Reproxy is a simple edge HTTP(s) server / reverse proxy supporting various providers (docker, static, file, consul catalog). One or more providers supply information about the requested server, requested URL, destination URL, and health check URL. It is distributed as a single binary or as a docker container.
Server (host) can be set as FQDN, i.e. `s.example.com`, `*` (catch all) or a regex. Exact match takes priority, so if there are two rules with servers `example.com` and `example\.(com|org)`, request to `example.com/some/url` will match the former. Requested url can be regex, for example `^/api/(.*)` and destination url may have regex matched groups in, i.e. `http://d.example.com:8080/$1`. For the example above `http://s.example.com/api/something?foo=bar` will be proxied to `http://d.example.com:8080/something?foo=bar`.
For convenience, requests with the trailing `/` and without regex groups expanded to `/(.*)`, and destinations in those cases expanded to `/$1`. I.e. `/api/` -> `http://127.0.0.1/service` will be translated to `^/api/(.*)` -> `http://127.0.0.1/service/$1`
Both HTTP and HTTPS supported. For HTTPS, static certificate can be used as well as automated ACME (Let's Encrypt) certificates. Optional assets server can be used to serve static files. Starting reproxy requires at least one provider defined. The rest of parameters are strictly optional and have sane default.
Reproxy distributed as a small self-contained binary as well as a docker image. Both binary and image support multiple architectures and multiple operating systems, including linux_x86_64, linux_arm64, linux_arm, macos_x86_64, macos_arm64, windows_x86_64 and windows_arm. We also provide both arm64 and x86 deb and rpm packages.
- docker container available on [Docker Hub](https://hub.docker.com/r/umputun/reproxy) as well as on [Github Container Registry](https://github.com/users/umputun/packages/container/reproxy/versions). I.e. `docker pull umputun/reproxy` or `docker pull ghcr.io/umputun/reproxy`.
Proxy rules supplied by various providers. Currently included - `file`, `docker`, `static` and `consul-catalog`. Each provider may define multiple routing rules for both proxied request and static (assets). User can sets multiple providers at the same time.
This is the simplest provider defining all mapping rules directly in the command line (or environment). Multiple rules supported. Each rule is 3 or 4 comma-separated elements `server,sourceurl,destination[,ping-url]`. For example:
-`example.com,/foo/bar,https://api.example.com/zzz,https://api.example.com/ping` - proxy all requests to `example.com` and with `/foo/bar` url to `https://api.example.com/zzz` and it sees `https://api.example.com/ping` for the health check.
The last (4th) element defines an optional ping url used for health reporting. I.e.`*,^/api/(.*),https://api.example.com/$1,https://api.example.com/ping`. See [Health check](#ping-and-health-checks) section for more details.
Docker provider supports a fully automatic discovery (with `--docker.auto`) with no extra configuration needed. By default, it redirects all requests like `http://<url>/<container name>/(.*)` to the internal IP of the given container and the exposed port. Only active (running) containers will be detected.
-`reproxy.enabled` - enable (`yes`, `true`, `1`) or disable (`no`, `false`, `0`) container from reproxy destinations.
Pls note: without `--docker.auto` the destination container has to have at least one of `reproxy.*` labels to be considered as a potential destination.
With `--docker.auto`, all containers with exposed port will be considered as routing destinations. There are 3 ways to restrict it:
- Exclude some containers explicitly with `--docker.exclude`, i.e. `--docker.exclude=c1 --docker.exclude=c2 ...`
- Allow only a particular docker network with `--docker.network`
- Set the label `reproxy.enabled=false` or `reproxy.enabled=no` or `reproxy.enabled=0`
If no `reproxy.route` defined, the default route is `^/<container_name>/(.*)`. In case if all proxied source should have the same prefix pattern, for example `/api/(.*)` user can define the common prefix (in this case `/api`) for all container-based routes. This can be done with `--docker.prefix` parameter.
Docker provider also allows to define multiple set of `reproxy.N.something` labels to match multiple distinct routes on the same container. This is useful as in some cases a single container may expose multiple endpoints, for example, public API and some admin API. All the labels above can be used with "N-index", i.e. `reproxy.1.server`, `reproxy.1.port` and so on. N should be in 0 to 9 range.
Consul Catalog provider calls Consul API periodically (every second by default) to obtain services, which has any tag with `reproxy.` prefix. User can redefine check interval with `--consul-catalog.interval` command line flag as well as consul address with `--consul-catalog.address` command line option. The default address is `http://127.0.0.1:8500`.
In case if rules set as a part of docker compose environment, destination with the regex group will conflict with compose syntax. I.e. attempt to use `https://api.example.com/$1` in compose environment will fail due to a syntax error. The standard solution here is to "escape" `$` sign by replacing it with `$$`, i.e. `https://api.example.com/$$1`. This substitution supported by docker compose and has nothing to do with reproxy itself. Another way is to use `@` instead of `$` which is supported on reproxy level, i.e. `https://api.example.com/@1`_
SSL mode (by default none) can be set to `auto` (ACME/LE certificates), `static` (existing certificate) or `none`. If `auto` turned on SSL certificate will be issued automatically for all discovered server names. User can override it by setting `--ssl.fqdn` value(s). In `auto` and `static` SSL mode, Reproxy will automatically add the `X-Forwarded-Proto` and `X-Forwarded-Port` headers. These headers are useful for services behind the proxy to know the original protocol (http or https) and port number used by the client.
Reproxy allows to sanitize (remove) incoming headers by passing `--drop-header` parameter (can be repeated). This parameter can be useful to make sure some of the headers, set internally by the services, can't be set/faked by the end user. For example if some of the services, responsible for the auth, sets `X-Auth-User` and `X-Auth-Token` it is likely makes sense to drop those headers from the incoming requests by passing `--drop-header=X-Auth-User --drop-header=X-Auth-Token` parameter or via environment `DROP_HEADERS=X-Auth-User,X-Auth-Token`
The opposite function, setting outgoing header(s) supported as well. It can be useful in many cases, for example enforcing some custom CORS rules, security related headers and so on. This can be done with `--header` parameter (can be repeated) or env `HEADER`. For example, this is how it can be done with the docker compose:
By default no request log generated. This can be turned on by setting `--logger.enabled`. The log (auto-rotated) has [Apache Combined Log Format](http://httpd.apache.org/docs/2.2/logs.html#combined)
User can also turn stdout log on with `--logger.stdout`. It won't affect the file logging above but will output some minimal info about processed requests, something like this:
Users may turn the assets server on (off by default) to serve static files. As long as `--assets.location` set it treats every non-proxied request under `assets.root` as a request for static files. The assets server can be used without any proxy providers; in this mode, reproxy acts as a simple web server for the static content. Assets server also supports "spa mode" with `--assets.spa` where all not-found request forwarded to `index.html`.
In addition to the common assets server, multiple custom assets servers are supported. Each provider has a different way to define such a static rule, and some providers may not support it at all. For example, multiple asset servers make sense in static (command line provider), file provider, and even useful with docker providers, however it makes very little sense with consul catalog provider.
1. static provider - if source element prefixed by `assets:` or `spa:` it will be treated as file-server. For example `*,assets:/web,/var/www,` will serve all `/web/*` request with a file server on top of `/var/www` directory.
3. docker provider - `reproxy.assets=web-root:location`, i.e. `reproxy.assets=/web:/var/www`. Switching to spa mode done by setting `reproxy.spa` to `yes` or `true`
Assets server supports caching control with the `--assets.cache=<duration>` parameter. `0s` duration (default) turns caching control off. A duration is a sequence of decimal numbers, each with optional fraction and a unit suffix, such as "300ms", "1.5h" or "2h45m". Valid time units are "ns", "us" (or "µs"), "ms", "s", "m", "h" and "d".
2. Custom duration for different mime types. It should include two parts - the default value and the pairs of mime:duration. In command line this looks like multiple `--assets.cache` options, i.e. `--assets.cache=48h --assets.cache=text/html:24h --assets.cache=image/png:2h`. Environment values should be comma-separated, i.e. `ASSETS_CACHE=48h,text/html:24h,image/png:2h`
Serving purely static content is one of the popular use cases. Usually this used for the separate frontend container providing UI only. With the assets server such a container is almost trivial to make. This is an example from the container serving [reproxy.io](http://reproxy.io)
All it needs is to copy stastic assets to some location and passing this location as `"--assets.location` to reproxy entrypoint.
## SPA-friendly mode
Some SPA applications counts on proxy to handle 404 on static asset in a special way, by redirecting it to "/index.html". This is similar to nginx's `try_files $uri $uri/ …` directive and, apparently, this functionality somewhat important for the modern web apps.
This mode is off by default and can be turned on by setting `--assets.spa` or `ASSETS_SPA=true` env.
By default reproxy treats destination as a proxy location, i.e. it invokes http call internally and returns response back to the client. However by prefixing destination url with `@code` this behaviour can be changed to a permanent (status code 301) or temporary (status code 302) redirects. I.e. destination set to `@301 https://example.com/something` with cause permanent http redirect to `Location: https://example.com/something`
-`--timeout.*` various timeouts for both server and proxy transport. See `timeout` section in [All Application Options](#all-application-options). A zero or negative value means there will be no timeout.
In order to eliminate the need to pass custom params/environment, the default `--listen` is dynamic and trying to be reasonable and helpful for the typical cases:
- If anything set by users to `--listen` all the logic below ignored and host:port passed in and used directly.
- If nothing set by users to `--listen` and reproxy runs outside the docker container, the default is `127.0.0.1:80` for http mode (`ssl.type=none`) and `127.0.0.1:443` for ssl mode (`ssl.type=auto` or `ssl.type=static`).
- If nothing set by users to `--listen` and reproxy runs inside the docker, the default is `0.0.0.0:8080` for http mode, and `0.0.0.0:8443` for ssl mode.
-`/ping` responds with `pong` and indicates what reproxy up and running
-`/health` returns `200 OK` status if all destination servers responded to their ping request with `200` or `417 Expectation Failed` if any of servers responded with non-200 code. It also returns json body with details about passed/failed services.
In addition to the endpoints above, reproxy supports optional live health checks. In this case (if enabled), each destination checked for ping response periodically and excluded failed destination routes. It is possible to return multiple identical destinations from the same or various providers, and the only passed picked. If numerous matches were discovered and passed - the final one picked according to `lb-type` strategy (by default random selection).
To turn live health check on, user should set `--health-check.enabled` (or env `HEALTH_CHECK_ENABLED=true`). To customize checking interval `--health-check.interval=` can be used.
Reproxy returns 502 (Bad Gateway) error in case if request doesn't match to any provided routes and assets. In case if some unexpected, internal error happened it returns 500. By default reproxy renders the simplest text version of the error - "Server error". Setting `--error.enabled` turns on the default html error message and with `--error.template` user may set any custom html template file for the error rendering. The template has two vars: `{{.ErrCode}}` and `{{.ErrMessage}}`. For example this template `oh my! {{.ErrCode}} - {{.ErrMessage}}` will be rendered to `oh my! 502 - Bad Gateway`
Reproxy allows to define system level max req/sec value for the overall system activity as well as per user. 0 values (default) treated as unlimited.
User activity limited for both matched and unmatched routes. All unmatched routes considered as a "single destination group" and get a common limiter which is `rate*3`. It means if 10 (req/sec) defined with `--throttle.user=10` the end user will be able to perform up to 30 request pers second for either static assets or unmatched routes. For matched routes this limiter maintained per destination (route), i.e. request proxied to s1.example.com/api will allow 10 r/s and the request proxied to s2.example.com will allow another 10 r/s.
Reproxy supports basic auth for all requests. This is useful for protecting endpoints during the development and testing, before allowing unrestricted access to them. This functionality is disabled by default and not granular enough to allow for per-route auth. I.e. enabled basic auth will affect all requests.
In order to enable basic auth for all requests, user should set the typical htpasswd file with `--basic-htpasswd=<file location>` or `env BASIC_HTPASSWD=<file location>`.
Reproxy allows restricting access to the routes with a list of comma-separated subnets or ips. This is useful for the development and testing, before allowing unrestricted access to them. It also can be used to restrict access to the internal services. By default, all the routes are open for all the clients.
To restrict access to the routes, user should set appropriate keys for the routes, i.e. `reproxy.remote` for docker and consul, and `remote` for file provider. The value should be a list of comma-separated subnets or ips or subnets. For example `127.0.0.1, 192.168.1.0/24`. For more details see [docker provider](#docker-provider) and [consul catalog provider](#consul-catalog-provider) sections.
By default, reproxy will check the remote address from the client's request. However, in some cases, it won't work as expected, for example behind of other proxy, or with docker bridge network. This can be altered with `--remote-lookup-headers` parameter allowing check the value of the header `X-Real-IP` or `X-Forwarded-For` (in this order) and use it for the check. If the header is not set, the check will be performed against the remote address of the client.
Checking headers should be used with caution, as it is possible to fake them. However, in some cases, it is the only way to get the real remote address of the client. Generally, it is recommended to use this option only if user is completely controlling all the headers and can guarantee the headers are not faked.
The core functionality of reproxy can be extended with external plugins. Each plugin is an independent process/container implementing [rpc server](https://golang.org/pkg/net/rpc/). Plugins registered with reproxy conductor and added to the chain of the middlewares. Each plugin receives request with the original url, headers and all matching route info and responds with the headers and the status code. Any status code >= 400 treated as an error response and terminates flow immediately with the proxy error. There are two types of headers plugins can set:
-`HeadersIn` - incoming headers. Those will be sent to the proxied url
-`HeadersOut` - outgoing headers. Will be sent back to the client
By default headers set by a plugin will be mixed with the original headers. In case if plugin need to control all the headers, for example drop some of them, `OverrideHeaders*` field can be set by a plugin indicating to the core reproxy process the need to overwrite all the headers instead of mixing them in.
To simplify the development process all the building blocks provided. It includes `lib.Plugin` handling registration, listening and dispatching calls as well as `lib.Request` and `lib.Response` defining input and output. Plugin's authors should implement concrete handlers satisfying `func(req lib.Request, res *lib.HandlerResponse) (err error)` signature. Each plugin may contain multiple handlers like this.
By default, the reproxy container runs under the root user to simplify the initial setup and access the docker's socket. This is needed to allow the docker provider discovery of the running containers. However, if such a discovery is not required or the docker provider not in use, it is recommended to change the user to some less-privileged one. It can be done on the docker-compose level and on docker level with `user` option.
Sometimes, even with inside-the-docker routing, it makes sense to disable the docker provider and setup rules with either static or file provider. All the containers running within a compose sharing the same network and accessible via local DNS. User can have a rule like this to avoid docker discovery: `- STATIC_RULES=*,/api/email/(.*),http://email-sender:8080/$$1`. This rule expects `email-sender` container defined inside the same compose. Please note: users can achieve the same result by using the docker network even if the destination service was defined in a different compose file. This way reproxy configuration can stay separate from the actual services.
There is nothing except reproxy binary inside the reproxy container, as it builds on top of an empty (scratch) image.
Each option can be provided in two forms: command line or environment key:value pair. Some command line options have a short form, like `-l localhost:8080` and all of them have the long form, i.e `--listen=localhost:8080`. The environment key (name) [listed](#all-application-options) for each option as a suffix, i.e. `[$LISTEN]`.
All size options support unit suffixes, i.e. 10K (or 10k) for kilobytes, 16M (or 16m) for megabytes, 10G (or 10g) for gigabytes. Lack of any suffix (i.e. 1024) means bytes.
Some options are repeatable, in this case user may pass it multiple times with the command line, or comma-separated in env. For example `--ssl.fqdn` is such an option and can be passed as `--ssl.fqdn=a1.example.com --ssl.fqdn=a2.example.com` or as env `SSL_ACME_FQDN=a1.example.com,a2.example.com`
The project is under active development and may have breaking changes till `v1` is released. However, we are trying our best not to break things unless there is a good reason. As of version 0.4.x, reproxy is considered good enough for real-life usage, and many setups are running it in production.
