What problem it solves
One VPS runs several services in containers, each on its own port: memos notes on 5230, pulse monitoring on 7655, plus a website. From the outside you want something else: names on the regular port 443, a real certificate, and the services’ own ports closed.
Without a proxy
http://203.0.113.10:5230— memoshttp://203.0.113.10:7655— pulsehttp://203.0.113.10:8080— the site
No HTTPS, every port open to the outside, each service with its own protection and logs, or none.
With a reverse proxy
https://memos.example.orghttps://pulse.example.orghttps://example.org
Only 80 and 443 open to the outside. Certificates, bans, admin passwords and the request log — in one place.
Only one program on a server can hold port 443. So one goes in front: it takes every request, looks at which name is being asked for, and passes the request to the right container. That’s a reverse proxy. Since all traffic goes through it, it’s also where the shared pieces live: certificates, the HTTP-to-HTTPS redirect, bans on attacking addresses, admin passwords, rate limits, and one log for all services.
There are many reverse proxies; the comparison is below. What sets Traefik apart is that it takes routes from Docker itself: a service declares its name with labels in its own compose file, and Traefik sees it as soon as the container starts.
A request’s path: entry point, router, middleware, service
The whole config rests on these four concepts. A request passes them in order:
browser · https://memos.example.org
- Entry point
https · :443A port Traefik listens on. It accepts the connection and presents the certificate; the shared HTTP redirect lives here too. - Router
Host(`memos.example.org`)A rule for “which requests are mine”. If no router matches, Traefik answers404 page not found. - Middleware
crowdsec, ratelimitChecks along the way, in the listed order: a ban gives 403, a rate limit 429; passwords, headers. - Service
memos:5230Where to send it: the container’s address and port. With several containers Traefik splits requests between them.
memos container · proxy network · port not published
Two kinds of config
Traefik’s settings come in two parts with different rules. Most “why didn’t it apply” questions come from here.
Static
traefik.yml- Defines entry points, where routes come from, certificates, logs, plugins.
- Applied only at startup: a change means restarting Traefik, and every site blinks.
${…}variables aren’t substituted (traefik.yml doesn’t substitute variables).
Dynamic
labels, dynamic/*.yml- Defines routers, middleware, services, TLS options.
- Applied live: Traefik watches Docker and the files.
- Compose substitutes variables in labels.
Traefik gets dynamic config from providers. On a single server with compose, two are enough: docker reads the
labels of running containers through docker.sock, and file reads YAML files from the dynamic/ directory. That’s
where you put what many services need at once: the shared ban middleware, rate limits, TLS options. There are also
providers for Kubernetes, Swarm, Consul and others.
An object’s name carries its provider as a suffix: crowdsec@file is a middleware from a file, memos@docker a
router from labels. A reference to an object from another provider needs the suffix; otherwise Traefik looks for it
in its own provider and doesn’t find it.
Where it’s used
Your own server with services in Docker
Each stack in /opt/<name>/ declares its name and protection in its own compose file. Remove the stack and the
route goes with it.
Several domains on one server
A wildcard certificate for each domain is requested on the entry point and covers all subdomains at once (Traefik and Let’s Encrypt).
Protection in one place
A CrowdSec ban via plugin (CrowdSec with Traefik behind Cloudflare), Basic Auth on admin panels, rate limits — middleware you attach to a router with one line.
A log of every request
The JSON access log is read by CrowdSec, vector and Loki (Traefik logs).
TCP services by name
Sometimes traffic has to pass through as is, without decryption: a TCP router with HostSNI and tls.passthrough
picks the recipient by the name in the TLS handshake on the same port 443.
Kubernetes and Swarm
An Ingress and Gateway API controller, installed by default in k3s. Ingress NGINX stopped getting updates in March 2026, and some clusters are moving off it.
Other ways to solve it
The alternatives are described from their documentation as of October 2026 and weren’t tested on a server.
| Where routes live | Certificates | Sees containers | |
|---|---|---|---|
| Traefik | labels in each service’s compose file, plus YAML files | Let’s Encrypt built in: HTTP-01, TLS-ALPN-01, DNS-01; Cloudflare and other DNS providers in the stock image | yes |
| Caddy | one Caddyfile, 2–3 lines per site: memos.example.org { reverse_proxy memos:5230 } |
HTTPS by default for any name in the config, Let’s Encrypt and ZeroSSL; wildcard only with a build for your DNS provider | via the third-party caddy-docker-proxy |
| nginx | server { … } files, applied with nginx -s reload |
separately: certbot or acme.sh with its own timer | no |
| Nginx Proxy Manager | a web UI on port 81 over nginx; settings in its SQLite database | Let’s Encrypt with a button in the UI | no |
| HAProxy | haproxy.cfg: frontends and backends by hand |
ACME since 3.2 (HTTP-01), DNS-01 since 3.3 via the Data Plane API | no |
| If… | Use |
|---|---|
| services live in Docker, come and go, and you want the route next to the service | Traefik |
| two or three sites and you want one short file | Caddy |
| nginx is already there, or you also need a static file server with cache tuning | nginx + certbot |
| you won’t write configs by hand and have only a few services; settings live in a database you can’t put in git (Server configs in git) | Nginx Proxy Manager |
| you balance load across many servers and fine-tune TCP | HAProxy |
What you get and what it costs
What you get
- The route lives next to the service. A new stack brings its own labels; the shared proxy config stays untouched.
- No restarts. Traefik listens to Docker events and the
dynamic/directory and rebuilds routes at most once every 2 s (providersThrottleDuration). Other sites aren’t interrupted. - Certificates on its own. It obtains them, stores them in
acme.jsonand renews 30 days before expiry. A wildcard through Cloudflare works with the officialtraefik:v3.7image, no custom build (Traefik and Let’s Encrypt). - Middleware out of the box: passwords, rate limits, redirects, headers, IP allowlists, compression. Plugins add what’s missing, such as the CrowdSec bouncer (CrowdSec with Traefik behind Cloudflare).
- A JSON log — a ready source for CrowdSec and Loki (Traefik logs).
What it costs
- Nobody checks the labels. Traefik silently accepts a typo in a router name and creates another router (Middleware on a missing router), and a second label with the same key overwrites the first (A router’s middlewares go in one line). The error shows only in the dashboard or the log.
- The config is spread across stacks. To answer “what’s open to the outside right now”, you go through every compose file or turn on the dashboard.
- Two kinds of config with different rules: the static one only via restart and without variables (traefik.yml doesn’t substitute variables). The first service takes longer to bring up than in Caddy, where two lines suffice.
- It needs
docker.sock, and socket access is root on the host (docker.sock). - Upgrades break things. v3 changed rule syntax:
HostRegexpnow uses Go regexes,HeadersbecameHeader. With Docker 29, Traefik before 3.6.1 stopped seeing containers. So pin the minor version in the tag:traefik:v3.7.
Where to start
- Traefik: what it is and when to pick it
- Traefik from scratch: the first service by name, with HTTPSBring up the proxy and open the first two services. Plus which changes apply live and which need a restart.
- traefik.yml block by block: what each setting meansEvery line with its default and when to change it.
- Traefik labels: router, middleware, serviceEveryday use: connect a service, a password, a rate limit, TCP.
- Traefik and Let’s Encrypt: wildcard, ACME, a certificate for neighboursOne certificate for every subdomain, the DNS challenge.
- CrowdSec with Traefik behind Cloudflare: two bouncers and X-Forwarded-ForBans at the proxy level, sites behind Cloudflare.
- Traefik logs: CrowdSec, vector, Loki, rotationWho reads the access log and how not to lose lines on rotation.
If it doesn’t work
A blank 404 page not found instead of the site
- Cause
- No router matched: the container lacks
traefik.enable=true(withexposedByDefault: falseTraefik doesn’t see it without that), there’s a typo inHost, the container isn’t running, or the router was disabled by an error — for example, a reference to a middleware that doesn’t exist. - Fix
- Check
docker logs traefik --since 10m 2>&1 | grep -iE 'err|error'and compare the name in the rule with the address bar.
The browser complains about the certificate, which says TRAEFIK DEFAULT CERT
- Cause
- There’s no matching certificate, so Traefik served its self-signed one: port 80 is closed for HTTP-01, the DNS token lacks permissions, or a Let’s Encrypt limit was hit (5 failed validations per name per hour).
- Fix
- Check
docker logs traefik 2>&1 | grep -i acme | tail -20; details — Traefik and Let’s Encrypt.
Gateway Timeout after about 30 seconds
- Cause
- Traefik knows the container but can’t reach it: the container isn’t on the shared
proxynetwork, or it’s on several networks and Traefik picked the wrong one. 30 s is the defaultdialTimeout. - Fix
- Add the container to the
proxynetwork and name it explicitly:traefik.docker.network=proxy.
Bad Gateway right away
- Cause
- The network is fine, but nothing listens on that port:
loadbalancer.server.portis wrong, or the service inside the container isn’t up yet. - Fix
- Compare the port with the image’s documentation and look at the service’s
docker logs.
After a Docker upgrade every site returns 404, and Traefik’s log says client version 1.24 is too old
- Cause
- Docker 29 raised the minimum API version to 1.44, while Traefik before 3.6.1 negotiates 1.24.
- Fix
- Use
traefik:v3.6.1or newer: 3.6.1 added API version negotiation.
A change in traefik.yml has no effect
- Cause
- Traefik reads the static config only at startup.
- Fix
docker compose restart traefik.
A large upload breaks off at about one minute
- Cause
- The entry point’s default
readTimeoutis 60 s — Traefik must read the whole request, body included, within that time. - Fix
- Raise
transport.respondingTimeouts.readTimeouton the entry point (traefik.yml block by block) and restart Traefik.