webdis

A Redis HTTP interface with JSON output

nicolasff
2853
304
C

Build

About Webdis

A very simple web server providing an HTTP interface to Redis. It embeds hiredis, jansson (with some local changes), and http-parser. It also depends on libevent, to be installed separately.

Build and run from source

Building Webdis requires the libevent development package. You can install it on Ubuntu by typing sudo apt-get install libevent-dev or on macOS by typing brew install libevent.

To build Webdis with support for encrypted connections to Redis, see Building Webdis with SSL support.

$ make clean all

$ ./webdis &

$ curl http://127.0.0.1:7379/SET/hello/world
→ {"SET":[true,"OK"]}

$ curl http://127.0.0.1:7379/GET/hello
→ {"GET":"world"}

$ curl -d "GET/hello" http://127.0.0.1:7379/
→ {"GET":"world"}

Configuration

Webdis is configured by a configuration file in JSON format, provided to the server as a command-line parameter:

./webdis /path/to/webdis.json

At start-up, Webdis will look for a file named webdis.json in the current directory if no file is specified on the command line.

This repository comes with a sample configuration file named webdis.json, to be used for local evaluation. Another file, webdis.prod.json, is provided as a starting point to build a production configuration file. Do not use either of these files in production without reviewing them.

The various features of Webdis are documented in this README.

Try in Docker

$ docker run --name webdis-test --rm -d -p 127.0.0.1:7379:7379 nicolas/webdis
0d2ce311a4834d403cc3e7cfd571b168ba40cede6a0e155a21507bb0bf7bee81

$ curl http://127.0.0.1:7379/PING
{"PING":[true,"PONG"]}

# To stop it:
$ docker stop webdis-test
0d2ce311a483

Docker repositories and Docker Content Trust

Webdis images are published on Docker Hub and Amazon ECR.

Docker Hub

$ docker pull nicolas/webdis:0.1.22
$ docker pull nicolas/webdis:latest

Starting from release 0.1.12 and including latest, Docker Hub images are signed (download public key). You should see the following key ID if you verify the trust:

$ docker trust inspect nicolas/webdis:0.1.22 --pretty

Signatures for nicolas/webdis:0.1.22

SIGNED TAG   DIGEST                                                             SIGNERS
0.1.22       5a7d342e3a9e5667fe05f045beae4b5042681d1d737f60843b7dfd11f96ab72f   (Repo Admin)

List of signers and their keys for nicolas/webdis:0.1.22

SIGNER      KEYS
nicolasff   dd0768b9d35d

Administrative keys for nicolas/webdis:0.1.22

  Repository Key:	fed0b56b8a8fd4d156fb2f47c2e8bd3eb61948b72a787c18e2fa3ea3233bba1a
  Root Key:	40be21f47831d593892370a8e3fc5bfffb16887c707bd81a6aed2088dc8f4bef

The signing keys are listed on this documentation page; please make sure they match what you see.
The same documentation page details how to verify the signatures of multi-architecture images, and the tree of manifests used to build them.

Amazon Elastic Container Registry (ECR)

$ docker pull public.ecr.aws/nicolas/webdis:0.1.22
$ docker pull public.ecr.aws/nicolas/webdis:latest

A note on ECR and trust: AWS does not support Notary v2 at the time of this writing, although a security talk from 2020 mentions that the feature could be available in 2021.

The consequence is that Webdis images on ECR are not signed at this time.

They can still be verified, since the images uploaded there use the exact same hash as the ones on Docker Hub, which are signed. This means that you can verify the signature using the docker trust inspect command described above, as long as you also make sure that the image hash associated with the image on ECR matches the one shown on Docker Hub.

For more details about Content Trust validation with ECR images, refer to the article titled Webdis and Docker Content Trust in the Webdis documentation.

Multi-architecture images

Starting with release 0.1.19, Docker images for Webdis are published as manifest lists supporting multiple architectures. Each release points to an x86-64 image and an ARM64v8 image:

$ docker manifest inspect nicolas/webdis:0.1.19 | jq -r '.manifests | .[] | .platform.architecture + " -> " + .digest'
amd64 -> sha256:2ced2d99146e1bcaf10541d17dbac573cffd02237c3b268875be1868138d3b54
arm64 -> sha256:d026c5675552947b6a755439dfd58360e44a8860436f4eddfe9b26d050801248

By default docker pull will download only the relevant image for your architecture, but you can specify the platform to download the image for a specific architecture, e.g.

$ docker pull nicolas/webdis:0.1.19 --platform linux/arm64/v8

Build and run a Docker image locally

Clone the repository and open a terminal in the webdis directory, then run:

$ docker build -t webdis:custom .
[...]

$ docker run --name webdis-test --rm -d -p 127.0.0.1:7379:7379 webdis:custom
f0a2763fd456ac1f7ebff80eeafd6a5cd0fc7f06c69d0f7717fb2bdcec65926e

$ curl http://127.0.0.1:7379/PING
{"PING":[true,"PONG"]}

To stop it:

$ docker stop webdis-test
f0a2763fd456

Docker images and embedded Redis

ℹ️ The Docker images provided on Docker Hub under nicolas/webdis contain both Webdis and an embedded Redis server. They were built this way to make it easy to try Webdis without having to configure a Docker deployment with two containers, but this is likely not the best way to run Webdis in production.

The following documentation pages cover various such use cases:

More articles are available in the Webdis documentation.

Building Webdis with SSL support

Webdis needs libraries that provide TLS support to encrypt its connections to Redis:

  • On Alpine Linux, install openssl-dev with apk-add openssl-dev.
  • On Ubuntu, install libssl-dev with apt-get install libssl-dev.
  • On macOS with HomeBrew, install OpenSSL with brew install [email protected].

Then, build Webdis with SSL support enabled:

$ make SSL=1

Configuring Webdis with SSL

Once Redis is configured with SSL support (see this guide for step-by-step instructions), you can configure Webdis to connect to Redis over encrypted connections.

Add a block to webdis.json under a key named "ssl" placed at the root level, containing the following object:

{
    "enabled": true,
    "ca_cert_bundle": "/path/to/ca.crt",
    "path_to_certs": "/path/to/trusted/certs",
    "client_cert": "/path/to/redis.crt",
    "client_key": "/path/to/redis.key",
    "redis_sni": "redis.mydomain.tld"
}

This means that "ssl" should be at the same level as "redis_host", "redis_port", etc.

Important: the presence of the "ssl" configuration block alone does not necessarily enable secure connections to Redis. The key "enabled" inside this block must also be set to true, otherwise Webdis will keep using unencrypted connections.

Use the following table to match the Redis configuration keys to the fields under "ssl" in webdis.json:

Redis field Webdis field Purpose
tls-cert-file client_cert Client certificate
tls-key-file client_key Client key
tls-ca-cert-file ca_cert_bundle CA certificate bundle

Two other keys have no equivalent in redis.conf:

  • path_to_certs is an optional directory path where trusted CA certificate files are stored in an OpenSSL-compatible format.
  • redis_sni is an optional Redis server name, used as a server name indication (SNI) TLS extension.

See also the Hiredis docs and Hiredis source code for more information.

Running Redis and Webdis with SSL in Docker Compose

For a full tutorial showing how to configure and run Redis and Webdis under Docker Compose with SSL connections between the two services, head to the docs folder and open Running Webdis & Redis in Docker Compose with SSL connections.

SSL troubleshooting

Follow this table to diagnose issues with SSL connections to Redis.

Error message or issue Cause Solution
Unexpected key or incorrect value in webdis.json: ‘ssl’ Webdis is not compiled with SSL support Build webdis with make SSL=1
Unexpected key or incorrect value under ‘ssl’ Invalid configuration One or more keys in the ssl object in was not recognized, make sure they are all valid
Failed to load client certificate Invalid client certificate Verify the file that client_cert points to
Failed to load private key Invalid client key Verify the file that client_key points to
Failed to load CA Certificate or CA Path Invalid CA certificate bundle Verify the file that ca_cert_bundle points to
All requests fail with HTTP 503, logs show “Error disconnecting: Connection reset by peer” SSL disabled in config but Webdis connected to an SSL port Make sure enabled is set to true and that Webdis connects to the SSL port for Redis
Logs show “Server closed the connection” at start-up SSL connection failed The client key and/or client certificate was missing. Make sure the configuration is valid.
No error but all requests hang Webdis connected to the non-SSL port Make sure Webdis is connecting to the port set under tls-port in redis.conf

Features

  • GET and POST are supported, as well as PUT for file uploads (see example of PUT usage here).
  • JSON output by default, optional JSONP parameter (?jsonp=myFunction or ?callback=myFunction).
  • Raw Redis 2.0 protocol output with .raw suffix.
  • MessagePack output with .msg suffix.
  • HTTP 1.1 pipelining (70,000 http requests per second on a desktop Linux machine.)
  • Multi-threaded server, configurable number of worker threads.
  • WebSocket support (Currently using the specification from RFC 6455).
  • Connects to Redis using a TCP or UNIX socket.
  • Support for secure connections to Redis (requires Redis 6 or newer).
  • Support for “Keep-Alive” connections to Redis: add "hiredis": { "keep_alive_sec": 15 } to webdis.json to enable it with the default value. See the Hiredis documentation for details, the value configured in webdis.json is the interval passed to redisEnableKeepAliveWithInterval. Important: note how it is used to set the value for TCP_KEEPALIVE (the same value) and to compute the value for TCP_KEEPINTVL (integer, set to 1/3 × interval).
  • Restricted commands by IP range (CIDR subnet + mask) or HTTP Basic Auth, returning 403 errors.
  • Support for Redis authentication in the config file: set redis_auth to a single string to use a password value, or to an array of two strings to use username+password auth (new in Redis 6.0).
  • Environment variables can be used as values in the config file, starting with $ and in all caps (e.g. $REDIS_HOST).
  • Pub/Sub using Transfer-Encoding: chunked, works with JSONP as well. Webdis can be used as a Comet server.
  • Drop privileges on startup.
  • Custom Content-Type using a pre-defined file extension, or with ?type=some/thing.
  • URL-encoded parameters for binary data or slashes and question marks. For instance, %2f is decoded as / but not used as a command separator.
  • Logs, with a configurable verbosity.
  • Configurable fsync frequency for the log file:
    • Set "log_fsync": "auto" (default) to let the file system handle file persistence on its own.
    • Set "log_fsync": N where N is a number to call fsync every N milliseconds.
    • Set "log_fsync": "all" (very slow) to persist the log file to its storage device on each log message.
  • Cross-origin requests, usable with XMLHttpRequest2 (Cross-Origin Resource Sharing - CORS).
  • File upload with PUT.
  • With the JSON output, the return value of INFO is parsed and transformed into an object.
  • Optionally run as a daemon process: set "daemonize": true and "pidfile": "/var/run/webdis.pid" in webdis.json.
  • Default root object: Add "default_root": "/GET/index.html" in webdis.json to substitute the request to / with a Redis request.
  • HTTP request limit with http_max_request_size (in bytes, set to 128 MB by default).
  • Database selection in the URL, using e.g. /7/GET/key to run the command on DB 7.

Ideas, TODO…

  • Add better support for PUT, DELETE, HEAD, OPTIONS? How? For which commands?
    • This could be done using a “strict mode” with a table of commands and the verbs that can/must be used with each command. Strict mode would be optional, configurable. How would webdis know of new commands remains to be determined.
  • MULTI/EXEC/DISCARD/WATCH are disabled at the moment; find a way to use them.
  • Support POST of raw Redis protocol data, and execute the whole thing. This could be useful for MULTI/EXEC transactions.
  • Enrich config file:
    • Provide timeout (maybe for some commands only?). What should the response be? 504 Gateway Timeout? 503 Service Unavailable?
  • Multi-server support, using consistent hashing.
  • SSL/TLS?
    • It makes more sense to terminate SSL with nginx used as a reverse-proxy.
  • SPDY?
    • SPDY is mostly useful for parallel fetches. Not sure if it would make sense for Webdis.
  • Send your ideas using the github tracker, on twitter @yowgi or by e-mail to [email protected].

HTTP error codes

  • Unknown HTTP verb: 405 Method Not Allowed.
  • Redis is unreachable: 503 Service Unavailable.
  • Matching ETag sent using If-None-Match: 304 Not Modified.
  • Could also be used:
    • Timeout on the redis side: 503 Service Unavailable.
    • Missing key: 404 Not Found.
    • Unauthorized command (disabled in config file): 403 Forbidden.

Command format

The URI /COMMAND/arg0/arg1/.../argN.ext executes the command on Redis and returns the response to the client. GET, POST, and PUT are supported:

  • GET /COMMAND/arg0/.../argN.ext
  • POST / with COMMAND/arg0/.../argN in the HTTP body.
  • PUT /COMMAND/arg0.../argN-1 with argN in the HTTP body (see section on file uploads.)

.ext is an optional extension; it is not read as part of the last argument but only represents the output format. Several formats are available (see below).

Special characters: / and . have special meanings, / separates arguments and . changes the Content-Type. They can be replaced by %2f and %2e, respectively.

Redis authentication

Webdis can connect to a Redis server that requires credentials.
For Redis versions before 6.0, provide the password as a single string in webdis.json using the key "redis_auth". For example:

    "redis_auth": "enter-password-here"

Redis 6.0 introduces a more granular access control system and switches from a single password to a pair of username and password. To use these two values with Webdis, set "redis_auth" to an array containing the two strings, e.g.

    "redis_auth": ["my-username", "my-password"]

This new authentication system is only supported in Webdis 0.1.13 and above.

ACL

Access control is configured in webdis.json. Each configuration tries to match a client profile according to two criteria:

Each ACL contains two lists of commands, enabled and disabled. All commands being enabled by default, it is up to the administrator to disable or re-enable them on a per-profile basis.

Examples:

{
    "disabled": ["DEBUG", "FLUSHDB", "FLUSHALL"],
},

{
    "http_basic_auth": "user:password",
    "disabled":        ["DEBUG", "FLUSHDB", "FLUSHALL"],
    "enabled":         ["SET"]
},

{
    "ip":      "192.168.10.0/24",
    "enabled": ["SET"]
},

{
    "http_basic_auth": "user:password",
    "ip":              "192.168.10.0/24",
    "enabled":         ["SET", "DEL"]
}

ACLs are interpreted in order, later authorizations superseding earlier ones if a client matches several. The special value “*” matches all commands.

ACLs and Websocket clients

These rules apply to WebSocket connections as well, although without support for HTTP Basic Auth filtering. IP filtering is supported.

For JSON-based WebSocket clients, a rejected command will return this object (sent as a string in a binary frame):

{"message": "Forbidden", "error": true, "http_status": 403}

The http_status code is an indicator of how Webdis would have responded if the client had used HTTP instead of a WebSocket connection, since WebSocket messages do not inherently have a status code.

For raw Redis protocol WebSocket clients, a rejected command will produce this error (sent as a string in a binary frame):

-ERR Forbidden\r\n

Environment variables

Environment variables can be used in webdis.json to read values from the environment instead of using constant values.
For this, the value must be a string starting with a dollar symbol and written in all caps. For example, to make the redis host and port configurable via environment variables, use the following:

{
    "redis_host": "$REDIS_HOST",
    "redis_port": "$REDIS_PORT",
}

JSON output

JSON is the default output format. Each command returns a JSON object with the command as a key and the result as a value.

Examples:

// string
$ curl http://127.0.0.1:7379/GET/y
{"GET":"41"}

// number
$ curl http://127.0.0.1:7379/INCR/y
{"INCR":42}

// list
$ curl http://127.0.0.1:7379/LRANGE/x/0/1
{"LRANGE":["abc","def"]}

// status
$ curl http://127.0.0.1:7379/TYPE/y
{"TYPE":[true,"string"]}

// error, which is basically a status
$ curl http://127.0.0.1:7379/MAKE-ME-COFFEE
{"MAKE-ME-COFFEE":[false,"ERR unknown command 'MAKE-ME-COFFEE'"]}

// JSONP callback:
$ curl  "http://127.0.0.1:7379/TYPE/y?jsonp=myCustomFunction"
myCustomFunction({"TYPE":[true,"string"]})

RAW output

This is the raw output of Redis; enable it with the .raw suffix.

// string
$ curl http://127.0.0.1:7379/GET/z.raw
$5
hello

// number
$ curl http://127.0.0.1:7379/INCR/a.raw
:2

// list
$ curl http://127.0.0.1:7379/LRANGE/x/0/-1.raw
*2
$3
abc
$3
def

// status
$ curl http://127.0.0.1:7379/TYPE/y.raw
+zset

// error, which is basically a status
$ curl http://127.0.0.1:7379/MAKE-ME-COFFEE.raw
-ERR unknown command 'MAKE-ME-COFFEE'

Custom content-type

Several content-types are available:

  • .json for application/json (this is the default Content-Type).
  • .msg for application/x-msgpack. See https://msgpack.org/ for the specs.
  • .txt for text/plain
  • .html for text/html
  • .xhtml for application/xhtml+xml
  • .xml for text/xml
  • .png for image/png
  • .jpg or .jpeg for image/jpeg
  • Any other with the ?type=anything/youwant query string.
  • Add a custom separator for list responses with ?sep=, query string.
$ curl -v "http://127.0.0.1:7379/GET/hello.html"
[...]
< HTTP/1.1 200 OK
< Content-Type: text/html
< Date: Mon, 03 Jan 2011 20:43:36 GMT
< Content-Length: 137
<
<!DOCTYPE html>
<html>
[...]
</html>

$ curl -v "http://127.0.0.1:7379/GET/hello.txt"
[...]
< HTTP/1.1 200 OK
< Content-Type: text/plain
< Date: Mon, 03 Jan 2011 20:43:36 GMT
< Content-Length: 137
[...]

$ curl -v "http://127.0.0.1:7379/GET/big-file?type=application/pdf"
[...]
< HTTP/1.1 200 OK
< Content-Type: application/pdf
< Date: Mon, 03 Jan 2011 20:45:12 GMT
[...]

File upload

Webdis supports file upload using HTTP PUT. The command URI is slightly different, as the last argument is taken from the HTTP body.
For example: instead of /SET/key/value, the URI becomes /SET/key and the value is the entirety of the body. This works for other commands such as LPUSH, etc.

Uploading a binary file to webdis:

$ file redis-logo.png
redis-logo.png: PNG image, 513 x 197, 8-bit/color RGBA, non-interlaced

$ wc -c redis-logo.png
16744 redis-logo.png

$ curl -v --upload-file redis-logo.png http://127.0.0.1:7379/SET/logo
[...]
> PUT /SET/logo HTTP/1.1
> User-Agent: curl/7.19.7 (x86_64-pc-linux-gnu) libcurl/7.19.7 OpenSSL/0.9.8k zlib/1.2.3.3 libidn/1.15
> Host: 127.0.0.1:7379
> Accept: */*
> Content-Length: 16744
> Expect: 100-continue
>
< HTTP/1.1 100 Continue
< HTTP/1.1 200 OK
< Content-Type: application/json
< ETag: "0db1124cf79ffeb80aff6d199d5822f8"
< Date: Sun, 09 Jan 2011 16:48:19 GMT
< Content-Length: 19
<
{"SET":[true,"OK"]}

$ curl -vs http://127.0.0.1:7379/GET/logo.png -o out.png
> GET /GET/logo.png HTTP/1.1
> User-Agent: curl/7.19.7 (x86_64-pc-linux-gnu) libcurl/7.19.7 OpenSSL/0.9.8k zlib/1.2.3.3 libidn/1.15
> Host: 127.0.0.1:7379
> Accept: */*
>
< HTTP/1.1 200 OK
< Content-Type: image/png
< ETag: "1991df597267d70bf9066a7d11969da0"
< Date: Sun, 09 Jan 2011 16:50:51 GMT
< Content-Length: 16744

$ md5sum redis-logo.png out.png
1991df597267d70bf9066a7d11969da0  redis-logo.png
1991df597267d70bf9066a7d11969da0  out.png

The file was uploaded and re-downloaded properly: it has the same hash and the content-type was set properly thanks to the .png extension.

WebSockets

Webdis supports WebSocket clients implementing RFC 6455.

Important: WebSocket support is currently disabled by default. To enable WebSocket support, set the key named "websockets" to value true in webdis.json, e.g.

{
    "daemonize": false,
    "websockets": true,
}

(start and end of file omitted).

WebSockets are supported with the following formats, selected by the connection URL:

  • JSON (on / or /.json)
  • Raw Redis wire protocol (on /.raw)

Example:

function testJSON() {
    var jsonSocket = new WebSocket("ws://127.0.0.1:7379/.json");
    jsonSocket.onmessage = function(messageEvent) {
        console.log("JSON received:", messageEvent.data);
    };
    jsonSocket.onopen = function() {
        console.log("JSON socket connected!");
        jsonSocket.send(JSON.stringify(["SET", "hello", "world"]));
        jsonSocket.send(JSON.stringify(["GET", "hello"]));
    };
}
testJSON();

This produces the following output:

JSON socket connected!
JSON received: {"SET":[true,"OK"]}
JSON received: {"GET":"world"}

WebSockets HTML demo

The Webdis repository contains a demo web page with JavaScript code that can be used to test WebSocket support.

In a terminal, check out Webdis, build it, and configure it with WebSocket support:

$ cd ~/src/webdis
$ make
$ vim webdis.json      # (edit the file to add "websockets": true)
$ grep websockets webdis.json
    "websockets": true,
$ ./webdis

Then go to the tests/ directory and open websocket.html with a web browser.

Pub/Sub with chunked transfer encoding

Webdis exposes Redis PUB/SUB channels to HTTP clients, forwarding messages in the channel as they are published by Redis. This is done using chunked transfer encoding.

Example using XMLHttpRequest:

var previous_response_length = 0
xhr = new XMLHttpRequest()
xhr.open("GET", "http://127.0.0.1:7379/SUBSCRIBE/hello", true);
xhr.onreadystatechange = checkData;
xhr.send(null);

function checkData() {
    if(xhr.readyState == 3)  {
        response = xhr.responseText;
        chunk = response.slice(previous_response_length);
        previous_response_length = response.length;
        console.log(chunk);
    }
};

Publish messages to redis to see output similar to the following:

{"SUBSCRIBE":["subscribe","hello",1]}
{"SUBSCRIBE":["message","hello","some message"]}
{"SUBSCRIBE":["message","hello","some other message"]}