# My app isn't reachable > Fix "This app didn't respond", "Nothing is deployed here", wrong ports, custom domains, certificates and services that can't reach each other. For agents: [llms.txt](https://ployz.dev/llms.txt) indexes the Ployz docs as Markdown. A visit to your app is a short trip: the browser looks up your domain, reaches one of your servers over https, and that server hands the request to your app on its port. When a page won't load, one leg of the trip is broken. ```mermaid flowchart LR browser(["Browser"]) -- "DNS" --> server["One of your servers"] server -- "PORT" --> app["Your app"] ``` Start with your domain's status in the service's **Settings → Networking**: it names most of the problems below. Your app's own errors are in the service's **Logs**. To see which server answered and what it knows about your service, [trace a request](https://ployz.dev/docs/troubleshooting/trace). When Ployz can't hand a request to your app, your server answers with its own page. Its title is a heading below, and its code (like `502 · NO_RESPONSE`) is on the page and in the `ployz-error` response header. ## My service has no address Services you add in the dashboard don't get a public address until you ask for one. In **Settings → Networking**, click **Generate Domain**, leave **Port** blank to use `PORT`, then deploy. ![The Generate Service Domain dialog](https://raw.githubusercontent.com/getployz/ployz/main/docs/user/images/generate-domain-dialog.png) ## This app didn't respond You'll see `502 · NO_RESPONSE`, then `503 · NO_RESPONSE` for the next 30 seconds while your server stops sending requests to a replica that didn't answer. Your server got the request, but your app didn't answer. 1. **Listen on `0.0.0.0` and `PORT`.** An app listening on `localhost` or `127.0.0.1` turns away requests from outside its container: | App | Listen on all addresses and `PORT` | | --- | --- | | Node.js, Express | `app.listen(process.env.PORT, "0.0.0.0")` | | Next.js | `next start` already does. With `output: "standalone"`, add the variable `HOSTNAME=0.0.0.0`. | | Vite preview | `vite preview --host 0.0.0.0 --port $PORT` | | gunicorn | `gunicorn --bind 0.0.0.0:$PORT app:app` | | uvicorn | `uvicorn main:app --host 0.0.0.0 --port $PORT` | | Rails, Puma | `bundle exec puma -b tcp://0.0.0.0:$PORT` | | Laravel | `php artisan serve --host=0.0.0.0 --port=$PORT` | Change it in your code or in **Settings → Deploy → Start command**. If Vite preview says "Blocked request. This host is not allowed.", add your domain to `preview.allowedHosts` in `vite.config`. 2. **Check the port matches.** See the next section. 3. **Check your app is running** in its **Logs**. A replica that crashes or fails its health check gets no traffic; see [Troubleshooting deployments](https://ployz.dev/docs/troubleshooting/deployments). 4. **Check the server running it is up.** A replica on a server that's down or out of reach doesn't answer either. See [Troubleshooting servers](https://ployz.dev/docs/troubleshooting/servers). When a server running one of your replicas powers off, a new connection to it gives up after half a second and the request goes to another replica. A request that had already reached it, or that went out on a connection the web server keeps open to it, waits about 14 seconds. Then a `GET`, like a page load, goes to another replica, and anything else, like a form post, gets `504 · NO_RESPONSE`. Your app may have handled it before the connection broke, so check whether it went through, like whether the order exists, before you send it again. The same 14-second limit applies to visitors. When a visitor's connection stops acknowledging data for about 13 seconds, say a phone on a bad network in the middle of a download, the web server drops it, and the visitor's browser or app has to retry. On a server upgraded from an earlier release, the web server Ployz already ran there keeps the old behavior: a request that had reached the powered-off server can hang for about 15 minutes. That lasts until the web server is created again. To do that now, open the server's page, turn **Takes web traffic** off, then on again. While it's off, that server doesn't answer on ports 80 and 443, so do one server at a time. A replica gets 60 seconds to start its response. If it hasn't by then, a `GET` goes to another replica and anything else gets `504 · NO_RESPONSE`, and a replica that hangs like this stops getting requests for a while. Server-sent events and websockets are fine as long as they answer straight away. For work that takes longer, like a big export, start it in the background and have the page check back for the result. ## This app isn't running You'll see `503 · NOT_RUNNING`. The service has no healthy replica, so there's nothing to send the request to. The page reloads itself every 10 seconds. - **A deploy is in progress.** The page goes away once the first replica passes its health check. - **The service is stopped.** Run `ployz service start `, or deploy it. See [Restart, stop or open a shell](https://ployz.dev/docs/services/services#restart-stop-or-open-a-shell). - **Every replica fails its health check.** See [Troubleshooting deployments](https://ployz.dev/docs/troubleshooting/deployments). ## My app listens on a different port The domain's row says where traffic goes. **→ Uses PORT** means the `PORT` variable, which is `8080` unless you set it. Either: - add a `PORT` variable set to your app's port, then deploy, or - click the pencil on the domain and set **Target port** to the port your app listens on. The health check follows whichever you pick. ## Nothing is deployed here You'll see `404 · NOT_DEPLOYED`. The domain leads to one of your servers, but no service answers for it there. - **Check the exact name.** `www.example.com` and `example.com` are different domains. Add each one you use to the service. - **On `http://`?** Ployz redirects http to https only once the domain has its certificate. Until then, wait for it to be issued. Behind Cloudflare, see [below](#my-domain-is-behind-cloudflare). - **On `https://`?** The domain isn't live yet. If its row says **Live after your next deploy**, click **Deploy**. - **Removed the service?** While your servers still hold a valid certificate for its domain, `https://` shows this page too, just like `http://`. Nothing renews that certificate, so once it expires, browsers show a certificate error instead. ## My custom domain is waiting for DNS You'll see **Waiting for DNS**, **DNS doesn't point here yet** or **DNS points to another server**. Your servers can't get a certificate until your domain leads to them. 1. Click **Show DNS records** on the domain's row and create exactly those records at your DNS provider. 2. Delete any other A and AAAA records for the same name, like your old host's. A leftover AAAA record alone can block the certificate. 3. Give DNS time to spread. Your servers keep checking on their own. ## The browser can't make a secure connection Your servers don't have a certificate for the domain yet. The domain shows **Issuing certificate**, a DNS status, or **Certificate failed; retrying at …**. Fix any DNS status first (see above). Otherwise give it a little time: your servers keep trying on their own. ## My domain is behind Cloudflare - Set **SSL/TLS** to **Full (strict)**. **Flexible** reaches your servers over plain http, which redirects back to https, so the browser says "too many redirects". - If the domain shows **Your proxy redirects to HTTPS: exempt /.well-known/acme-challenge/* from it**, Cloudflare's redirect is blocking your certificate. Set the DNS record to **DNS only**, or turn off **Always Use HTTPS** and add a redirect rule that sends http to https for every path except `/.well-known/acme-challenge/*`. - Your app sees the visitor's address, not Cloudflare's, in `X-Forwarded-For` and `X-Real-IP`. Your servers take it from Cloudflare's `CF-Connecting-IP` header, and only from requests that come from [Cloudflare's addresses](https://www.cloudflare.com/ips/). They replace both headers before a request reaches your app, so a visitor can't fake their address in either. - `CF-Connecting-IP` reaches your app only on requests that really came through Cloudflare or your Tunnel. Your servers remove it from any other request. Read `X-Forwarded-For` if your app takes traffic both ways. If your app forwards a request's headers to another site behind Cloudflare, drop the `CF-*` headers first: Cloudflare refuses a request that already carries `CF-Connecting-IP` with `403 error code: 1000`. ## Port 80 is closed You'll see **Port 80 is closed on 203.0.113.10. Open port 80 in the Server's firewall (for example ufw allow 80/tcp)**. Your servers use port 80 to get certificates, so Ployz leaves that server out of your organization's domain until it's open. When your other servers still take the traffic, your domains stay live and only **Domain** in **Organization → General** shows **Port 80 closed on 203.0.113.10.** An offline server is left out without a warning here: it shows **Offline** on the Servers page instead. Open TCP port 80 and TCP and UDP port 443 to the internet on every server that takes web traffic, in your hosting provider's firewall (like a Hetzner Cloud Firewall or an AWS security group) as well as the server's own (with UFW, `ufw allow 80/tcp`). Then click **Check again** under **Domain** in **Organization → General**. ## Something else answers on port 80 You'll see **Something else answers on port 80 of 203.0.113.10**. The server answers on port 80, but not as Ployz: another web server holds the port, often Caddy or nginx in front of a [self-hosted Ployz Cloud](https://ployz.dev/docs/self-hosting#choose-where-it-runs). Until Ployz gets the port back, Ployz leaves that server out of your organization's domain. When your other servers still take the traffic, your domains stay live and only **Domain** in **Organization → General** shows **Port 80 taken on 203.0.113.10.** See what holds it: ```sh sudo ss -ltnp '( sport = :80 or sport = :443 )' ``` Stop that web server, or move it to another server. To keep Ployz Cloud on the same server, publish it through a Cloudflare Tunnel instead, or serve your apps through one with [`ployz ingress use tunnel`](https://ployz.dev/docs/services/cloudflare-tunnel). Once port 80 is free, the server starts its web server on its own within a minute. You don't need to remove and re-add the server. Then click **Check again** under **Domain** in **Organization → General**. ## My servers only have IPv6 addresses A server with no IPv4 route needs NAT64 and DNS64 to set Ployz up; see [IPv6-only servers](https://ployz.dev/docs/servers/add-a-server#ipv6-only-servers). Ployz Cloud checks port 80 from its own network, which may have no IPv6 route. When a server with only an IPv6 address doesn't answer that check, Ployz doesn't call its port closed. **Domain** in **Organization → General** shows **Ployz Cloud can't reach 2001:db8::1 over IPv6, so it didn't check port 80 there.** - When every server is IPv6-only, your organization's domain still points at them. - When another server answers the check, the domain points only at the servers that answered, and **Domain** says traffic skips the IPv6 ones. Ployz can't confirm such a server takes traffic, so check it yourself from a network with IPv6: `curl -I http://web.`. Open TCP port 80 and TCP and UDP port 443 for IPv6 in every firewall, as in [Port 80 is closed](#port-80-is-closed). To reach your apps from an IPv4-only network, give a server an IPv4 address too. ## My apps go through a Cloudflare Tunnel and don't answer Run `ployz ingress`. Every server that takes web traffic should list `tunnel`. One that lists **closed** has no web server running; see [Ingress isn't running](#ingress-isnt-running). One that still lists `public` never switched; run `ployz ingress use tunnel` again and read its row. When every server lists `tunnel` and the domain still doesn't answer, check the tunnel in Cloudflare under **Zero Trust → Networks → Tunnels**. It should show one connector per server, and a public hostname for the domain with service `http://localhost:8480`. A hostname with any other service gets no answer, since the connector reaches nothing else. A tunnel whose token you rotated needs the new token: `ployz ingress use tunnel --token-stdin`. A domain behind the tunnel never shows **Issuing certificate**; Cloudflare serves it. If the browser says "too many redirects", set **SSL/TLS** to **Full** or **Flexible** in Cloudflare. ## My custom domain isn't reaching my Cloudflare Tunnel You'll see **Not reaching your tunnel** when Ployz asked the domain for a page only your servers answer and something else answered, or DNS sends the name nowhere. You'll see **No answer over HTTPS** when nothing answered at all: the connection was refused, its certificate failed, or it timed out, as with a name still pointing at an old server. 1. Click **Show DNS records** and create that CNAME at your DNS provider. In Cloudflare, turn **Proxied** on. 2. Delete any other A, AAAA or CNAME record for the same name. 3. A name in your Cloudflare zone needs no record of its own, unless one exists: a record for the exact name overrides the zone's `*` CNAME. Replace it with the CNAME shown, or delete it and check that the `*` CNAME to the tunnel still exists under **DNS → Records**. 4. With the record right, check that the tunnel has a connector for each server, as [above](#my-apps-go-through-a-cloudflare-tunnel-and-dont-answer). Ployz asks the name again each time you open the page. **Check** asks now. A name two levels under your zone, like `a.shop.example.com`, shows **The free certificate covers one level under example.com**. Cloudflare's free certificate doesn't cover it, so browsers warn. Use a name one level down, like `shop-a.example.com`. ## Ingress isn't running You'll see **Ingress isn't running on web-1**, under **Domain** in **Organization → General** or on a deployment. The web server Ployz runs on web-1 is missing or stopped, so web-1 takes no web traffic. When your other servers still take the traffic, your domains stay live and only **Domain** in **Organization → General** says so. Deploy any project to bring it back: ```sh ployz deploy ``` A deploy creates or starts the missing web server on every server that takes web traffic. `ployz deploy --plan` lists it first, as `ingress on web-1: would create`. If it still isn't running, the deploy says why. When several servers are missing it at once, one deploy may bring it back on only some of them, and it lists the rest as not running. Deploy again until none is listed. Bringing the web server back never stops it on a server where it already runs, including one another deploy just brought back. When the web server keeps crashing and restarting, the deploy leaves it alone and warns with the `docker logs` command that shows why. `ployz server add` and enrollment also show its exit code and last line. When something else holds port 80 or 443, see [above](#something-else-answers-on-port-80). ## None of my servers takes web traffic You'll see **No Server receives traffic** when every server has web traffic turned off, and **No ingress Server has a public IP** when the servers that take web traffic have no public address Ployz knows. [Add a server](https://ployz.dev/docs/servers/add-a-server), or open a server and turn on **Takes web traffic** under **Web traffic**. The dashboard can't set a public IP yet, so tell Ployz a server's address from the CLI: ```sh ployz server set web-1 --public-ip 203.0.113.10 ``` A public IP you set this way stays until you set it again: Ployz no longer looks it up when the server restarts. `ployz server set` and `ployz server add` refuse an address no visitor can reach, such as `0.0.0.0`, `127.0.0.1`, `::1` or `fe80::1`. It warns about a private one, such as `10.0.0.5`, which only its own network reaches. `--public-ip none` clears the address, and Ployz doesn't look one up afterwards either. A server that takes web traffic but whose web server is missing or stopped needs a deploy, not these settings. See [Ingress isn't running](#ingress-isnt-running). ## One service can't reach another You'll see errors like `ECONNREFUSED 127.0.0.1:5432` or `getaddrinfo ENOTFOUND api.internal`. - Reference the other service's address, like `${{ postgres.PLOYZ_PRIVATE_DOMAIN }}`, never `localhost`. Its private name is in that service's **Settings → Networking**, in the **Private address** row. - Use the port it listens on inside its container, like `5432`. - For a web service, use `http://`, not `https://`: traffic between services stays on your private network. - A name that doesn't resolve can mean the other service isn't healthy (check its **Logs**) or is in another environment. - Services on different servers need UDP port 51820 open between those servers. See [Private networking](https://ployz.dev/docs/services/private-networking). ## Good to know - **Scripts get JSON.** A request with `Accept: application/json` gets the same code as JSON, like `{"error":{"status":503,"code":"NOT_RUNNING",…}}`, and other clients get plain text. Every error page carries a request id, also in the `ployz-request-id` header. Include it when you report a problem. - **A redirect fails the health check.** Point **Healthcheck path** at a page your app serves directly. See [Deployments](https://ployz.dev/docs/deploy/deployments). - **Private names stay inside one environment.** In a branch, reference `${{ api.PLOYZ_PRIVATE_DOMAIN }}` rather than typing the name.