Who met with the problem of slow work of WS under Apache in linux version 1C 8.2?

Question

linux

A 1C 8.2 server is configured with a PostgreSQL base, WebService is raised on it under Apache. Linux - most likely Debian, if necessary - I will clarify.
There is SOAP function which returns the data in the form of XML.
In the configurator, a request with a response of half a megabyte is completed in 7 seconds, through WS - in 50+ seconds.
Some requests with a response of 30Kb are returned by the web service in 50+ seconds, others, even larger ones, in the region of a couple of seconds.
It was not possible to find a relationship between such a depressing time difference between different requests.
Timeouts everywhere are increased.
The question is, who had a similar problem and who fought it?

Answer 1

There were no such problems (there were many others that are much worse - you don’t even suspect how many pitfalls there are in the 1C Linux server).
Let's look at common sense.
1) Requests are not executed in the configurator - only configuration and administration takes place there. You most likely performed queries in a thin or thick client that is connected to the server. Since you mentioned the configurator, I am inclined to conclude that we are talking about a thick client with a connection to the server on the local network via TCP.
2) Your web services are raised on Apache and are served using the Apache module, which, like a thin / thick client, connects to the 1C server via TCP, receives data from the server, and then returns a response to the client via HTTP (S).
On the surface, there is a conclusion that in both cases the server works at the same speed (if at the time of testing the load was identical), but your extra 50 seconds are spent serving HTTP (s) requests / responses by the web server. Check in this direction - you are 70% sure that either your Apache is configured through w, or other loaded applications are spinning on it, in addition to the published 1C, and time for the 1C module is allocated according to the residual principle.
Another option is that you did not tell everything. My second version is that you get the request data in a thick client in a single sample, but you do not make a web service for a single request, but re-establish a ws connection for each element of the sample. With this approach, even on a dedicated server, the speed of web services in the case of big data will be hundreds and thousands of times slower than working in 1C clients.