pleskpanel
Regular Pleskian
TITLE:
ACTUAL RESULT:
Plesk Onyx 17.8 PHP 7.3 Handler Logs All Levels to Syslog
PRODUCT, VERSION, OPERATING SYSTEM, ARCHITECTURE:17.8.11 Update #71, CentOS Linux 7.7.1908
PROBLEM DESCRIPTION:Following addition of the Plesk PHP 7.3.10 handler version via the Plesk components tool, all PHP errors that use it are being logged into the /var/log/plesk-php73-fpm/error.log file regardless of log_level.
This is not a domain-level concern (Apache error log for a particular domain took a lot of disk space on a server deals with a site writing lots of errors to the site's error logs).
Similar issue noted here: Issue - Since PHP 7.3 many function errors (FPM)
STEPS TO REPRODUCE:This is not a domain-level concern (Apache error log for a particular domain took a lot of disk space on a server deals with a site writing lots of errors to the site's error logs).
Similar issue noted here: Issue - Since PHP 7.3 many function errors (FPM)
- Add PHP 7.3 handler support via Plesk GUI
- Change a website plan to use PHP 7.3 handler
- Uncomment and set log_level = alert in /opt/plesk/php/7.3/etc/php-fpm.conf
- Restart services or server for good measure
PHP warnings (not handler alert-level warnings) are being written to /var/log/plesk-php73-fpm/error.log, causing very large syslog files.
(they continue to be written to the proper site-level error log as well)
EXPECTED RESULT:(they continue to be written to the proper site-level error log as well)
Should work as previous Plesk-installed PHP handler versions (PHP 7.2 and below) as documented in Log files occupy a lot of disk space because PHP log_level = notice. Is it possible to change the default value? in which only handler alerts were logged to the syslog file.
ANY ADDITIONAL INFORMATION:Added this to the "Additional Directives" of the PHP 7.3 plan via Plesk to suppress errors:
[php-fpm-pool-settings]
catch_workers_output = no
This seems to have stopped writing all errors entirely to the syslog.
YOUR EXPECTATIONS FROM PLESK SERVICE TEAM:[php-fpm-pool-settings]
catch_workers_output = no
This seems to have stopped writing all errors entirely to the syslog.
Confirm bug