Resolved Statistic collection saturates disk

[Pascal_Netenvie]

Server operating system version

Debian 11.6

Plesk version and microupdate number

18.0.51u1

Hello,
I use Plesk + 360 monitoring on 15 to 20 servers.
Most of servers runs on Infomaniak Public cloud with Block storage perf 2 (1000 IOPS and 400 Mo/s).
On those servers no problems (apparently).

But recently i builded a new server with a perf 1 Block storage (limited to 200 Mo/s and 500 IOPS).
On this new server i have disk busy at 100% all the time and after some test i found that the process generating that seems to be server monitoring.
After i disabled Monitoring extension and killed Statictics_collect task the disk usage came back to normal.

So i wonder if Monitoring is well optimized and don't use too much disk IOPS ...
Because if it use too much disk read, for sure it slow a bit disk access for other processes even on servers with faster disk ...

I would like to have a real feedback from other users and Plesk 360 team.

 

[Peter Debik]

To get the desired feedback I suggest to open a support ticket on this question. Only engineers who know the software from the inside will be able answer what exactly it is doing.

Personally, from my experience I think it is quite likely that statistics require a lot of disk operations, because they will need to parse log files.

 

[Pascal_Netenvie]

Thanks for you answer.
I have rebooted then reinstalled Monitoring agent and the problem have not appeared since that.
I think the "Statictics_collect..." process is more likley connected to log rotation or analytics as you said.
But normally it would not run all day ...
I will follow that and come back here with more informations.

 

[Pascal_Netenvie]

Ok the problem come from this process :
/opt/psa/admin/bin/statistics_collector

The process start during night (around 4h20) then never start. And it keep disk busy 40 to 95% all the time.
I don't see that on any other server.

An idea or solution to solve that ?

 

[Pascal_Netenvie]

i wanted to say : "then never stop" ... not "never start"

 

[Peter Debik]

I suggest to open a support ticket on this question. Only engineers who know the software from the inside will be able answer what exactly it is doing.

 

[Pascal_Netenvie]

Do you have lastest URL to open ticket ?
It seems it changed recently ...
I was on Plesk Support site then tried to create an account but it bring me back to Plesk 360 where i already have an account but there is no link to support.

 

[Peter Debik]

It's still https://support.plesk.com, but if you cannot login there, please follow these steps:

https://support.plesk.com/hc/en-us/articles/13545238026007

 

[Pascal_Netenvie]

Ok done and now it work. Thx.

 

[Pascal_Netenvie]

This is support answer, if it can help other users :

The issue is caused by a combination of two factors - the disk subsystem of the server is quite limited in resources, and because both websites have a lot of files with a huge number of nested directories. Even basic Linux commands such as du and find perform exceptionally slow. As for statistics_collector - it tries to calculate the size of directories in accordance with its business logic, so it is also affected by the issue.

Due to the fact that the issue is system related - it is not possible to fix that from Plesk side. But there are few workarounds that can be applied:

• Disable web statistics for domains as described in How to enable/disable web statistics for a domain in Plesk article.
• Do adjustments in Tools & Settings > Server Statistics Settings - disabling some of the features should reduce the load on disk during calculation. Please check this documentation page for more information.
• Use Linux ionice feature to limit I/O throughput for statistics_collector. To do this:

1. Move /usr/local/psa/admin/sbin/statistics_collector to another location, for example /usr/local/psa/admin/sbin/real_statistics_collector.
2. Create a bash script named/usr/local/psa/admin/sbin/statistics_collector that will call for real_statistics_collector via ionice:

#!/bin/sh
ionice -c [scheduling class] -n [level] /usr/local/psa/admin/sbin/real_statistics_collector $@

Please note that it is only an example, and it is required to set scheduling class and level with values from ionice manual. You may check them here: ionice(1) - Linux man page. And it is also should be noted that this substitution is going to be removed during a Plesk update installation, so it will be required to reconfigure that after the updates.

 

[Petrus]

They have not found any solution and no one intends to do so. Other topics were answered with: Get a faster SSD. I have a server on which statistic_coll takes about 2-3 days. And all access_logs from the sites are only every 2 days. Now the last one is from December 26 and the current one. Probably next year I will get another dedicated server and obviously without Plesk, which made me desperate. It slows down the server and the memory and everything. I don't sell web space, I don't need statistics, I have the server only for my sites. I know how much space is occupied by each one, I know what traffic is being done, I don't need any statistics. And there is nowhere to disable it. I will probably become a former customer of yours and give you negative reviews everywhere if you don't correct this. I'm not saying that once in a while I wake up with extensions installed without being asked if I agree with their installation or not. And there is nowhere to block the installation of new extensions. I've had plesk for many years, but now I got really angry and said I should make an account to express my disappointment. Either you will do something with the tasks that make statistics or I will have to give up. Plesk is ok but the statistics literally destroy my server.

 

[Peter Debik]

@Petrus I have removed your other two identical venting posts, and for this here I ask you to open your own, new thread instead of hijacking others. If you experience issues, it could have been an option to ask for assistance by other Plesk users here on the forum before. You can also contact official Plesk support. They will be happy to assist. Instead of blaming unreasonable guesses on Plesk maybe it is better to find a solution. I also much dislike the blackmail type thread to write negative reviews. Again, it may be better if you seek analysis and solution for an issue instead of publishing threats.

 

[Petrus]

@Petrus I have removed your other two identical venting posts, and for this here I ask you to open your own, new thread instead of hijacking others. If you experience issues, it could have been an option to ask for assistance by other Plesk users here on the forum before. You can also contact official Plesk support. They will be happy to assist. Instead of blaming unreasonable guesses on Plesk maybe it is better to find a solution. I also much dislike the blackmail type thread to write negative reviews. Again, it may be better if you seek analysis and solution for an issue instead of publishing threats.

I didn't hijack anything. It's exactly the same problem. And the fact that you saw that I have two similar posts on similar topics, what does this show? That I studied the problem and looked for solutions, but they don't exist. And not only here but also on other forums and Plesk's answer was variations on: It cannot be given up as it is part of the basic functions. If you want, I'll give you the links. Are you saying I can contact Plesk support? It means that you don't know that if you have a reseller license you have to pay for assistance. I have a dedicated server from IBM and the support said that since it was bought through a reseller, I have to pay for the support. https://support.plesk.com/hc/en-us/articles/12388090147095-How-to-get-support-directly-from-Plesk- And IBM can't help me because it's not their product. No one is happy to help me. I also contacted Elena from the assistant, she gave me the link above where she told me to sign up for technical support. And I blame Plesk for real things, not unreasonable assumptions. Yes, the solution is to give up plesk. It is also proven by your message, which instead of helping me comes with hater's speech. It wasn't blackmail, but I was saying that I was very disappointed and a disappointed customer obviously can't say good things about what/who disappointed him. And you would rather give me a solution than have a hate speech. Obviously I did not expect a solution because, as I said in the first message, they do not exist because I searched for many days on various websites. Actually there is, I will give up plesk. And the aggressive manner of your answer confirms to me that this is the correct solution and that you do not know what answer to give to the subject. I not want to putting in an SSD or changing the HDD or giving up the millions of files on the server now. If you have a technical solution, I would be very happy to know it, if you are going to do the theory again, I am not interested.

 

[Pascal_Netenvie]

I didn't hijack anything. It's exactly the same problem. And the fact that you saw that I have two similar posts on similar topics, what does this show? That I studied the problem and looked for solutions, but they don't exist. And not only here but also on other forums and Plesk's answer was variations on: It cannot be given up as it is part of the basic functions. If you want, I'll give you the links. Are you saying I can contact Plesk support? It means that you don't know that if you have a reseller license you have to pay for assistance. I have a dedicated server from IBM and the support said that since it was bought through a reseller, I have to pay for the support. https://support.plesk.com/hc/en-us/articles/12388090147095-How-to-get-support-directly-from-Plesk- And IBM can't help me because it's not their product. No one is happy to help me. I also contacted Elena from the assistant, she gave me the link above where she told me to sign up for technical support. And I blame Plesk for real things, not unreasonable assumptions. Yes, the solution is to give up plesk. It is also proven by your message, which instead of helping me comes with hater's speech. It wasn't blackmail, but I was saying that I was very disappointed and a disappointed customer obviously can't say good things about what/who disappointed him. And you would rather give me a solution than have a hate speech. Obviously I did not expect a solution because, as I said in the first message, they do not exist because I searched for many days on various websites. Actually there is, I will give up plesk. And the aggressive manner of your answer confirms to me that this is the correct solution and that you do not know what answer to give to the subject. I not want to putting in an SSD or changing the HDD or giving up the millions of files on the server now. If you have a technical solution, I would be very happy to know it, if you are going to do the theory again, I am not interested.

Did you tried solution i posted above ? Especially disable web statistics for a domain in Plesk ?

 

[Petrus]

Did you tried solution i posted above ? Especially disable web statistics for a domain in Plesk ?

Yes. I disabled them the first time and no difference. Then I uninstalled them and for nothing. Both on AWstats and Webalizer. I have my own analysis software based on apache logs. Believe me, I searched for solutions for days on end and I understood that they cannot be disabled because it is an essential part of plesk. To uninstall I followed the steps from: https://support.plesk.com/hc/en-us/articles/12377141744919-How-to-disable-web-statistics-in-Plesk-
I appreciate that you wanted to help me, but the problem has been around for at least 2 years, but now it has become more pronounced. I have uploaded several million images on the server and pdf files during these years, and now statistic_coll no longer ends in 24 hours but lasts over 2 days. I'm not saying that vhostmng-find also takes hours on end for sites with many files and again no official solution. Question - vhostmng-find processe consume CPU Someone somehow solved the problem in another similar discussion saying that he put SSD instead of HDD and the time was reduced a lot. But this is not really a solution. It's like saying that Windows Vista is slow and the solution is a supercomputer. All the best.

 

[Pascal_Netenvie]

Hi this is the 2 entire answers from Plesk support that helped me in that case (on a server with limited Disk ressources) :

FIRST :

The issue is caused by a combination of two factors - the disk subsystem of the server is quite limited in resources, and because both websites have a lot of files with a huge number of nested directories. Even basic Linux commands such as du and find perform exceptionally slow. As for statistics_collector - it tries to calculate the size of directories in accordance with its business logic, so it is also affected by the issue.
 
Due to the fact that the issue is system related - it is not possible to fix that from Plesk side. But there are few workarounds that can be applied:

    Disable web statistics for domains as described in How to enable/disable web statistics for a domain in Plesk article.
    Do adjustments in Tools & Settings > Server Statistics Settings - disabling some of the features should reduce the load on disk during calculation. Please check this documentation page for more information.
    Use Linux ionice feature to limit I/O throughput for statistics_collector. To do this:

1. Move /usr/local/psa/admin/sbin/statistics_collector to another location, for example /usr/local/psa/admin/sbin/real_statistics_collector.
2. Create a bash script named/usr/local/psa/admin/sbin/statistics_collector that will call for real_statistics_collector via ionice:

#!/bin/sh
ionice -c [scheduling class] -n [level] /usr/local/psa/admin/sbin/real_statistics_collector $@

Please note that it is only an example, and it is required to set scheduling class and level with values from ionice manual. You may check them here: ionice(1) - Linux man page. And it is also should be noted that this substitution is going to be removed during a Plesk update installation, so it will be required to reconfigure that after the updates.

SECOND :

The process vhostmng-find is a Plesk internal utility that performs various operations related to scanning virtual hosts directories. For example, it is used to index files in vhosts directories (indexing is mainly done to have faster responses from File Manager) and detecting document root subdirectories for applications. It is also used by WordPress Toolkit and by some other tasks (including statistics calculation).
The reasons why vhostmng-find stresses the IO subsystem of the server to its limit are the same as before - websites directory and files are highly nested and there is a 500 IOPS limit. To reduce IO usage by vhostmng-find you may try applying the similar workaround with usage of ionice, but it can result in slow work of Plesk in general and various errors in interface and logs:
1. Move /usr/local/psa/admin/sbin/vhostmng-find to /usr/local/psa/admin/sbin/vhostmng-find-real (for example).
2. Create a bash script named/usr/local/psa/admin/sbin/shostmng-find that will call for vhostmng-find-rea with ionice:

#!/bin/sh
ionice -c [scheduling class] -n [level] /usr/local/psa/admin/sbin/vhostmng-find-real $@

 
So, in case you face issues with Plesk after applying this workaround, I can only suggest reverting it back and then contact with website developers asking if they can reorganize website structure. Or you may also contact with hosting provider technical support and ask them if they can switch storage subsystem to other performance tier (with higher IOPS limit) for additional price.

Hope that will help !

 

[Petrus]

Hi this is the 2 entire answers from Plesk support that helped me in that case (on a server with limited Disk ressources) :

FIRST :

The issue is caused by a combination of two factors - the disk subsystem of the server is quite limited in resources, and because both websites have a lot of files with a huge number of nested directories. Even basic Linux commands such as du and find perform exceptionally slow. As for statistics_collector - it tries to calculate the size of directories in accordance with its business logic, so it is also affected by the issue.
 
Due to the fact that the issue is system related - it is not possible to fix that from Plesk side. But there are few workarounds that can be applied:

    Disable web statistics for domains as described in How to enable/disable web statistics for a domain in Plesk article.
    Do adjustments in Tools & Settings > Server Statistics Settings - disabling some of the features should reduce the load on disk during calculation. Please check this documentation page for more information.
    Use Linux ionice feature to limit I/O throughput for statistics_collector. To do this:

1. Move /usr/local/psa/admin/sbin/statistics_collector to another location, for example /usr/local/psa/admin/sbin/real_statistics_collector.
2. Create a bash script named/usr/local/psa/admin/sbin/statistics_collector that will call for real_statistics_collector via ionice:

#!/bin/sh
ionice -c [scheduling class] -n [level] /usr/local/psa/admin/sbin/real_statistics_collector $@

Please note that it is only an example, and it is required to set scheduling class and level with values from ionice manual. You may check them here: ionice(1) - Linux man page. And it is also should be noted that this substitution is going to be removed during a Plesk update installation, so it will be required to reconfigure that after the updates.

SECOND :

The process vhostmng-find is a Plesk internal utility that performs various operations related to scanning virtual hosts directories. For example, it is used to index files in vhosts directories (indexing is mainly done to have faster responses from File Manager) and detecting document root subdirectories for applications. It is also used by WordPress Toolkit and by some other tasks (including statistics calculation).
The reasons why vhostmng-find stresses the IO subsystem of the server to its limit are the same as before - websites directory and files are highly nested and there is a 500 IOPS limit. To reduce IO usage by vhostmng-find you may try applying the similar workaround with usage of ionice, but it can result in slow work of Plesk in general and various errors in interface and logs:
1. Move /usr/local/psa/admin/sbin/vhostmng-find to /usr/local/psa/admin/sbin/vhostmng-find-real (for example).
2. Create a bash script named/usr/local/psa/admin/sbin/shostmng-find that will call for vhostmng-find-rea with ionice:

#!/bin/sh
ionice -c [scheduling class] -n [level] /usr/local/psa/admin/sbin/vhostmng-find-real $@

 
So, in case you face issues with Plesk after applying this workaround, I can only suggest reverting it back and then contact with website developers asking if they can reorganize website structure. Or you may also contact with hosting provider technical support and ask them if they can switch storage subsystem to other performance tier (with higher IOPS limit) for additional price.

Hope that will help !

Thank you for the answer. I knew this solution and already applied it and almost no difference. As I said in the first post, plesk not have solutions. An SSD or fewer files or giving up plesk. It is not possible to reduce the number of files. I have millions of images that are all useful. And pdf files as well. It does not mean that plesk is bad, but that it is not the optimal solution in my case. Plesk was and is very good for small/medium sites with few files. When you have over 15 million files... you need another solution. I will probably buy another simple dedicated server on which to put the static resources (images, pdfs, etc.) without plesk or other interfaces. All the best to everyone.