Apache workers
Jump to navigation
Jump to search
Apache worker model ⌘
- Apache can function using different models
- Most common - prefork
Apache prefork ⌘
- One central Apache process to accept incoming connections, read configuration etc
- Multiple child processes to serve requests
- Each child can serve only one request at a time
Apache prefork settings ⌘
``` StartServers 8 MinSpareServers 5 MaxSpareServers 20 ServerLimit 256 MaxClients 256 MaxRequestsPerChild 4000 ```
Tuning Apache prefork ⌘
- StartServers
- Number of Apache processes to create on first start
- MinSpareServers
- Minimum number of spare Apache processes that should be available at any time. If less spare workers than this are available, Apache will start more
- MaxSpareServers
- Maximum number of spare workers.If more than this are available, Apache will kill them off at a rate of 1/second to bring it below this.
Tuning Apache prefork ⌘
- Default EL5 config has a maximum of 256 child processes
- This means maximum 256 concurrent requests
- In reality, 256 may well be too high
- When using PHP, Apache processes are often 30MB+
- 30 * 256 = 7680MB of RAM required, just for workers
- Under these conditions, 256 MaxClients suitable only for machines with at 8GB RAM
Sizing MaxClients ⌘
- Tool ps_mem.py can be used
- http://www.pixelbeat.org/scripts/ps_mem.py
- Shows actual RAM in use by each process
- Should be used on a server with a representative workload
- Set MaxClients too high, and you'll start swapping
What are the workers doing? ⌘
- Apache module `mod_status` shows this data
- Module is loaded by default but not enabled
- Uncomment the section in /etc/httpd/conf/httpd.conf to enable
- Change `.example.com` to `127.0.0.1`
- Once enabled, run `service httpd fullstatus` to see data
Exercise ⌘
- Enable mod_status and find the status of each worker
- How much memory is each httpd process using on your VM?
- What is a suitable value for MaxClients on this VM based on the above?
