Configuration for RAM and CPU

[gcelano]

My Annis instance (4.12.2) seems to get clogged when I open more windows (sometimes, not even the initial Annis logo appears), as if there are not enough resources available on the server (even if it has 256 CPUs and 200GB for RAM). I have imposed no RAM limit on the service (annis.service), but I set up Nice=7 for it (ApacheHTTPServer has MemoryLimit=104857600000). No matter how many windows I open, the RAM memory consumption does not increase (while all CPUs are used), and therefore I am not sure whether there is some limit in place within Annis and/or ApacheHTTPServer preventing more/faster access to Annis. I changed Apache mpm_event.conf, but it does not seem to affect anything:

StartServers            4
MinSpareThreads         25
MaxSpareThreads         75
ThreadLimit             128
ThreadsPerChild         50
MaxRequestWorkers       300
MaxConnectionsPerChild  0

my service.toml is

[auth]
anonymous_access_all_corpora = true

[database]
graphannis = "/var/local/annis/.annis/v4"
sqlite = "/var/local/annis/.annis/v4/service_data.sqlite3"
disk_based= false
cache = {PercentOfFreeMemory = 50.0}

Do you have any suggestion? Should I change service.toml in some respect?

[thomaskrause]

Hi, 200GB of RAM seems plenty and even half of it would be used for loading the graph in memory, it should still be enough for the other services.

There can be several bottlenecks, and you already checked Apache and the graphANNIS service. You can check that the graphANNIS service configuration is actually used by the service by looking at the output/log and searching for the line Starting the bundled graphANNIS service with configuration file /tmp/.... This shows a path to the actual file used, which might be a modified one derived from the one you configure. You can open it and compare it with the configuration you intended and check that the relevant parts are the same (there might be some differences for the auth configuration).

Another bottleneck can be the interaction between the java-based server that services the web application (which is based on Jetty) and the Apache2 server which acts as a proxy. The Apache2 needs to proxy Websocket requests, which sometimes get interrupted and this looks like the website is not responding anymore, even if the initial frontend page was loaded. You can check with the Firefox or Chrome developer tools if the network connections are not failing. If there is a publicly reachable instance of your server, I can also check if I see any errors from the browser when trying to access it.

P.S. I converted the issue to a discussion, because it is not clear yet where a possible bug in ANNIS might be. We should try to narrow the issue down and then create issues for the specific found bug(s) in ANNIS