Varnish Cache v4后端健康检查异常：CMS宕机仍返回正常状态

Hey there! Let's figure out why Varnish isn't recognizing your CMS backend as down when you shut it down manually—this is a common gotcha with Varnish's default health checking behavior.

1. The Core Issue: Missing Health Probe Configuration #

By default, Varnish 4 doesn't perform active health checks on backends unless you explicitly define a probe. Without a probe, std.health(req.backend_hint) will always return true, even if your CMS is completely offline. Varnish assumes the backend is healthy unless told otherwise via a probe or manual CLI intervention.

2. Fix: Add a Backend Health Probe #

You need to define a probe in your VCL to let Varnish actively check if your CMS is responding. Here's a sample configuration tailored to your use case:

backend cms {
    .host = "your-cms-hostname-or-ip";
    .port = "80"; # or 443 if using HTTPS (adjust accordingly)
    
    # Define health check probe
    .probe = {
        .url = "/"; # Use a lightweight endpoint like /health if your CMS has one
        .interval = 5s;  # Check every 5 seconds
        .timeout = 2s;   # Fail if no response within 2 seconds
        .window = 5;     # Evaluate the last 5 check attempts
        .threshold = 3;  # Mark backend as sick if 3+ checks fail
    }
}

• Use a dedicated health endpoint (like /health) if your CMS provides one—it's more reliable than the homepage, which might have heavy assets.
• Adjust interval, timeout, window, and threshold based on your tolerance for downtime detection speed.

3. Ensure Grace Mode is Properly Configured #

Even with a working probe, you need to make sure Varnish uses cached content during the grace period when the backend is down. Add these sections to your VCL:

Set Grace Period on Backend Responses #

sub vcl_backend_response {
    # Set how long Varnish can use stale cached content when backend is down
    set beresp.grace = 24h; # Adjust to your desired grace period
}

Use Grace Content When Backend is Unhealthy #

sub vcl_recv {
    # Tell Varnish to allow grace content if backend is sick
    if (!std.health(req.backend_hint)) {
        set req.grace = 24h; # Match the beresp.grace value
    }
}

sub vcl_hit {
    # Deliver stale content from grace period if backend is down
    if (!std.health(req.backend_hint)) {
        if (obj.ttl + obj.grace > 0s) {
            return (deliver);
        }
    }
}

4. Verify the Fix #

After updating your VCL and reloading Varnish (varnishreload), test it out:

1. Shut down your CMS manually.
2. Wait for the probe's detection window (e.g., 15 seconds if using the sample probe settings).
3. Check the backend health status via the Varnish CLI:

   varnishadm backend.list
   

   You should see your cms backend marked as sick instead of healthy (no probe).
4. Send a request to Varnish—you should get the cached content instead of a 503.

Quick Notes #

• If you're using HTTPS for your backend, make sure you're using backend_ssl (or appropriate SSL configuration for Varnish 4) and adjust the probe URL/port accordingly.
• Avoid setting the probe interval too low (e.g., <1s) as it can add unnecessary load to your CMS.

Give these steps a shot, and you should see Varnish correctly detect the down backend and serve cached grace content as expected!

内容的提问来源于stack exchange，提问作者David Beaudway