Facebook
Twitter
burnley
Regular Pleskian
User name: burnley
TITLE
Obsidian on CentOS 7: php-fpm socket for a vhost disappears if another php-fpm vhost is added or removed
PRODUCT, VERSION, OPERATING SYSTEM, ARCHITECTURE
Plesk Obsidian Version 18.0.25 Update #2, last updated on April 8, 2020 09:38 PM
CentOS Linux 7.8.2003 (Core)
PROBLEM DESCRIPTION
Yesterday at 12:03pm AEST I had a website, let's call it broken.com.au, running php73-php-fpm starting to return HTTP 503. At that time Plesk has reconfigured this particular php-fpm pool as part of deleting several *other* subscriptions, unrelated to this site. Starting at that time php-fpm started to log into it's error_log file:
[12-Jul-2020 12:03:40] ERROR: epoll: unable to add fd 392 (which was at the time the file descriptor used for the crashed site)
I've fixed the site by reconfiguring it to run another php-fpm handler, namely php74-php-fpm. All good until today 12:02pm when it happened again: Plesk has deleted several other sites using php73-php-fpm and, at the same time, this broken.com.au site crashed again, even though now it's on a different php-fpm handler. Why did it crash? Because /var/www/vhosts/system/broken.com.au/php-fpm.sock went missing again exactly at 12:02pm. To fix it I had to run "systemctl restart php74-php-fpm", reload hasn't helped.
So I'm thinking we have some internal Plesk state inconsistency (in the psa db maybe?) and tried to run some tests on the affected server
Test1: Created a new test subscription using php73-php-fpm handler. As soon as Plesk finished creating the subscription and refreshed the window /var/www/vhosts/system/broken.com.au/php-fpm.sock disappeared. Had to run "systemctl restart php74-php-fpm"
Test2: Deleted the previously created test subscription in Plesk. As soon as Plesk finished deleting it and refreshed the window /var/www/vhosts/system/broken.com.au/php-fpm.sock disappeared. Had to run "systemctl restart php74-php-fpm" again.
So right now pretty much every action that involves reconfiguring php73-php-fpm handler causes the deletion of /var/www/vhosts/system/broken.com.au/php-fpm.sock. Why and where do I start looking for it?
STEPS TO REPRODUCE
Don't know how to reproduce it, but on the affected server we have right now the conditions to replicate the problem. See above.
ACTUAL RESULT
Erroneous /var/www/vhosts/system/broken.com.au/php-fpm.sock deletion. HTTP 503 service unavailable for the affected subscriptions.
EXPECTED RESULT
No /var/www/vhosts/system/broken.com.au/php-fpm.sock deletion if the Plesk actions are affecting other websites
ANY ADDITIONAL INFORMATION
YOUR EXPECTATIONS FROM PLESK SERVICE TEAM
Confirm bug
TITLE
Obsidian on CentOS 7: php-fpm socket for a vhost disappears if another php-fpm vhost is added or removed
PRODUCT, VERSION, OPERATING SYSTEM, ARCHITECTURE
Plesk Obsidian Version 18.0.25 Update #2, last updated on April 8, 2020 09:38 PM
CentOS Linux 7.8.2003 (Core)
PROBLEM DESCRIPTION
Yesterday at 12:03pm AEST I had a website, let's call it broken.com.au, running php73-php-fpm starting to return HTTP 503. At that time Plesk has reconfigured this particular php-fpm pool as part of deleting several *other* subscriptions, unrelated to this site. Starting at that time php-fpm started to log into it's error_log file:
[12-Jul-2020 12:03:40] ERROR: epoll: unable to add fd 392 (which was at the time the file descriptor used for the crashed site)
I've fixed the site by reconfiguring it to run another php-fpm handler, namely php74-php-fpm. All good until today 12:02pm when it happened again: Plesk has deleted several other sites using php73-php-fpm and, at the same time, this broken.com.au site crashed again, even though now it's on a different php-fpm handler. Why did it crash? Because /var/www/vhosts/system/broken.com.au/php-fpm.sock went missing again exactly at 12:02pm. To fix it I had to run "systemctl restart php74-php-fpm", reload hasn't helped.
So I'm thinking we have some internal Plesk state inconsistency (in the psa db maybe?) and tried to run some tests on the affected server
Test1: Created a new test subscription using php73-php-fpm handler. As soon as Plesk finished creating the subscription and refreshed the window /var/www/vhosts/system/broken.com.au/php-fpm.sock disappeared. Had to run "systemctl restart php74-php-fpm"
Test2: Deleted the previously created test subscription in Plesk. As soon as Plesk finished deleting it and refreshed the window /var/www/vhosts/system/broken.com.au/php-fpm.sock disappeared. Had to run "systemctl restart php74-php-fpm" again.
So right now pretty much every action that involves reconfiguring php73-php-fpm handler causes the deletion of /var/www/vhosts/system/broken.com.au/php-fpm.sock. Why and where do I start looking for it?
STEPS TO REPRODUCE
Don't know how to reproduce it, but on the affected server we have right now the conditions to replicate the problem. See above.
ACTUAL RESULT
Erroneous /var/www/vhosts/system/broken.com.au/php-fpm.sock deletion. HTTP 503 service unavailable for the affected subscriptions.
EXPECTED RESULT
No /var/www/vhosts/system/broken.com.au/php-fpm.sock deletion if the Plesk actions are affecting other websites
ANY ADDITIONAL INFORMATION
YOUR EXPECTATIONS FROM PLESK SERVICE TEAM
Confirm bug
