执行systemctl reload httpd时部分进程启动异常的技术问询

Let’s dig into this odd alternating behavior you’re seeing—where every odd-numbered systemctl reload httpd only spins up 11 processes, but even runs get you the full 21. This is almost always tied to how Apache handles graceful reloads or misconfigured process limits. Here’s how to fix it:

1. Audit Apache’s MPM Configuration #

First stop: your Multi-Processing Module (MPM) settings. If you’re using mpm_prefork or mpm_worker, mismatched limits can cause this flip-flop:

• Open your MPM config file (typically /etc/httpd/conf.modules.d/00-mpm.conf)
• Look for directives like ServerLimit, MaxRequestWorkers, StartServers, and MinSpareServers
• A common culprit is ServerLimit being set lower than MaxRequestWorkers. When you reload, Apache can’t spawn more processes than ServerLimit allows on the first try, but a second reload might bypass this stale limit. Ensure ServerLimit matches or exceeds MaxRequestWorkers for your MPM.

2. Test Restart vs. Reload Behavior #

A reload tells Apache to gracefully replace child processes, but sometimes old processes hang around instead of exiting properly on odd runs. Let’s isolate this:

• First, run systemctl restart httpd (a full restart, not reload)
• Check your process count with:

  ps aux | grep httpd | grep -v grep | grep -v log | wc -l
  

• If the count stays consistent after restart, the issue is with how reload handles process termination. Add or adjust GracefulShutdownTimeout in your Apache config (e.g., GracefulShutdownTimeout 30) to give old processes more time to exit before spawning new ones.

3. Clean Up Stale Lock/PID Files #

Occasionally, Apache leaves behind stale PID files or locks that mess with reloads:

• Stop Apache first: systemctl stop httpd
• Check /var/run/httpd/ for any unexpected files (like duplicate PID files or lock files)
• Delete stale files, then start Apache back up with systemctl start httpd
• Test the reload sequence again to see if the process count stabilizes.

4. Verify the Systemd Service File #

Systemd’s ExecReload directive might be misconfigured, causing inconsistent reload behavior:

• Open the httpd service file (usually /usr/lib/systemd/system/httpd.service)
• Look for the ExecReload line—it should look like:

  ExecReload=/usr/sbin/httpd $OPTIONS -k graceful
  

• If it’s missing the -k graceful flag or has incorrect options, that could be why reloads aren’t working reliably. Correct the line, run systemctl daemon-reload to apply changes, then test again.

Original Issue Context #

When executing systemctl reload httpd, some processes fail to start on odd-numbered runs. The process count alternates between 11 and 21, as shown below:

$ systemctl reload httpd
$ ps aux | grep httpd | grep -v grep | grep -v log | wc -l
11
$ systemctl reload httpd
$ ps aux | grep httpd | grep -v grep | grep -v log | wc -l
21
$ systemctl reload httpd
$ ps aux | grep httpd | grep -v grep | grep -v log | wc -l
11

内容的提问来源于stack exchange，提问作者Naoki