Tortuga

Tortuga is an HTTP CGI-specific server written in Rust.

Badges

Book

For design goals, non-goals and more see the Tortuga Web Server Book.

RFC

This project attempts to implement the CGI RFC.

Testing

Local Install

To test the command-line interface, install the crate locally from the root of the repository with:

cargo install --path ./

Cargo Tests

To run the unit and documentation tests, use:

cargo test

Endianness

While the system sends all numbers in network byte order (i.e., big endian), WebAssembly uses little-endian for its numbers. Therefore, the system will handle mapping the integers between the types of endianness. See https://tools.ietf.org/html/draft-newman-network-byte-order-01

Examples

Some basic CGI programs can be found in the /examples directory.

Versioning

Tortuga adheres to Semantic Versioning. You can use tortuga version or tortuga -V to determine the version of a Tortuga installation.

Benchmark

All benchmarking code was run on a 16-core Codespace with the default image. Apache was installed using the apt package and wrk was installed by building from source.

Apache

Configured Apache to serve the debug.cgi script using the default configuration to get a baseline for the performance we want to aim for without much load.

Script:

sudo cp examples/*.cgi /usr/lib/cgi-bin
wrk -t5 -c5 -d30s -R20 --latency 'http://localhost/cgi-bin/debug.cgi/extra/path?foo+bar+--me%202'
wrk -t5 -c5 -d30s -R340 --latency 'http://localhost/cgi-bin/empty.cgi/extra/path?foo+bar+--me%202'

Output:

Running 30s test @ http://localhost/cgi-bin/debug.cgi/extra/path?foo+bar+--me%202
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    78.12ms    2.29ms  97.98ms   81.50%
    Req/Sec     3.83      2.88     6.00    100.00%
  Latency Distribution (HdrHistogram - Recorded Latency)
 50.000%   77.63ms
 75.000%   78.97ms
 90.000%   80.51ms
 99.000%   86.14ms
 99.900%   98.05ms
 99.990%   98.05ms
 99.999%   98.05ms
100.000%   98.05ms

#[Mean    =       78.119, StdDeviation   =        2.294]
#[Max     =       97.984, Total count    =          400]
#[Buckets =           27, SubBuckets     =         2048]
----------------------------------------------------------
  600 requests in 30.01s, 682.72KB read
Requests/sec:     20.00
Transfer/sec:     22.75KB

Running 30s test @ http://localhost/cgi-bin/empty.cgi/extra/path?foo+bar+--me%202
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     2.89ms  632.16us   5.86ms   60.74%
    Req/Sec    71.72     48.95   111.00     68.41%
  Latency Distribution (HdrHistogram - Recorded Latency)
 50.000%    2.87ms
 75.000%    3.40ms
 90.000%    3.70ms
 99.000%    4.22ms
 99.900%    5.22ms
 99.990%    5.81ms
 99.999%    5.86ms
100.000%    5.86ms
#[Mean    =        2.886, StdDeviation   =        0.632]
#[Max     =        5.856, Total count    =         6795]
#[Buckets =           27, SubBuckets     =         2048]
----------------------------------------------------------
  10205 requests in 30.00s, 1.18MB read
Requests/sec:    340.13
Transfer/sec:     40.20KB

Test without server

Running the empty.cgi script without an HTTP server takes about 5 milliseconds. Therefore, we will not be able to do much better than half that time (since the test command starts 2 processes). Apache takes approximately 3 milliseconds on an empty CGI script, so there is very little overhead.

cargo build --release
time target/release/tortuga test -s examples/debug.cgi

CGI versus WCGI

On a 4-core Intel CPU, I ran the following quick benchmarks:

WCGI without Compilation Cache

The initial compilation slows down the first request.

$ wrk -c 1 -t 1 -d 1s 'http://localhost:3000/cgi-bin/echo.wcgi/extra/path?--foo+bar'
Running 1s test @ http://localhost:3000/cgi-bin/echo.wcgi/extra/path?--foo+bar
  1 threads and 1 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.19ms    4.79ms  44.02ms   96.79%
    Req/Sec     2.55k   462.45     3.01k    90.91%
  2791 requests in 1.10s, 272.56KB read
Requests/sec:   2535.83
Transfer/sec:    247.64KB

WCGI with Compilation Cache

The cache is enabled by pointing the server at a file created via wasmtime config new.

$ wrk -c 1 -t 1 -d 1s 'http://localhost:3000/cgi-bin/echo.wcgi/extra/path?--foo+bar'      
Running 1s test @ http://localhost:3000/cgi-bin/echo.wcgi/extra/path?--foo+bar
  1 threads and 1 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   408.90us  266.19us   3.76ms   92.34%
    Req/Sec     2.59k   260.40     2.84k    81.82%
  2828 requests in 1.10s, 276.17KB read
Requests/sec:   2570.45
Transfer/sec:    251.02KB

CGI

$ wrk -c 1 -t 1 -d 1s 'http://localhost:3000/cgi-bin/echo.cgi/extra/path?--foo+bar'
Running 1s test @ http://localhost:3000/cgi-bin/echo.cgi/extra/path?--foo+bar
  1 threads and 1 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     6.68ms    1.08ms   9.74ms   69.33%
    Req/Sec   150.10     13.25   170.00     80.00%
  150 requests in 1.00s, 14.65KB read
Requests/sec:    149.35
Transfer/sec:     14.58KB

Static File

$ wrk -c 1 -t 1 -d 1s 'http://localhost:3000/index.html'
Running 1s test @ http://localhost:3000/index.html
  1 threads and 1 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   221.81us  118.00us   1.45ms   90.75%
    Req/Sec     4.62k   497.28     5.10k    81.82%
  5058 requests in 1.10s, 503.82KB read
Requests/sec:   4598.99
Transfer/sec:    458.10KB