Memory leak in Core when overloading with requests

[OnixGH]

When running Passenger with valgrind and spamming the server with small POST requests locally, it ends up with leaked memory. E.g. on Ubuntu 14.04 with 5.0.27 (also occurs with 5.0.26):

==20840== 8,994,816 bytes in 549 blocks are definitely lost in loss record 13 of 13
==20840==    at 0x4C2D110: memalign (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==20840==    by 0x4C2D227: posix_memalign (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==20840==    by 0x47C1F4: call_memalign(unsigned long, unsigned long) (palloc.cpp:47)
==20840==    by 0x47C2C0: psg_create_pool(unsigned long) (palloc.cpp:62)
==20840==    by 0x5F3C70: Passenger::ServerKit::HttpServer<Passenger::Core::Controller, Passenger::Core::Client>::reinitializeRequest(Passenger::Core::Client*, Passenger::Core::Request*) (HttpServer.h:870)
==20840==    by 0x5E6070: Passenger::Core::Controller::reinitializeRequest(Passenger::Core::Client*, Passenger::Core::Request*) (Hooks.cpp:121)
==20840==    by 0x5FA5E2: Passenger::ServerKit::HttpServer<Passenger::Core::Controller, Passenger::Core::Client>::handleNextRequest(Passenger::Core::Client*) (HttpServer.h:253)
==20840==    by 0x5F3A75: Passenger::ServerKit::HttpServer<Passenger::Core::Controller, Passenger::Core::Client>::onClientAccepted(Passenger::Core::Client*) (HttpServer.h:707)
==20840==    by 0x5E5DAC: Passenger::Core::Controller::onClientAccepted(Passenger::Core::Client*) (Hooks.cpp:92)
==20840==    by 0x607153: Passenger::ServerKit::BaseServer<Passenger::Core::Controller, Passenger::Core::Client>::onClientsAccepted(Passenger::Core::Client**, unsigned int) (Server.h:540)
==20840==    by 0x55810C: Passenger::ServerKit::BaseServer<Passenger::Core::Controller, Passenger::Core::Client>::feedNewClients(int const*, unsigned int) (Server.h:797)
==20840==    by 0x55138E: Passenger::ServerKit::AcceptLoadBalancer<Passenger::Core::Controller>::feedNewClient(Passenger::Core::Controller*, int) (AcceptLoadBalancer.h:174)

• worker_connections is set to 10000 to avoid the "not enough workers" error
• passenger_socket_backlog is also set higher than that, although I still see the "connect() to unix:/tmp/passenger.M6yCfQv/agents.s/core failed (11: Resource temporarily unavailable) while connecting to upstream: error", which is something that I didn't expect either

The POST request is something simple, and every upload gets dumped to the same file called "last_upload":

POST /upload/create HTTP/1.1
Host: 127.0.0.1:3000
Connection: close
Content-length: 10

1234567890

The load script is also simple:

#!/bin/bash
for i in `seq 1 1000`;
do
for j in `seq 1 7`;
do
cat post.txt | nc 127.0.0.1 3000 -q 10 >/dev/null &
done
perl -e 'select(undef,undef,undef,.001)'
done