In prior releases of Payara Server, it was not possible to control the maximum number of concurrent Stateless EJB instances in Payara Server. It was, however, possible to control the number of pooled Stateless EJB instances, as well as concurrent MDB instances. These features were available in Oracle GlassFish Server 3.1 and earlier but not in the GlassFish Open Source editions (3.1.2.x and 4.x).
In the current release of Payara Server (171), it is now possible to limit concurrent Stateless EJB instances that are dispatched, allowing fine-grained control of resources, limiting surface area for DDOS attacks and making applications run more smoothly and efficiently.
The key difference is that previously it was not possible to limit the number of concurrent threads dispatched for the same Stateless EJB, the only limit was the number of pooled beans. When the pool ran out, new EJB instances were created, with no upper bound. In the current release, this upper bound can be established so the maximum number of threads is not exceeded beyond this value.
These limits are controlled on a per-EJB basis, using
bean-pool elements in the
glassfish-ejb-jar.xml deployment descriptor file:
<max-pool-size>This element controls the maximum concurrent instances that are dispatched for this EJB (number of threads). The default is configured in the domain; this has not changed from previous versions.
<max-wait-time-in-millis>This element controls what to do when the number of requests exceeds the amount of beans available in the pool. This is a new feature.
- Value of -1 (default)
Current behaviour is unchanged, when the pooled number is exceeded, a new EJB instance is created, there is no upper bound
- Value of 0
This is the new behaviour, when the pooled number is exceeded, the request will wait for a bean to free up in the pool. This effectively caps the number of threads that are concurrently created for a this particular EJB request.
- Value between 1 - MAX_INTEGER
This is the new behaviour, when the pool number is exceeded, the request will wait for a bean to free up in the pool, up to the number of milliseconds specified here. After this time has expired, a new EJB instance is created and dispatched, thus the configured pool size acts as a soft upper bound, one that can be exceeded once the timer has expired.
- Value of -1 (default)
Additionally, a new system property
fish.payara.ejb-container.max-wait-time-in-millis can be set to change the default global value of
<max-wait-time-in-millis> for all Stateless EJB bean pools. Unless overridden in the deployment descriptor file, this will become the new default value and can be used to cap the upper bound of all concurrent invocations of any Stateless EJB pools.
<steady-pool-size> element controls the minimum number of beans in the bean pool. This will increase performance at the expense of memory footprint.
Here's a complete example: