Most slow reports spend their time in the database. Find out where the time goes before you change server settings.

Find the slow query

In edit mode, point to the component, open ⋮ and select Execution cost. It lists the SQL statements the engine sent for that component. Run them in your database tool: if they are slow there too, tune the database (indexes, partitions, statistics) or the model, not Datafor. See Report Editor Basics.

Datafor settings

Lever	Where	Effect
Engine limits	Settings › Data › Query engine: Max concurrent queries, Max result rows, Query timeout (seconds)	Caps the load one server puts on the databases and stops runaway queries. See Query Engine.
SQL threads	mondrian.rolap.maxSqlThreads (default 100) and mondrian.rolap.maxSqlThreadsPerQuery in bi-server/pentaho-solutions/system/datafor/mondrian.properties	How many SQL statements run at once in total and per query. See Query Engine.
Rows per component	Page › Settings › Performance › Max query records; the default (5,000) is Default chart query row limit in Settings › General › System configuration	Fewer rows load and render faster. See Page Settings and System Configuration.
Model cache	Cache expire in the model; the default for new models is Default model cache expiration in System configuration (0 = never expires)	Repeated queries are answered from the cache instead of the database. Use an expiration that matches how often the data is loaded. See Creating an Analysis Model.
Connection pool	Pooling in the datasource; server defaults in dbcp-defaults in pentaho-solutions/system/pentaho.xml (maxActive 20)	A busy datasource may need a larger pool. A larger pool does not make a slow query faster. See Configuring MySQL Data Source.
Aggregation tables	Model	Large fact tables are answered from pre-aggregated tables. See Use Aggregation Tables.
AI Agent workers	AI_AGENT_DISPATCH_WORKERS in bi-server/ai-agent/.env (1–8, default 2)	Questions to the AI Agent stop queuing behind each other. See Managing High Concurrency.

Memory (heap size)

The Datafor server runs in a Java virtual machine (Java is bundled). Its heap is set with two options:

• -Xms: heap size at start. Default 2048m.
• -Xmx: maximum heap size. Default 6144m (6 GB).

Raise -Xmx when the server logs java.lang.OutOfMemoryError: Java heap space, or when many users open large reports at the same time. Size it so that the heap, the built-in PostgreSQL database, the AI Agent and the operating system together fit in physical memory: a heap larger than the free RAM makes the server swap and slows it down. For example, on a 16 GB server that also runs the built-in database and the AI Agent, -Xmx10240m leaves about 6 GB for the rest.

Linux and Windows (start script)

1. Stop Datafor.

2. Open the start script in the bi-server folder:

   • Linux: start-server.sh
   • Windows: start-server.bat

3. Find the line that sets CATALINA_OPTS and change the two values:

   # start-server.sh
   CATALINA_OPTS="-Xms2048m -Xmx10240m -Dsun.rmi.dgc.client.gcInterval=3600000 ..."

   rem start-server.bat
   set CATALINA_OPTS=-Xms2048m -Xmx10240m -Dsun.rmi.dgc.client.gcInterval=3600000 ...

   Change only the numbers and leave the other options on the line as they are.

4. Start Datafor. The Linux script prints the options it used (Using CATALINA_OPTS: …).

An update package replaces start-server.*, set-env.* and tomcat/conf/server.xml; the replaced files are kept in bi-server/update/backup<time stamp>/. Note your -Xms/-Xmx values before every update and restore them in the CATALINA_OPTS line afterwards.

Windows service

If Datafor runs as the Windows service DataforSolutionServer, the start script is not used. The service takes its heap from the values that tomcat\bin\service.bat passes when the service is installed (JvmMs 2048 and JvmMx 6144, in MB). To change them on an installed service, run as administrator:

bi-server\tomcat\bin\tomcat9w.exe //ES//DataforSolutionServer

On the Java tab, set Initial memory pool and Maximum memory pool (in MB), click OK, and restart the service. Alternatively, change JvmMs and JvmMx in service.bat, then remove and reinstall the service.

Docker

The container runs the same /opt/bi-server/start-server.sh, which sets CATALINA_OPTS itself, so passing CATALINA_OPTS or JAVA_OPTS with docker run -e has no effect. Edit the script in the container and restart it:

docker cp datafor-ee:/opt/bi-server/start-server.sh .
# edit -Xmx in start-server.sh
docker cp start-server.sh datafor-ee:/opt/bi-server/start-server.sh
docker exec -u root datafor-ee chown biadmin:biadmin /opt/bi-server/start-server.sh
docker exec datafor-ee /opt/bi-server/stop-server.sh
docker restart datafor-ee

Edit the file with an editor that keeps Linux line endings. The change survives a new container only if /opt/bi-server is on a volume (see Deploying Datafor Using Docker). If you limit the container's memory with --memory, keep the limit well above -Xmx, because PostgreSQL runs in the same container.

Tomcat

The web connector is in bi-server/tomcat/conf/server.xml. It listens on port 28080 and already compresses responses (compression="on", compressionMinSize="2048" for HTML, JavaScript, CSS, JSON and text), so there is nothing to enable.

It sets no maxThreads, so Tomcat handles at most 200 requests at the same time. Raise it only if many users work at once and requests wait while CPU and memory are still free. Add the attribute to the existing <Connector … port="28080" …> element, keep its other attributes, and restart Datafor:

<Connector URIEncoding="UTF-8" ... port="28080" protocol="HTTP/1.1"
           connectionTimeout="20000"
           redirectPort="28443"
           maxThreads="400"
           ... />

For HTTPS, caching of static files and long-running requests, put Nginx in front of Datafor. See Deploying Datafor Behind Nginx.

Garbage collection log

To see whether pauses come from garbage collection, add these options to the CATALINA_OPTS line of the start script (see Memory (heap size)) and restart:

-Xloggc:/opt/bi-server/tomcat/logs/gc.log -XX:+PrintGCDetails -XX:+PrintGCDateStamps -XX:+UseGCLogFileRotation -XX:NumberOfGCLogFiles=5 -XX:GCLogFileSize=20M

Use your own installation path. These options are for the bundled Java 8. If the server runs on Java 9 or later, use -Xlog:gc*:file=/opt/bi-server/tomcat/logs/gc.log instead: Java 9 and later do not start with some of the Java 8 options (for example -XX:+PrintGCDateStamps), and Java 8 does not start with -Xlog. No JIT options are needed; Java 8 already enables tiered compilation and compressed object pointers.