From cronfy at gmail.com  Wed Sep  8 00:39:12 2010
From: cronfy at gmail.com (cronfy)
Date: Wed, 8 Sep 2010 02:39:12 +0400
Subject: [mpm-itk] Why process connection with fork() ?
Message-ID: <AANLkTik0j2svzaLhZ=p7xoQ+uz9Q8UjUjjr=ijFs-G40@mail.gmail.com>

Hello,

I am interested, why connection processing is implemented via fork()
in child_main(). Looks like the only reason is to write a message to
log if something goes wrong with a child:

        /*
         * We now have a connection, so set it up with the appropriate
         * socket options, file descriptors, and read/write buffers.
         */

        {
            pid_t pid = fork(), child_pid;
            int status;
            switch (pid) {
            case -1:
                ap_log_error(APLOG_MARK, APLOG_ERR, errno, NULL,
"fork: Unable to fork new process");
                break;
            case 0: /* child */
                apr_proc_mutex_child_init(&accept_mutex, ap_lock_fname, pchild);
                current_conn = ap_run_create_connection(ptrans,
ap_server_conf, csd, my_child_num, sbh, bucket_alloc);
                if (current_conn) {
                    ap_process_connection(current_conn, csd);
                    ap_lingering_close(current_conn);
                }
                exit(0);
            default: /* parent; just wait for child to be done */
                do {
                    child_pid = waitpid(pid, &status, 0);
                } while (child_pid == -1 && errno == EINTR);

                if (child_pid != pid || !WIFEXITED(status)) {
                    if (WIFSIGNALED(status)) {
                        ap_log_error(APLOG_MARK, APLOG_ERR, 0,
ap_server_conf, "child died with signal %u", WTERMSIG(status));
                    } else if (WEXITSTATUS(status) != 0) {
                        ap_log_error(APLOG_MARK, APLOG_ERR, 0,
ap_server_conf, "child exited with non-zero exit status %u",
WEXITSTATUS(status));
                    } else {
                        ap_log_error(APLOG_MARK, APLOG_ERR, errno,
NULL, "waitpid() failed");
                    }
                    clean_child_exit(1);
                }
                break;
            }
        }


Isn't it better to just process a connection (execute case 0: block)
and exit immediately, without fork()ing?

With this implementation we always need 2*N + 1 processes to serve N
simultaneous requests, which does not look good for performance.

-- 
// cronfy



From sgunderson at bigfoot.com  Wed Sep  8 00:43:19 2010
From: sgunderson at bigfoot.com (Steinar H. Gunderson)
Date: Wed, 8 Sep 2010 00:43:19 +0200
Subject: [mpm-itk] Why process connection with fork() ?
In-Reply-To: <AANLkTik0j2svzaLhZ=p7xoQ+uz9Q8UjUjjr=ijFs-G40@mail.gmail.com>
References: <AANLkTik0j2svzaLhZ=p7xoQ+uz9Q8UjUjjr=ijFs-G40@mail.gmail.com>
Message-ID: <20100907224319.GB12486@samfundet.no>

On Wed, Sep 08, 2010 at 02:39:12AM +0400, cronfy wrote:
> I am interested, why connection processing is implemented via fork()
> in child_main(). Looks like the only reason is to write a message to
> log if something goes wrong with a child:

The reason is that a fork is much cheaper than a full restart of the child.
Also, the respawn logic (which we just copy) is not tuned at all for children
dying constantly.

> With this implementation we always need 2*N + 1 processes to serve N
> simultaneous requests, which does not look good for performance.

Sleeping processes do not cost much, unless you have an extremely bad
scheduler.

/* Steinar */
-- 
Homepage: http://www.sesse.net/




From roman at rs-labs.com  Mon Sep 20 19:11:05 2010
From: roman at rs-labs.com (Roman Medina-Heigl Hernandez)
Date: Mon, 20 Sep 2010 19:11:05 +0200
Subject: [mpm-itk] Mod_evasive
Message-ID: <4C9795A9.5030006@rs-labs.com>

Hello,

Did anybody get mod_evasive to work with mpm-itk? I didn't and I suspect
may be due to incompatibility with itk. Any idea?

TIA.

-- 

Saludos,
-Roman

PGP Fingerprint:
09BB EFCD 21ED 4E79 25FB  29E1 E47F 8A7D EAD5 6742
[Key ID: 0xEAD56742. Available at KeyServ]