X
X
x2sp2015-04-01 18:58:06
1C
x2sp, 2015-04-01 18:58:06

How to solve the problem with a long connection to 8.3 databases (with working 8.2 databases)?

Technical part (initial):
1) 1s cluster.
OS - 2k8 r2 (VM)
4 GB of RAM 2 cores
is used only as a 1s cluster and license issuance
2) sql 2008 (VM)
OS - 2r8 r2
16 GB of RAM 4 cores
is used only for SQL
3) sql 2008 (VM)
OS - 2k8 r2
12 GB of RAM 4 cores
are used only for SQL
Physics:
1.2 VMs are in a 2k8 r2 cluster consisting of 2
e5645 nodes (2 pcs) and 64 GB hp360g7 + sas iscsi hp p2000g3
3 VMs are in standaline 2k8 r2
e5- 2690 (2pcs) and 64gb hp360g8
Network: Cisco 2960g (1 subnet) - 1gb
It is planned to switch to version 8.3, so at the moment two platforms 8.2.19.83 and 8.3.5.1486 are working. (key 64 bit)
7 bases 8.2 with the number of fields not more than 30 (sql1) 5 bases 8.3 with the number of fields from
1-2 (sql2)
bases 8.3 (from 30 -5 minutes)
and take off half and due to lack of memory.
The following works have been done:
a) Optimization of cluster memory 8.3 cluster
parameters:
Restart interval: 86400
Memory exceedance interval: 30
Stop running processes after: 30
Memory priority
working server parameters:
Safe memory consumption 512 MB
The amount of memory for working processes, up to which the server is considered to be the driver 700 mb
Number of IS per process: 1
Number of connections per process 25
Solved the problem with crashes
2) Reconfiguring clients to search for a license from the server side with fixing the file on the server nethasp.ini
NH_SERVER_ADDR = 192.168.112.18
NH_USE_BROADCAST = Disabled
In this case, we have 30 software licenses and a key for 10 on server 18.
3) After the start of the agent 8.3 service, a problem was fixed in the logs:
Failed application ragent.exe, version 8.3.5.1486, time stamp 0x54f76689, failed module rtrsrvc.dll, version 8.3.5.1486, time stamp 0x54f7685d, exception code 0xc0000005, error offset 0x00000000000020f4, process id 0xcb4, application start time
4) There was an attempt to rearrange the platform (8.3) from scratch, but the error remained.
5) It was decided to combine the bases with the first chin.
6) Technical journal 1c was given to ITS support, no errors were found.
At the moment, all VMs lie in the Hyper-V cluster. Of the proposed solutions, only 2:
a) Separate services for different users
6) Change the port range on the 8.3 service
c) Raise 2012 r2 (it is planned to switch the infrastructure to hyper-v 3.0) and install the school inside the machine, see what happens.
When the 1c cluster functioned only as 1c 8.2, the amount of memory was - 2 GB with 1 core. And never had any performance issues. Therefore, an increase in RAM up to 4 GB and cores up to 2 is due to the fact that the databases are working in test mode. Considering that all transactions are processed by the school and requests are made by the user as a fat client. I did not see the point in setting a larger value at that time. Moreover, the problems were fixed mainly in the morning, it does not matter one person enters or 4 at once. In the middle of the day there were no problems.
PS 8.2 works as before, everything is fine with it.

Answer the question

In order to leave comments, you need to log in

3 answer(s)
X
x2sp, 2015-05-04
@x2sp

The victory was not easy...
I will try to describe everything in detail, although quite a lot of time has passed. I hope the information I have collected will help system administrators and give me something to think about.
I moved the databases to 1 school and reassigned the user to the 8.3 agent - it didn’t help ...
I traveled our Runet up and down and found two interesting articles that I suggest you read.
For the first time, a fairly detailed test between two platforms: the PS-middle variant and the dual-core server. It turns out quite interesting points that were also emphasized by me while working in other companies.
efsol.ru/articles/tuning-1c.html
The second one refers to 1c tests on virtual machines. In it, I just saw the reason for the delay in starting the configuration:
efsol.ru/articles/performance-comparison-1c.html
After digging around for about a week and trying several methods, I could not solve the problem with the first launch. But I noticed one interesting pattern... When a thin client is running, the second launch of the system occurs almost instantly. Changed the settings for the number of IS per process: 8 (bases on 8.3 so far 5). As a result, since the server stopped wasting time on creating RPHOST when entering the trail. the base and the rest he spent only on unloading the conf from whining. Reduced the start time of the second bases by 10-7 seconds.
In principle, this option suits me completely, given that 7-10 users work with each database, the conf is constantly kept in RPHOSTe and the login time is 4-8 seconds with authentication together.
If you have problems that the base is not opened often, then as an option I can offer to file a small reg. user login task to each database and set the service to restart in the evening (either through services or restart interval). I think this should help, although here we run into the presence of a license, so we have to think)))
But another unpleasant moment surfaced, on one of the forums I received the following answer:
Offtop. Do you have CORP licenses?
Extended features of the 1C:Enterprise 8.3 CORP level server compared to the 64-bit PROF level server:
* safe memory consumption per call;
* number of IS per process;
* the amount of worker process memory up to which the server is considered productive;
* the maximum amount of memory of worker processes;
*balancing strategy (by memory, by performance);
The use of the listed functionalities with the help of 1C:Enterprise 8. Server license (x86-64) products of the PROF level, that is, those that do not have the CORP designation in the name, is illegal.
I decided to check with the guys from three different companies if they know that there is a corp and prof platform. To which an answer was received in the form of: a spinning finger at the temple and a package to the 1c forum. And here is the answer from support 1c below:
1) >>> I would like to clarify the differences between the platform 8.3 corp and 8.3 prof.
https://partners.v8.1c.ru/forum/message/1301566#m_...
www.1c.ru/news/info.jsp?id=16733
In fact, when using the PROF Platform, according to the license, you can only use the default settings of the cluster.
If you have problems with the "default" cluster settings (lack of memory, inability to update the configuration, etc.), then
this behavior is an error (either of the platform or of this application solution).
The request with concrete examples to create requests for correction.
At the time of correction of the error, permission can be issued in writing (signed by the director of 1C CJSC) to use the functionality of the Corp. license.
2) >>> Please clarify, i.e. there are two platforms 8.3?
No. The platform is not currently alone.
There is also no software control of this license at the moment.
So the CORP license is more of a legal concept.
To be honest, I do not believe in various types of conspiracies, but when a company has an almost 100% monopoly in the SME market, thoughts are different.
“There will be an update of some kind of configuration for reporting, which will require an update of the platform in which program control will already be implemented ... And then you (1s) will pay us (1c) in full ... we are children.”
PS I want to clarify that my server is based on 2008r2 and there may be differences from 2012. All the same, the kernel is cut in detail there and hyper-v 3.0 also grew like buns. But as they say "IT`s Alive!!!" and work 1s on virtual machines is not only possible, but also welcomed. As a result, we have 30 users 8.2 and 20 users 8.3. Good luck to you all, be patient and never give up)))

C
Cool Admin, 2015-04-01
@ifaustrue

Where do users end up? Perhaps the problem is there. When starting a thick client, 1C intensively caches a bunch of files in the user profile and at the pace, if there is a slow disk or a queue full, the start and connection will be delayed.

Y
Yuri Goncharuk, 2015-04-02
@yukon39

Number of IS per process: 1

Take a closer look here. In this form, when connecting to a database that has no connections, the system launches a new process in the OS, opens a connection to IB in it, initializes the configuration data, and only then responds to the user. Perhaps "problems are fixed mostly in the morning" just from here.
offtop. Do you have CORP licenses?
Extended features of the 1C:Enterprise 8.3 CORP level server compared to the 64-bit PROF level server:
* safe memory consumption per call;
* number of IS per process;
* the amount of worker process memory up to which the server is considered productive;
* the maximum amount of memory of worker processes;
*balancing strategy (by memory, by performance);
The use of the listed functionalities with the help of 1C:Enterprise 8. Server license (x86-64) products of the PROF level, that is, those that do not have the CORP designation in the name, is illegal.

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question