeners --load-balancer-arn "$ALB_ARN" \
  --query 'Listeners[].[Port,Protocol,DefaultActions[0].Type]' --output table
# With Full or Full (strict) there must be a row: 443 | HTTPS

If the 443 listener is missing, add it with a certificate. An ACM certificate works with Full (strict) as long as it covers the hostname Cloudflare uses. You can also import a Cloudflare Origin CA certificate into ACM, but remember that only Cloudflare trusts it, so the origin must be reachable through the proxy only.

resource "aws_lb_listener" "https" {
  load_balancer_arn = aws_lb.app.arn
  port              = 443
  protocol          = "HTTPS"
  ssl_policy        = "ELBSecurityPolicy-TLS13-1-2-2021-06"
  certificate_arn   = aws_acm_certificate.app.arn

  default_action {
    type             = "forward"
    target_group_arn = aws_lb_target_group.app.arn
  }
}

Step 3: Confirm the application is running and listening on the right interface

For EC2 instances that take traffic directly:

sudo systemctl status nginx        # or apache2, php-fpm, your app
sudo ss -tlnp | grep -E ':(80|443)\b'
# Look for 0.0.0.0:443 or [::]:443. A 127.0.0.1:443 entry means local-only.
sudo tail -n 50 /var/log/nginx/error.log

If the process dies on restarts or deploys, that is the real cause. Run it under a systemd unit with Restart=on-failure and alarm on instance status.

Step 4: Rule out a block on Cloudflare’s ranges

Check the host firewall, ModSecurity and fail2ban the same way as in the error 522 runbook (Steps 3 and 5). The difference here is that a 521 block is active: the server answers with a refusal rather than silence. An error that hits only some visitors, or only now and then, very often means just part of Cloudflare’s address space is blocked.

# Banned addresses (fail2ban 0.11+). Compare them with api.cloudflare.com/client/v4/ips
sudo fail2ban-client banned

Step 5: Watch for a redirect loop after changing the mode

There is one more trap tied to the SSL mode. If the ALB redirects all HTTP to HTTPS while Cloudflare runs in Flexible mode, Cloudflare connects over HTTP and gets a redirect every time, which the browser reports as ERR_TOO_MANY_REDIRECTS. The fix is Full (strict) with a 443 listener on the ALB. The 80-to-443 redirect can stay for clients that reach the ALB without going through Cloudflare.

Validation

# 1. The origin answers on the right port and protocol
curl -sk -o /dev/null -w "%{http_code}\n" \
  --resolve app.example.com:443:$(dig +short my-alb-123456.eu-west-1.elb.amazonaws.com | head -1) \
  https://app.example.com/
# Expected: 200 (or a 301/302 from the app), not a connection error

# 2. The site works through Cloudflare
curl -sI https://app.example.com/ | head -3
# Expected: HTTP/2 200, server: cloudflare

# 3. Strict mode passes certificate validation
curl -sv --resolve app.example.com:443:ORIGIN_IP https://app.example.com/ 2>&1 | grep -E "SSL certificate verify|subject:"

Once it is fixed, set up Cloudflare’s origin error notifications or a CloudWatch alarm on HTTPCode_ELB_5XX_Count, so the next outage is not first reported by your users.

Frequently asked questions about Cloudflare error 521

What does Cloudflare error 521 mean?

Error 521 (Web server is down) means your origin server refused the connection from Cloudflare. The usual causes are an application that is not running or not listening on the port the chosen SSL/TLS mode expects, or a firewall at the origin blocking Cloudflare's addresses.

Why did I get a 521 right after switching the SSL mode to Full?

In Full and Full (strict) modes Cloudflare connects to the origin over HTTPS on port 443. If your ALB has only a port 80 listener, or the server does not speak HTTPS, the connection is refused. Add an HTTPS listener with an ACM or Cloudflare Origin CA certificate before you change the mode.

Is a 521 the same as a 503 from my ALB?

No. A 521 is generated by Cloudflare and means the connection to the origin was not accepted at all. A 503 with the header server: awselb/2.0 comes from the ALB itself, which did accept the connection but has no healthy targets. The first is a network and listener problem, the second a health check and application problem.

 

Jerzy Kopaczewski

Cloudflare in front of AWS and the 52x errors will not go away?

Book a free 30-minute call. We set up SSL modes, ALB listeners, certificates and traffic filtering so that Cloudflare and AWS work together without surprises.

Book a call