activemq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "John Fletcher (JIRA)" <j...@apache.org>
Subject [jira] Updated: (AMQ-2945) KahaDB corrupted when too many messages accumulate: EOFException
Date Mon, 27 Sep 2010 08:42:40 GMT

     [ https://issues.apache.org/activemq/browse/AMQ-2945?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

John Fletcher updated AMQ-2945:
-------------------------------

    Description: 
I am finding the KahaDB on a broker is becoming corrupted after a certain time of being disconnected
(or perhaps rather, after a certain amount of messages are queued up). My config is extremely
simple - originally 5.4.0 and now 5.4.1 on Windows with the default config, except adding
a connection to a broker on an external machine with
<networkConnectors>
  <networkConnector uri="static:(tcp://machine1:61616,tcp://machine2:61616)"/>
</networkConnectors>

In this case the external machine/consumer is only on during business hours whereas the broker
producing the messages is on 24/7. We're finding that generally the producer instance survives
weeknights, during which it accumulates about 75,000 messages and then offloads them in the
morning to
the consumer instance, but not weekends. Last time it failed after about collecting about
160,000 messages.

I get the excepton: ERROR | Failed to start ActiveMQ JMS Message Broker. Reason: java.io.EOFException:
Chunk stream does not exist at page: 0 | org.apache.activemq.broker.BrokerService | WrapperSimpleAppMain
- and plenty more, I'll attach the log.

On 5.4.0 I found that only after deleting the file db.data was I able to restart the broker
again. On one test run on 5.4.1 I'm finding that deleting db.data is not solving the problem.

I noticed two other users described experiencing exactly the same problem:
http://activemq.2283324.n4.nabble.com/Kaha-corruption-in-5-4-SNAPSHOT-td2366036.html

To replicate the problem, I think it would probably suffice to start a broker with default
config and let 300,000 messages accumulate without being consumed - I'm guessing the problem
should occur before you get there. But perhaps you'll need to create a networkconnector as
I have done and try the same thing. Will attach log...

  was:
I am finding the KahaDB on a broker is becoming corrupted after a certain time of being disconnected
(or perhaps rather, after a certain amount of messages are queued up). My config is extremely
simple - originally 5.4.0 and now 5.4.1 on Windows with the default config, except adding
a connection to a broker on an external machine with
<networkConnectors>
  <networkConnector uri="static:(tcp://machine1:61616,tcp://machine2:61616)"/>
</networkConnectors>

In this case the external machine/consumer is only on during business hours whereas the broker
producing the messages is on 24/7. We're finding that generally the producer instance survives
weeknights, during which it accumulates about 75,000 messages and then offloads them in the
morning to
the consumer instance, but not weekends. Last time it failed after about collecting about
160,000 messages.

I get the excepton: ERROR | Failed to start ActiveMQ JMS Message Broker. Reason: java.io.EOFException:
Chunk stream does not exist at page: 0 | org.apache.activemq.broker.BrokerService | WrapperSimpleAppMain
- and plenty more, I'll attach the log.

On 5.4.0 I found that only after deleting the file db.data was I able to restart the broker
again. On one test run on 5.4.1 I'm finding that deleting db.data is not solving the problem.

I noticed two other users described experiencing exactly the same problem:
http://activemq.2283324.n4.nabble.com/Kaha-corruption-in-5-4-SNAPSHOT-td2366036.htm

To replicate the problem, I think it would probably suffice to start a broker with default
config and let 300,000 messages accumulate without being consumed - I'm guessing the problem
should occur before you get there. But perhaps you'll need to create a networkconnector as
I have done and try the same thing. Will attach log...


> KahaDB corrupted when too many messages accumulate: EOFException
> ----------------------------------------------------------------
>
>                 Key: AMQ-2945
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2945
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Message Store
>    Affects Versions: 5.4.0, 5.4.1
>         Environment: Windows XP, running as service.
>            Reporter: John Fletcher
>         Attachments: 5.4.0.log
>
>
> I am finding the KahaDB on a broker is becoming corrupted after a certain time of being
disconnected (or perhaps rather, after a certain amount of messages are queued up). My config
is extremely simple - originally 5.4.0 and now 5.4.1 on Windows with the default config, except
adding a connection to a broker on an external machine with
> <networkConnectors>
>   <networkConnector uri="static:(tcp://machine1:61616,tcp://machine2:61616)"/>
> </networkConnectors>
> In this case the external machine/consumer is only on during business hours whereas the
broker producing the messages is on 24/7. We're finding that generally the producer instance
survives weeknights, during which it accumulates about 75,000 messages and then offloads them
in the morning to
> the consumer instance, but not weekends. Last time it failed after about collecting about
160,000 messages.
> I get the excepton: ERROR | Failed to start ActiveMQ JMS Message Broker. Reason: java.io.EOFException:
Chunk stream does not exist at page: 0 | org.apache.activemq.broker.BrokerService | WrapperSimpleAppMain
- and plenty more, I'll attach the log.
> On 5.4.0 I found that only after deleting the file db.data was I able to restart the
broker again. On one test run on 5.4.1 I'm finding that deleting db.data is not solving the
problem.
> I noticed two other users described experiencing exactly the same problem:
> http://activemq.2283324.n4.nabble.com/Kaha-corruption-in-5-4-SNAPSHOT-td2366036.html
> To replicate the problem, I think it would probably suffice to start a broker with default
config and let 300,000 messages accumulate without being consumed - I'm guessing the problem
should occur before you get there. But perhaps you'll need to create a networkconnector as
I have done and try the same thing. Will attach log...

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message