Eu tenho meu servidor apache configurado com o trabalhador MPM da seguinte maneira:
<IfModule mpm_worker_module>
ServerLimit 20
ThreadLimit 50
StartServers 3
MaxClients 1000
MinSpareThreads 50
MaxSpareThreads 200
ThreadsPerChild 50
MaxRequestsPerChild 10000
</IfModule>
O servidor recebe cerca de 2 milhões de solicitações todos os dias. Quando as solicitações forem recebidas, o servidor apache deixará de aceitar nova conexão (o navegador não pode se conectar). E o processo permanecerá na lista de processos. O uso da memória é de cerca de 10% e a CPU é de 0%. Parece que o servidor apache não conseguiu reciclar os processos quando deveria.
Encontrei algo relacionado à falha nos logs de erro:
[Mon Aug 15 12:36:24 2011] [error] child process 11444 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:24 2011] [error] child process 12090 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:24 2011] [error] child process 10698 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:24 2011] [error] child process 12144 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:24 2011] [error] child process 12032 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:24 2011] [error] child process 10699 still did not exit, sending a SIGKILL
[Mon Aug 15 12:36:25 2011] [notice] caught SIGTERM, shutting down
[Mon Aug 15 12:37:31 2011] [notice] mod_python: Creating 8 session mutexes based on 20 max processes and 50 max threads.
[Mon Aug 15 12:37:31 2011] [notice] mod_python: using mutex_directory /tmp
[Mon Aug 15 12:37:31 2011] [notice] Apache/2.2.17 (Ubuntu) mod_python/3.3.1 Python/2.7.1+ mod_ssl/2.2.17 OpenSSL/0.9.8o configured -- resuming normal operations
[Mon Aug 15 15:36:46 2011] [error] server is within MinSpareThreads of MaxClients, consider raising the MaxClients setting
[Mon Aug 15 15:38:48 2011] [error] server reached MaxClients setting, consider raising the MaxClients setting
[Mon Aug 15 16:17:31 2011] [warn] child process 766 still did not exit, sending a SIGTERM
[Mon Aug 15 16:17:31 2011] [warn] child process 767 still did not exit, sending a SIGTERM
Alguma ideia? Obrigado!