Feature/cache

[subnetmarco]

This pull request is the first step in implementing #15, with the final step being having an invalidation system in place.

It implements a time-expiration caching system configurable through kong.yaml configuration file.

# Cache configuration
cache:
  expiration: 5 # In seconds

I'm creating the pull request without merging, and will post the results of the benchmark below.

[subnetmarco]

Benchmarks

The benchmarks have been executed on the EC2 environment, using:

siege -c 300 -t 60S ...

Without cache (version 0.0.1-beta)

Transactions:          70626 hits
Availability:          98.53 %
Elapsed time:          21.80 secs
Data transferred:         0.88 MB
Response time:            0.09 secs
Transaction rate:      3239.72 trans/sec
Throughput:           0.04 MB/sec
Concurrency:          297.06
Successful transactions:       70626
Failed transactions:          1052
Longest transaction:          1.37
Shortest transaction:         0.01

With cache (default=5s)

Transactions:         535318 hits
Availability:          99.77 %
Elapsed time:          43.45 secs
Data transferred:         6.64 MB
Response time:            0.02 secs
Transaction rate:     12320.32 trans/sec
Throughput:           0.15 MB/sec
Concurrency:          295.57
Successful transactions:      535318
Failed transactions:          1209
Longest transaction:          5.02
Shortest transaction:         0.00

[thibaultcha]

nice

[samek]

Is this already in the 0.4.1 ?

[thibaultcha]

This is in Kong since the first release.

[subnetmarco]

@samek bear in mind that is the internal cache of Kong for its internal entities, not an HTTP caching layer for the upstream APIs (#411, #412)

[samek]

@thefosk exactly what I thought it was for (http upstream). :(

Any plans for that ?

[mashapedeployment]

@samek https://github.com/Mashape/kong/issues?q=caching+is%3Aissue

[samek]

ok in the mean time I've added it to the nginx config in kong.yml - for docker.
If someone want's to try (it'll be caching all apis's).

Add parts between the ##CUSTOM CACHING##

 --- a/config.docker/kong.yml
 +++ b/config.docker/kong.yml
 @@ -35,7 +35,8 @@ database: cassandra
  databases_available:
    cassandra:
      properties:
 -      hosts: "cassandra"
 +      hosts: 
 +CASSANDRAHOSTS
        port: 9042
        timeout: 1000
        keyspace: kong
 @@ -124,8 +125,13 @@ nginx: |
      ';

      init_worker_by_lua 'kong.exec_plugins_init_worker()';
 +    ##CUSTOM CACHING##
 +    proxy_cache_path  /tmp/ levels=1 keys_zone=apicache:10m;
 +    ##CUSTOM CACHING##

      server {
 +      set_real_ip_from 0.0.0.0/0;
 +      real_ip_header Cres-Client-IP;
        server_name _;
        listen {{proxy_port}};
        listen {{proxy_ssl_port}} ssl;
 @@ -144,6 +150,15 @@ nginx: |
          # Authenticate the user and load the API info
          access_by_lua 'kong.exec_plugins_access()';

 +        ##CUSTOM CACHING##
 +        proxy_cache apicache;
 +        proxy_cache_use_stale updating;
 +        proxy_cache_lock on;
 +        proxy_cache_valid 200 10s;
 +        proxy_ignore_headers X-Accel-Expires Expires Cache-Control;
 +        add_header X-Proxy-Cache $upstream_cache_status;
 +        ##CUSTOM CACHING##
 +
          # Proxy the request
          proxy_set_header X-Real-IP $remote_addr;
          proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
 @@ -203,4 +218,4 @@ nginx: |
        # Do not remove, additional configuration placeholder for some plugins
        # {{additional_configuration}}
      }

[samek]

Btw. Looks like analitics only counts misses.