question

Upvotes
Accepted
1 0 1 2

路透RFA消息队列的消息量很大,导致内存溢出

我是中国进出口银行的路透数据源开发厂商,有个问题我想咨询一下

我使用JAVA RFA,版本付下

Manifest-Version: 1.0
Specification-Title: Classes for RFAJ 5.1
Implementation-Version: Build 2004.12.15
Created-By: 1.4.0_01 (Sun Microsystems Inc.)
Specification-Vendor: Reuters

连接路透数据源服务器,连接配置详见附件subscribe.txt

报文订阅,详见附件code.txt

报文处理,详见附件comsume.txt

在处理一段时间候,出现了如附件outofmemory.jpg的内存溢出问题outofmemofy.jpg

treprfarfa-api
outofmemofy.zip (349.5 KiB)
subscribe.txt (6.0 KiB)
icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
Accepted
281 1 3 7

The latest release of RFA Java that supports the Market Data programming interface (SSL connection) is RFA Java 7.6.1.L1 release. However this would require the upgrade of Java version as well since RFA Java 7.4 or higher requires Java 7.

From the given heap screen snapshot, large amount of memory was retained by EventQueue object, which basically holds data events received from the infrastructure. This indicates the possibility that a lot of data events were pending (for processing) in the event queue and the application was likely experiencing the slow consumer condition - the application did not process events fast enough.

Generally, there are two main approaches to address the slow consumer problem, i.e. a) improving the event processing capability of the application, and b) reducing the number of updates (by applying update conflation).

To improve the event processing, the developer could review and optimize the event processing routine so that less time is spent on processing events. The application can also be revised to have multiple sessions and connections (having more threads) to distribute the workload across different CPU cores.

On the other hand, if the application cannot be changed, performing the update conflation (on the infrastructure) should be considered. This would reduce the number of updates the application has to handle; however, at the expense of not having all ticks of updates.

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvote
1.2k 23 31 44

Please see the following answer investigating RFA memory growth. Rising data rates with slow consumers, either by hardware limits or by complex processing will see additional buffer consumption and eventual memory exhaustion.

It is recommended to investigate enabling just-in-time conflation in TREP to reduce the data rates being presented to the consuming application during peak trading hours.

Note that RFA/Java 5.x, 6.x, 7.0, and 7.2 are no longer supported, similarly JRE versions 4, 5, and 6. Code changes will be required to upgrade, and it may be easier to migrate to the new Elektron SDK (EMA) instead.

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Write an Answer

Hint: Notify or tag a user in this post by typing @username.

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.