ActiveMQ問題分析和解決

百聯達發表於2014-11-28

1)KahaDb和AMQ Message Store兩種持久方式如何選擇?

官方:

From 5.3 onwards - we recommend you use KahaDB - which offers improved scalability and recoverability over the AMQ Message Store.

The AMQ Message Store which although faster than KahaDB - does not scales as well as KahaDB and recovery times take longer.

 

非官方:

kaha檔案系統實際上上是一個檔案索引系統,有兩部分組成,一個是資料檔案系統,由一個個獨立的檔案組成,預設檔案大小是32M大(可配置),另外一個是索引檔案系統,記錄訊息在資料檔案中的位置資訊以及資料檔案中的空閒塊資訊。資料檔案是儲存到硬碟上的,索引檔案是快取在記憶體中的。所以這個儲存系統對大訊息儲存有利,象我們的memberId之類的文字訊息,實際上是浪費,索引比訊息還大,哈。

 

我方分析:

推薦: Amq持久方式

理由:雖然官方推薦使用KahaDB持久方式,但其提到的優勢:可伸縮性和恢復性較好,對於我們實際的應用意義不大。從我們自己的使用經驗來看,KahaDB持久方式,Data檔案是一個大檔案(感覺檔案過大後,造成佇列服務癱死的可能性會增大),從官網的相關配置(附錄1)也找不到哪裡可以設定資料的檔案的最大Size。)而Amq持久方式可以設定Data檔案最大Size,這樣可以保證即時訊息積壓很多,Data檔案也不至於過大。

 

 

2)錯誤:Channel was inactive for too long

解決方法:

在建立連線的Uri中加入: wireFormat.maxInactivityDuration=0

 

參考資源:

http://jinguo.iteye.com/blog/243153

 

You can do the following to fix the issues:

1) Append max inactivity duration to your Uri in the format below: wireFormat.maxInactivityDuration=0

2) Use the same Uri at the client side as well as at the server side

Regards,

 

如果不這樣設定,對應的錯誤會出現:

2008-05-07 09:22:56,343 [org.apache.activemq.ActiveMQConnection]-[WARN] Async exception with no exception listener: org.apache.activemq.transport.InactivityIOException: Channel was inactive for too long: localhost/127.0.0.1:61616

org.apache.activemq.transport.InactivityIOException: Channel was inactive for too long: localhost/127.0.0.1:61616

ActiveMQ的tcp url:tcp://localhost:61616後面要加入?wireFormat.maxInactivityDuration=0 這樣的引數,否則當一段時間沒有訊息傳送時會丟擲 "Channel was inactive for too long"異常

 

3)錯誤:Wire format negotiation timeout: peer did not send his wire format.

解決方法:

1)關閉ActiveMqLog4j

開啟:conf/log4j.properties

將:log4j.rootLogger=INFO, console, logfile

修改為:log4j.rootLogger=OFF

 

2)在建立連線的Uri中加入: maxInactivityDurationInitalDelay=30000

 

例如北京的測試環境連線Uri:

tcp://192.168.199.80:61616?wireFormat.maxInactivityDuration=0&maxInactivityDurationInitalDelay=30000&connection.AsyncSend=true

 

參考資源:

 

If you get exception like this,it can mean one of three things:

1. You're connecting to the port not used by ActiveMQ TCP transport

Make sure to check that you're connecting to the appropriate host:port

 

2. You're using log4j JMS appender and doesn't filter out ActiveMQ log messages

Be sure to read How do I use log4j JMS appender with ActiveMQ and more importantly to never send ActiveMQ log messages to JMS appender

 

3. Your broker is probably under heavy load (or network connection is unreliable), so connection setup cannot be completed in a reasonable time

If you experience sporadic exceptions like this, the best solution is to use failover transport, so that your clients can try connecting again if the first attempt fails. If you're getting these kind of exceptions more frequently you can also try extending wire format negotiation period (default 10 sec). You can do that by using wireFormat.maxInactivityDurationInitalDelay property on the connection URL in your client.

For example

tcp://localhost:61616?wireFormat.maxInactivityDurationInitalDelay=30000

will use 30 sec timeout.(貌似有問題!!!)

 

4)錯誤:Out of memory

解決方法:

1)  設定Java最大記憶體限制為合適大小:

Bin/activemq.bat 中ACTIVEMQ_OPTS=-Xmx512M(預設是512)

2)Activemq.xml配置節:systemUsage/ systemUsage配置大小合適,並且特別注意:大於所有durable desitination設定的memoryUsage之和。

 

備註:

1)尖括號:“>”代表萬用字元

2)ACTIVEMQ_OPTS的配置〉=memoryUsage中配置〉=所有durable desitination設定之和

 

 

 

3)SystemUsage配置設定了一些系統記憶體和硬碟容量,當系統消耗超過這些容量設定時,amq會“slow down producer”,還是很重要的。

 

 

參考資料:

http://m.oschina.net/blog/26216

參考--

       對於MQ的內容實用是可管理和可配置的。首先需要判斷的是MQ的哪部分系統因記憶體不足而導致洩漏,是JVM,broker還是消費者、生產者?

       一、記憶體管理

       JVM記憶體管理:

       1. 用bin/activemq命令在獨立JVM中執行broker。用-Xmx和-Xss命令即可(activemq.bat檔案中修改ACTIVEMQ_OPTS選項引數即可);

       2. 預設情況下,MQ用512M的JVM;

       broker記憶體管理:

       1. broker使用的記憶體並不是由JVM的記憶體決定的。雖然受到JVM的限制,但broker確實獨立管理器記憶體;

       2. systemUsage和destination的記憶體限制與broker記憶體息息相關;

       3. MQ中記憶體的關係是:JVM->Broker->broker features;

       4. 所有destination的記憶體總量不能超過broker的總記憶體;

       消費者:

       1. 由於訊息大小可以配置,prefetch limit往往是導致記憶體溢位的主要原因;

       2. 減少prefetch limit的大小,會減少消費者記憶體中儲存的訊息數量;

       生產者:

       1. 除非訊息數量超過了broker資源的限制,否則生產者不會導致記憶體溢位;

       2. 當記憶體溢位後,生產者會收到broker的阻塞資訊提示;

       二、其他

       將訊息緩衝之硬碟:

       1. 只有當訊息在記憶體中儲存時,才允許訊息的快速匹配與分發,而當消費者很慢或者離開時,記憶體可能會耗盡;

       2. 當destination到達它的記憶體臨界值時,broker會用訊息遊標來快取非持久化的訊息到硬碟。

       3. 臨界值在broker中透過memoryUsage和systemUsage兩個屬性配置,請參考activemq.xml;

       4. 對於緩慢的消費者,當尚未耗盡記憶體或者轉變為生產者併發控制模式前,這個特性允許生產者繼續傳送訊息到broker;

       5. 當有多個destination的時候,預設的記憶體臨界值可能被打破,而這種情況將訊息快取到硬碟就顯得很有意義;

       6. precentUsage配置:使用百分比來控制記憶體使用情況;

       多個執行緒:

       1. 預設情況下,MQ每個destination都對應唯一的執行緒;

       2. -Dorg.apache.activema.UseDedicatedTaskRunner=false(activemq.bat檔案中修改ACTIVEMQ_OPTS選項引數即可),用執行緒池來限制執行緒的數量,從而減少記憶體消耗;

       大資料傳輸:

       1. destination policies--maxPageSize:控制進入記憶體中的訊息數量;lazyDispatch:增加控制使用當前消費者列表的預取值;

       2. 使用blogMessage或者streamsMessage型別來進行大量檔案的傳輸;

       洩漏JMS資源:

       1. 當session或者producer或者consumer大量存在而沒有關閉的時候;

       2. 使用PooledConnectionFactory;

5)採用failover方式連線導致卡死

解決方法

不採用failover連線

 

分析

採用failover方式連線,如果所要連線的伺服器或者Activemq服務宕了,那麼程式會一直處於等待狀態,不超時,不報錯。

二)附錄

1)KahaDB持久配置

property name

default value

Comments

directory

activemq-data

the path to the directory to use to store the message store data and log files

indexWriteBatchSize

1000

number of indexes written in a batch

indexCacheSize

10000

number of index pages cached in memory

enableIndexWriteAsync

false

if set, will asynchronously write indexes

journalMaxFileLength

32mb

a hint to set the maximum size of the message data logs

enableJournalDiskSyncs

true

ensure every non transactional journal write is followed by a disk sync (JMS durability requirement)

cleanupInterval

30000

time (ms) before checking for a discarding/moving message data logs that are no longer used

checkpointInterval

5000

time (ms) before checkpointing the journal

ignoreMissingJournalfiles

false

If enabled, will ignore a missing message log file

checkForCorruptJournalFiles

false

If enabled, will check for corrupted Journal files on startup and try and recover them

checksumJournalFiles

false

create a checksum for a journal file - to enable checking for corrupted journals

Available since version 5.4:

   

archiveDataLogs

false

If enabled, will move a message data log to the archive directory instead of deleting it.

directoryArchive

null

Define the directory to move data logs to when they all the messages they contain have been consumed.

databaseLockedWaitDelay

10000

time (ms) before trying to get acquire a the database lock (used by shared master/slave)

maxAsyncJobs

10000

the maximum number of asynchronous messages that will be queued awaiting storage (should be the same as the number of concurrent MessageProducers)

concurrentStoreAndDispatchTopics

false

enable the dispatching of Topic messages to interested clients to happen concurrently with message storage

concurrentStoreAndDispatchQueues

true

enable the dispatching of Queue messages to interested clients to happen concurrently with message storage

 

2)AMQ持久配置

property name

default value

Comments

directory

activemq-data

the path to the directory to use to store the message store data and log files

useNIO

true

use NIO to write messages to the data logs

syncOnWrite

false

sync every write to disk

maxFileLength

32mb

a hint to set the maximum size of the message data logs

persistentIndex

true

use a persistent index for the message logs. If this is false, an in-memory structure is maintained

maxCheckpointMessageAddSize

4kb

the maximum number of messages to keep in a transaction before automatically committing

cleanupInterval

30000

time (ms) before checking for a discarding/moving message data logs that are no longer used

indexBinSize

1024

default number of bins used by the index. The bigger the bin size - the better the relative performance of the index

indexKeySize

96

the size of the index key - the key is the message id

indexPageSize

16kb

the size of the index page - the bigger the page - the better the write performance of the index

directoryArchive

archive

the path to the directory to use to store discarded data logs

archiveDataLogs

false

if true data logs are moved to the archive directory instead of being deleted

 

3)systemUsage配置

property name

default value

Comments

memoryUsage

20M

amq使用記憶體大小,照amq論壇上說,這個值應該大於所有durable desitination設定的

memoryUsage之和,否則會導致硬碟swap,影響效能。

storeUsage

1G

kaha資料儲存大小,如果設定不足,效能會下降到1個1個發

tempUsage

100M

非persistent的訊息儲存在temp區域

4)其他配置

 

4.1)Failover Transport Options

Option Name

Default

Description

transport.timeout

-1

Time that a send operation blocks before failing.

transport.initialReconnectDelay

10

Time in Milliseconds that the transport waits before attempting to reconnect the first time.

transport.maxReconnectDelay

30000

The max time in Milliseconds that the transport will wait before attempting to reconnect.

transport.backOffMultiplier

2

The amount by which the reconnect delay will be multiplied by if useExponentialBackOff is enabled.

transport.useExponentialBackOff

true

Should the delay between connection attempt grow on each try up to the max reconnect delay.

transport.randomize

true

Should the Uri to connect to be chosen at random from the list of available Uris.

transport.maxReconnectAttempts

0

Maximum number of time the transport will attempt to reconnect before failing (0 means infinite retries)

transport.startupMaxReconnectAttempts

0

Maximum number of time the transport will attempt to reconnect before failing when there has never been a connection made. (0 means infinite retries) (included in NMS.ActiveMQ v1.5.0+)

transport.reconnectDelay

10

The delay in milliseconds that the transport waits before attempting a reconnection.

transport.backup

false

Should the Failover transport maintain hot backups.

transport.backupPoolSize

1

If enabled, how many hot backup connections are made.

transport.trackMessages

false

keep a cache of in-flight messages that will flushed to a broker on reconnect

transport.maxCacheSize

256

Number of messages that are cached if trackMessages is enabled.

transport.updateURIsSupported

true

Update the list of known brokers based on BrokerInfo messages sent to the client.

 

4.2)Connection Options

Option Name

Default

Description

connection.AsyncSend

false

Are message sent Asynchronously.

connection.AsyncClose

true

Should the close command be sent Asynchronously

connection.AlwaysSyncSend

false

Causes all messages a Producer sends to be sent Asynchronously.

connection.CopyMessageOnSend

true

Copies the Message objects a Producer sends so that the client can reuse Message objects without affecting an in-flight message.

connection.ProducerWindowSize

0

The ProducerWindowSize is the maximum number of bytes in memory that a producer will transmit to a broker before waiting for acknowledgement messages from the broker that it has accepted the previously sent messages. In other words, this how you configure the producer flow control window that is used for async sends where the client is responsible for managing memory usage. The default value of 0 means no flow control at the client. See also

connection.useCompression

false

Should message bodies be compressed before being sent.

connection.sendAcksAsync

false

Should message acks be sent asynchronously

connection.messagePrioritySupported

true

Should messages be delivered to the client based on the value of the Message Priority header.

connection.dispatchAsync

false

Should the broker to the connection's consumers.

 

4.3)OpenWire Options

Option Name

Default

Description

wireFormat.stackTraceEnabled

false

Should the stack trace of exception that occur on the broker be sent to the client? Only used by openwire protocol.

wireFormat.cacheEnabled

false

Should commonly repeated values be cached so that less marshalling occurs? Only used by openwire protocol.

wireFormat.tcpNoDelayEnabled

false

Does not affect the wire format, but provides a hint to the peer that TCP nodelay should be enabled on the communications Socket. Only used by openwire protocol.

wireFormat.sizePrefixDisabled

false

Should serialized messages include a payload length prefix? Only used by openwire protocol.

wireFormat.tightEncodingEnabled

false

Should wire size be optimized over CPU usage? Only used by the openwire protocol.

wireFormat.maxInactivityDuration

30000

The maximum inactivity duration (before which the socket is considered dead) in milliseconds. On some platforms it can take a long time for a socket to appear to die, so we allow the broker to kill connections if they are inactive for a period of time. Use by some transports to enable a keep alive heart beat feature. Set to a value <= 0 to disable inactivity monitoring.

maxInactivityDurationInitalDelay

10000

The initial delay in starting the maximum inactivity checks (and, yes, the word 'Inital' is supposed to be misspelled like that)

 

5)安全配置

1、控制檯安全配置,開啟conf/jetty.xml檔案,找到

   

       

       

       

   

   將“false”改為“true”即可。使用者名稱和密碼存放在conf/jetty-realm.properties檔案中。

2、生產者和消費者連線MQ需要密碼

   開啟conf/activemq.xml檔案,在標籤裡的標籤前加入:

     

  

  

  

  

  

  

 注意必須在標籤前,否則啟動ActiveMQ會報錯。

 

使用者名稱和密碼存放在conf/credentials.properties檔案中

來自 “ ITPUB部落格 ” ,連結:http://blog.itpub.net/28624388/viewspace-1349867/,如需轉載,請註明出處,否則將追究法律責任。

相關文章