Hi Christophe.
Seeing the data that you have stored and the time that you have been running the environment, it is possible that the database is fragmented and for that reason there may be a degradation in the reading and writing with the database.
Anyway, if you do not have problems of unknowns and queuing in the servers, it is possible that you do not have problems with the current data stored in database. As you have the community version you cannot install a second historical database reducing the data in the main one to 15 days to make it work faster and sending the rest of the data to the historical one, being able to store in it up to years. https://pandorafms.com/manual/en/documentation/05\_big\_environments/07\_server\_management?s\[\]=historical&s\[\]=database#history\_database
A question, do you have monitored the exact memory consumption of the pandora_server process? Do you notice an increase in it inside the ps aux command? It would be nice to see how the graph of this module looks like...
module_begin
module_name Pandora Server memory
module_type generic_data
module_exec ps aux | grep pandora_server | grep "D" | gawk '{print $4}'
module_end
In the case that it is the free memory that is decreasing in the server and the available memory is kept because it starts to pass to buffer/cache memory, it is a normal behavior of the server due to the memory management of the linux systems,
Some time ago a memory leak was discovered in openssl / curl libraries. I have seen that you have enabled the webserver, do you have active web monitoring to URL with https? In the Enterprise version it is compiled into the binary and stopped happening ( webserver recently only was Enterprise), but it is possible that the openssl, curl and perl-Net-SSLeay versions may cause extra memory consumption.
You can try to uncomment the following line ( web_engine lwp ) in the pandora_server.conf to use LWP instead of curl in case you have monitoring of this type added and it can cause to these libraries.
Regards