<Container id="Foo" type="STATELESS">
AccessTimeout = 30 seconds
MaxSize = 10
MinSize = 0
StrictPooling = true
MaxAge = 0 hours
ReplaceAged = true
ReplaceFlushed = false
MaxAgeOffset = -1
IdleTimeout = 0 minutes
GarbageCollection = false
SweepInterval = 5 minutes
CallbackThreads = 5
CloseTimeout = 5 minutes
UseOneSchedulerThreadByBean = false
EvictionThreads = 1
</Container>
Resources
All containers will be created automatically - which means you don’t need to define them if you don’t need to tune their configuration - when a bean of their type if found.
To avoid that use openejb.offline
property and set it to true
. See Server Configuration for more detail.
@Stateless
A @Stateless
container.
Declarable in tomee.xml via
Declarable in properties via
Foo = new://Container?type=STATELESS
Foo.AccessTimeout = 30 seconds
Foo.MaxSize = 10
Foo.MinSize = 0
Foo.StrictPooling = true
Foo.MaxAge = 0 hours
Foo.ReplaceAged = true
Foo.ReplaceFlushed = false
Foo.MaxAgeOffset = -1
Foo.IdleTimeout = 0 minutes
Foo.GarbageCollection = false
Foo.SweepInterval = 5 minutes
Foo.CallbackThreads = 5
Foo.CloseTimeout = 5 minutes
Foo.UseOneSchedulerThreadByBean = false
Foo.EvictionThreads = 1
Configuration
AccessTimeout
Specifies the time an invokation should wait for an instance of the pool to become available.
After the timeout is reached, if an instance in the pool cannot be obtained, the method invocation will fail.
Usable time units: nanoseconds, microsecons, milliseconds, seconds, minutes, hours, days. Or any combination such as "1 hour and 27 minutes and 10 seconds"
Any usage of the jakarta.ejb.AccessTimeout
annotation will
override this setting for the bean or method where the
annotation is used.
MaxSize
Specifies the size of the bean pools for this stateless SessionBean container. If StrictPooling is not used, instances will still be created beyond this number if there is demand, but they will not be returned to the pool and instead will be immediately destroyed.
MinSize
Specifies the minimum number of bean instances that should be in
the pool for each bean. Pools are prefilled to the minimum on
startup. Note this will create start order dependencies between
other beans that also eagerly start, such as other @Stateless
beans with a minimum or @Singleton
beans using @Startup
. The
start order.
The minimum pool size is rigidly maintained. Instances in the
minimum side of the pool are not eligible for IdleTimeout
or
GarbageCollection
, but are subject to MaxAge
and flushing.
If the pool is flushed it is immediately refilled to the minimum
size with MaxAgeOffset
applied. If an instance from the minimum
side of the pool reaches its MaxAge
, it is also immediately
replaced. Replacement is done in a background queue using the
number of threads specified by CallbackThreads
.
StrictPooling
StrictPooling tells the container what to do when the pool reaches it’s maximum size and there are incoming requests that need instances.
With strict pooling, requests will have to wait for instances to
become available. The pool size will never grow beyond the the
set MaxSize
value. The maximum amount of time a request should
wait is specified via the AccessTimeout
setting.
Without strict pooling, the container will create temporary instances to meet demand. The instances will last for just one method invocation and then are removed.
Setting StrictPooling
to false
and MaxSize
to 0
will result in
no pooling. Instead instances will be created on demand and live
for exactly one method call before being removed.
MaxAge
Specifies the maximum time that an instance should live before it should be retired and removed from use. This will happen gracefully. Useful for situations where bean instances are designed to hold potentially expensive resources such as memory or file handles and need to be periodically cleared out.
Usable time units: nanoseconds, microsecons, milliseconds,
seconds, minutes, hours, days. Or any combination such as
1 hour and 27 minutes and 10 seconds
ReplaceAged
When ReplaceAged
is enabled, any instances in the pool that
expire due to reaching their MaxAge
will be replaced immediately
so that the pool will remain at its current size. Replacement
is done in a background queue so that incoming threads will not
have to wait for instance creation.
The aim of his option is to prevent user requests from paying
the instance creation cost as MaxAge
is enforced, potentially
while under heavy load at peak hours.
Instances from the minimum side of the pool are always replaced
when they reach their MaxAge
, this setting dictates the
treatment of non-minimum instances.
ReplaceFlushed
When ReplaceFlushed
is enabled, any instances in the pool that
are flushed will be replaced immediately so that the pool will
remain at its current size. Replacement is done in a background
queue so that incoming threads will not have to wait for
instance creation.
The aim of his option is to prevent user requests from paying the instance creation cost if a flush performed while under heavy load at peak hours.
Instances from the minimum side of the pool are always replaced when they are flushed, this setting dictates the treatment of non-minimum instances.
A bean may flush its pool by casting the SessionContext
to
Flushable
and calling flush()
. See SweepInterval
for details on
how flush is performed.
import jakarta.annotation.Resource;
import jakarta.ejb.SessionContext;
import jakarta.ejb.Stateless;
import java.io.Flushable;
import java.io.IOException;
public class MyBean {
private SessionContext sessionContext;
public void flush() throws IOException {
((Flushable) sessionContext).flush();
}
}
MaxAgeOffset
Applies to MaxAge usage and would rarely be changed, but is a nice feature to understand.
When the container first starts and the pool is filled to the
minimum size, all those "minimum" instances will have the same
creation time and therefore all expire at the same time dictated
by the MaxAge
setting. To protect against this sudden drop
scenario and provide a more gradual expiration from the start
the container will spread out the age of the instances that fill
the pool to the minimum using an offset.
The MaxAgeOffset
is not the final value of the offset, but
rather it is used in creating the offset and allows the
spreading to push the initial ages into the future or into the
past. The pool is filled at startup as follows:
for (int i = 0; i < poolMin; i++) {
long ageOffset = (maxAge / poolMin * i * maxAgeOffset) % maxAge;
pool.add(new Bean(), ageOffset));
}
The default MaxAgeOffset
is -1 which causes the initial
instances in the pool to live a bit longer before expiring. As
a concrete example, let’s say the MinSize is 4 and the MaxAge is
100 years. The generated offsets for the four instances created
at startup would be 0, -25, -50, -75. So the first instance
would be "born" at age 0, die at 100, living 100 years. The
second instance would be born at -25, die at 100, living a total
of 125 years. The third would live 150 years. The fourth 175
years.
A MaxAgeOffset
of 1 would cause instances to be "born" older
and therefore die sooner. Using the same example MinSize
of 4
and MaxAge
of 100 years
, the life spans of these initial four
instances would be 100, 75, 50, and 25 years respectively.
A MaxAgeOffset
of 0 will cause no "spreading" of the age of the
first instances used to fill the pool to the minimum and these
instances will of course reach their MaxAge at the same time.
It is possible to set to decimal values such as -0.5, 0.5, -1.2,
or 1.2.
IdleTimeout
Specifies the maximum time that an instance should be allowed to sit idly in the pool without use before it should be retired and removed.
Usable time units: nanoseconds, microsecons, milliseconds, seconds, minutes, hours, days. Or any combination such as "1 hour and 27 minutes and 10 seconds"
GarbageCollection
Allows Garbage Collection to be used as a mechanism for shrinking
the pool. When set to true all instances in the pool, excluding
the minimum, are eligible for garbage collection by the virtual
machine as per the rules of java.lang.ref.SoftReference
and can be
claimed by the JVM to free memory. Instances garbage collected
will have their @PreDestroy
methods called during finalization.
In the OpenJDK VM the -XX:SoftRefLRUPolicyMSPerMB
flag can adjust
how aggressively SoftReferences are collected. The default
OpenJDK setting is 1000, resulting in inactive pooled instances
living one second of lifetime per free megabyte in the heap, which
is very aggressive. The setting should be increased to get the
most out of the GarbageCollection
feature of the pool. Much
higher settings are safe. Even a setting as high as 3600000 (1
hour per free MB in the heap) does not affect the ability for the
VM to garbage collect SoftReferences in the event that memory is
needed to avoid an OutOfMemoryException
.
SweepInterval
The frequency in which the container will sweep the pool and
evict expired instances. Eviction is how the IdleTimeout
,
MaxAge
, and pool "flush" functionality is enforced. Higher
intervals are better.
Instances in use are excluded from sweeping. Should an instance
expire while in use it will be evicted immediately upon return
to the pool. Effectively MaxAge
and flushes will be enforced as
a part of normal activity or sweeping, while IdleTimeout is only
enforcable via sweeping. This makes aggressive sweeping less
important for a pool under moderate load.
Usable time units: nanoseconds, microsecons, milliseconds,
seconds, minutes, hours, days. Or any combination such as
1 hour and 27 minutes and 10 seconds
CallbackThreads
When sweeping the pool for expired instances a thread pool is
used to process calling @PreDestroy
on expired instances as well
as creating new instances as might be required to fill the pool
to the minimum after a Flush or MaxAge
expiration. The
CallbackThreads
setting dictates the size of the thread pool and
is shared by all beans deployed in the container.
CloseTimeout
PostConstruct methods are invoked on all instances in the pool
when the bean is undeployed and its pool is closed. The
CloseTimeout
specifies the maximum time to wait for the pool to
close and PostConstruct
methods to be invoked.
Usable time units: nanoseconds, microsecons, milliseconds,
seconds, minutes, hours, days. Or any combination such as
1 hour and 27 minutes and 10 seconds
UseOneSchedulerThreadByBean
back to previous behavior (TomEE 1.x) where 1 scheduler thread was used for stateless eviction by bean (ie for 500 stateless beans you get 500 eviction threads)
EvictionThreads
number of threads to associate to eviction threads (1 is not bad for most applications)
@Stateful
A @Stateful
container.
Declarable in tomee.xml via
<Container id="Foo" type="STATEFUL">
AccessTimeout = 30 seconds
Cache = org.apache.openejb.core.stateful.SimpleCache
Passivator = org.apache.openejb.core.stateful.SimplePassivater
TimeOut = 20
Frequency = 60
Capacity = 1000
BulkPassivate = 100
</Container>
Declarable in properties via
Foo = new://Container?type=STATEFUL
Foo.AccessTimeout = 30 seconds
Foo.Cache = org.apache.openejb.core.stateful.SimpleCache
Foo.Passivator = org.apache.openejb.core.stateful.SimplePassivater
Foo.TimeOut = 20
Foo.Frequency = 60
Foo.Capacity = 1000
Foo.BulkPassivate = 100
Configuration
AccessTimeout
Specifies the maximum time an invocation could wait for the
@Stateful
bean instance to become available before giving up.
After the timeout is reached a jakarta.ejb.ConcurrentAccessTimeoutException
will be thrown.
Usable time units: nanoseconds, microsecons, milliseconds, seconds, minutes, hours, days. Or any combination such as "1 hour and 27 minutes and 10 seconds"
Any usage of the jakarta.ejb.AccessTimeout
annotation will
override this setting for the bean or method where the
annotation is used.
Cache
The cache is responsible for managing stateful bean instances. The cache can page instances to disk as memory is filled and can destroy abandoned instances. A different cache implementation can be used by setting this property to the fully qualified class name of the Cache implementation.
Passivator
The passivator is responsible for writing beans to disk
at passivation time. Different passivators can be used
by setting this property to the fully qualified class name
of the PassivationStrategy
implementation. The passivator
is not responsible for invoking any callbacks or other
processing, its only responsibly is to write the bean state
to disk.
Known implementations:
-
org.apache.openejb.core.stateful.RAFPassivater
-
org.apache.openejb.core.stateful.SimplePassivater
TimeOut
Specifies the time a bean can be idle before it is removed by the container.
This value is measured in minutes. A value of 5 would result in a time-out of 5 minutes between invocations. A value of -1 would mean no timeout. A value of 0 would mean a bean can be immediately removed by the container.
Any usage of the jakarta.ejb.StatefulTimeout
annotation will
override this setting for the bean where the annotation is used.
Frequency
Specifies the frequency (in seconds) at which the bean cache is checked for idle beans.
Capacity
Specifies the size of the bean pools for this stateful SessionBean container.
BulkPassivate
Property name that specifies the number of instances to passivate at one time when doing bulk passivation.
@Singleton
A @Singleton
container.
Declarable in tomee.xml via
<Container id="Foo" type="SINGLETON">
AccessTimeout = 30 seconds
</Container>
Declarable in properties via
Foo = new://Container?type=SINGLETON
Foo.AccessTimeout = 30 seconds
Configuration
AccessTimeout
Specifies the maximum time an invocation could wait for the
@Singleton
bean instance to become available before giving up.
After the timeout is reached a jakarta.ejb.ConcurrentAccessTimeoutException
will be thrown.
Usable time units: nanoseconds, microsecons, milliseconds,
seconds, minutes, hours, days. Or any combination such as
1 hour and 27 minutes and 10 seconds
Any usage of the jakarta.ejb.AccessTimeout
annotation will
override this setting for the bean or method where the
annotation is used.
@MessageDriven
A MDB container.
Declarable in tomee.xml via
<Container id="Foo" type="MESSAGE">
ResourceAdapter = Default JMS Resource Adapter
MessageListenerInterface = jakarta.jms.MessageListener
ActivationSpecClass = org.apache.activemq.ra.ActiveMQActivationSpec
InstanceLimit = 10
FailOnUnknowActivationSpec = true
</Container>
Declarable in properties via
Foo = new://Container?type=MESSAGE
Foo.ResourceAdapter = Default JMS Resource Adapter
Foo.MessageListenerInterface = jakarta.jms.MessageListener
Foo.ActivationSpecClass = org.apache.activemq.ra.ActiveMQActivationSpec
Foo.InstanceLimit = 10
Foo.FailOnUnknowActivationSpec = true
Configuration
ResourceAdapter
The resource adapter delivers messages to the container
MessageListenerInterface
Specifies the message listener interface handled by this container
ActivationSpecClass
Specifies the activation spec class
InstanceLimit
Specifies the maximum number of bean instances that are allowed to exist for each MDB deployment.
FailOnUnknowActivationSpec
Log a warning if true or throw an exception if false is an activation spec can’t be respected
@Managed
A managed bean container.
Declarable in tomee.xml via
<Container id="Foo" type="MANAGED" />
Declarable in properties via
Foo = new://Container?type=MANAGED
CMP entity
A CMP bean container.
Declarable in tomee.xml via
<Container id="Foo" type="CMP_ENTITY">
CmpEngineFactory = org.apache.openejb.core.cmp.jpa.JpaCmpEngineFactory
</Container>
Declarable in properties via
Foo = new://Container?type=CMP_ENTITY
Foo.CmpEngineFactory = org.apache.openejb.core.cmp.jpa.JpaCmpEngineFactory
Configuration
CmpEngineFactory
The engine to use for this container. By default TomEE only provides the JPA implementation.
BMP entity
A BMP entity container.
Declarable in tomee.xml via
<Container id="Foo" type="BMP_ENTITY">
PoolSize = 10
</Container>
Declarable in properties via
Foo = new://Container?type=BMP_ENTITY
Foo.PoolSize = 10
Configuration
PoolSize
Specifies the size of the bean pools for this bmp entity container.