It's also worth noting that paging is a palliative measure. The broker is not designed to be consistently in a state where production > consumption such that more and more messages keep accumulating. Paging is meant to provide temporary relief.
If you find that production > consumption more often than not then you might consider mixing something like a database into your architecture somewhere to store your message data. It's a classic anti-pattern to use a message broker as a database [1]. Justin On Tue, Jun 6, 2017 at 3:15 AM, Денис Кирпиченков < denis.kirpichen...@gmail.com> wrote: > Miroslav, thanks for advice! > I've decreased memory usage by tunning broker.xml configuration. > JournalImpl is not consuming a lot of memory now. But problem still exists > - org.apache.activemq.artemis.core.paging.impl.PaginManagerImpl has > become > a problem. It holds an information about all transactions in a hashmap > org.apache.activemq.artemis.core.paging.impl.PagingManagerImpl. > transactions. > And when queue has, lets say 100k messages, it may cause an > OutOfMemoryError for apps with small heap (-Xmx256m) > > May be there are another configuration options I'm missing? > > > On Fri, Jun 2, 2017 at 2:49 PM, Miroslav Novak <mno...@redhat.com> wrote: > > > Hi, > > > > I think you need to configure max-size-bytes in address-settings. See [1] > > for more information. Basically you specify how much memory can every > > queue/topic consume. Next you can set address-full-policy to BLOCK, if > you > > don't want to page messages to disc if max-size-bytes was exceeded. It > will > > hold producers until some messages are consumed. > > > > I would also recommend to configure consumer-window-size to 0 if your > > consumers are slow. It will avoid buffering of messages on client side > > message buffer so those messages can be consumed by other consumers on > > queue which are idle. > > > > I hope this helps, > > Mirek > > > > [1] https://github.com/apache/activemq-artemis/blob/master/ > > docs/user-manual/en/paging.md#configuration-1 > > > > ----- Original Message ----- > > > From: "Денис Кирпиченков" <denis.kirpichen...@gmail.com> > > > To: dev@activemq.apache.org > > > Sent: Friday, June 2, 2017 11:20:12 AM > > > Subject: org.apache.activemq.artemis.core.journal.impl.JournalImpl > > > > > > Hello, All! > > > > > > I work on an application that uses artemis in embedded mode and > sometimes > > > (due to unexpected external coditions) the app processes any single > > message > > > very slow. But in any cases artemis has constantly high rate of input > > > messages. For example, artemis receives 100-200 messages per second, > but > > > due to slow consumers it could deliver only 10-20 messages per second. > > > As you can image, described conditions cause an OutOfMemoryError, > > > eventually. > > > After an investigation I found that ConcurrentHashMap<Long, > > > JournalRecord> records in > > > org.apache.activemq.artemis.core.journal.impl.JournalImpl holds a lot > of > > > information about messages which is waiting to be delivered. > > > > > > Artemis is configured to have a persisted journal, but anyway > > > JournalImpl.records grows on any new received message as far as JVM > heap > > > allows. > > > > > > And my questions, is there any workaround? maybe configuration tweaks? > > > How do you think is it ok to make JournalImpl.records backed by > > > disk-storage, to limit it's heap usage. > > > > > > -- > > > Best regards, > > > Denis > > > > > >
