TITLE:
Plesk updates website configuration concurrently instead of sequential
PRODUCT, VERSION, OPERATING SYSTEM, ARCHITECTURE:Plesk Onyx 17.8.11
Debian 9
amd64/x86_64
PROBLEM DESCRIPTION:Debian 9
amd64/x86_64
When we made adjustments to default PHP config, for example /etc/php/7.0/cgi/php.ini we need to inform subscriptions about these changes. In order for us to do this we run:
This updates all php.ini files for all subscriptions with the updated version as is expected. However, this also executes some sort of async process.
On servers with a high domain count (eg. over 500) this is causing the server to run out of memory, each max process count, unable to fork errors, etc, etc. We sometimes see over 3000 of these 'async plesk processes' that spit out errors like unable to fork.
Additionally, this sometimes seems to trigger some kind of race condition in FPM that causes FPM to hang
STEPS TO REPRODUCE:
Code:
/opt/psa/bin/php_settings -u
This updates all php.ini files for all subscriptions with the updated version as is expected. However, this also executes some sort of async process.
On servers with a high domain count (eg. over 500) this is causing the server to run out of memory, each max process count, unable to fork errors, etc, etc. We sometimes see over 3000 of these 'async plesk processes' that spit out errors like unable to fork.
Additionally, this sometimes seems to trigger some kind of race condition in FPM that causes FPM to hang
1. Create a plesk machine with a high domain count (over 500) with PHP enabled.
2. Run /opt/psa/bin/php_settings -u
3. In another session, see your process counter go skyhigh. Server load goes through the roof, etc, etc.
ACTUAL RESULT:2. Run /opt/psa/bin/php_settings -u
3. In another session, see your process counter go skyhigh. Server load goes through the roof, etc, etc.
1. the async process sometimes fails to fork, possibly resulting in sites not properly adjusted.
2. Server reaches max process count
3. Server gets very slow and very busy
4. Sometimes triggers a race conditiion in FPM, where fpm requires a manual restart
EXPECTED RESULT:2. Server reaches max process count
3. Server gets very slow and very busy
4. Sometimes triggers a race conditiion in FPM, where fpm requires a manual restart
The 'Plesk async process' should be run sequentially (one domain after another) and the processes should be queued.
Now it seems they are all fired after another as a seperate process/thread. This isn't an issue on a server with maybe 100 domains, but it is on servers with lots of domains.
ANY ADDITIONAL INFORMATION:Now it seems they are all fired after another as a seperate process/thread. This isn't an issue on a server with maybe 100 domains, but it is on servers with lots of domains.
YOUR EXPECTATIONS FROM PLESK SERVICE TEAM:Confirm bug
Last edited: