Demo of how using a Berkeley DB concurrently without locking can break under load

Install

• nginx
• memcached
• siege
• libmemcached

Setup

• Start nginx with the supplied config (tweaking port, location of fastcgi_params, etc. as necessary).

• Start memcached on localhost, port 11211 like this:

  memcached -d

• Start the fcgi backends like this (adjust -n 4 to the number of CPU cores available / 2):

  ./script/demo_corruptdb_fastcgi.pl -l 127.0.0.1:8999 -n 4 -e

Running

Run siege against http:://localhost:8080/ like this (adjust -c 4 to the number of CPU cores available):

siege -c 4 http:://localhost:8080/

or

siege -c 4 http:://localhost:8080/memcached/

Select the implementation of Berkeley DB to test in lib/Demo/CorruptDB/Controller/Root.pm by adding/removing comments:

BEGIN {
    # @AnyDBM_File::ISA = qw(DB_File) # Breaks
    # @AnyDBM_File::ISA = qw(GDBM_File) # Breaks
    @AnyDBM_File::ISA = qw(NDBM_File) # Breaks, but only sometimes?
    # @AnyDBM_File::ISA = qw(ODBM_File) # Breaks
} 

And watch how it breaks under load with siege. It might be necessary to remove corrupt DB's between runs (with 'rm /tmp/corruptdb*').

The NDBM implementation breaks more easily when siege is run without the -c option.

The memcached version was added for comparison. It doesn't break :)