activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gilles Harloux <gilles.harl...@gmail.com>
Subject Re: KahaDB: No messages, but log files not reclaimed and StorePercentUsage above 100
Date Tue, 23 Oct 2012 15:53:45 GMT
filed as https://issues.apache.org/jira/browse/AMQ-4128. I agree that
AMQ-3866 doesn't say a word about mKahaDB. It just caught my eye
because of the symtoms, and the logs look alike; Then, I saw you asked
if there was XA involved, hence my reaction: I am not using XA
explicitly, my best guess is that its use comes from mKahaDB /
cross-store transactions.

Anyway, thanks for your help!

On Tue, Oct 23, 2012 at 4:13 PM, Gary Tully <gary.tully@gmail.com> wrote:
> ok, maybe open a  new issue, and attach your logs and configure a data
> small journal data file so that it can easily be attached once you
> find the block in gc.
>
> If recovery of local xa multi store transactions is the problem, I
> would expect some log information from:
> org.apache.activemq.store.kahadb.MultiKahaDBTransactionStore
>
> use setJournalMaxFileLength=24k (it needs to be > 8k) on the
> persistence adapter.
>
> It may require killing the broker in the debugger to get to the bottom
> of this or using a system property that can kill the broker to force
> recovery in this case.
>
> An issue/jira ticket will put it on the radar in any event. I don't
> see that AMQ-3866 is using mkahadb.
>
> On 23 October 2012 14:47, Gilles Harloux <gilles.harloux@gmail.com> wrote:
>> I doubt I can reproduce it as a unit test: What I see happens
>> (sometimes, sometimes not) because I sigkill the process. If I replace
>> the kill with an orderly shutdown, everything works out ok.
>>
>> On Tue, Oct 23, 2012 at 3:38 PM, Gary Tully <gary.tully@gmail.com> wrote:
>>> do you think you could come up with a unit test that shows growth. So
>>> something that uses mkahadb and produces or consumed in a transaction
>>> from destinations that span the stores.
>>>
>>> You could configure the size of journal data files to 16k or
>>> something, such that any issue shows up very quickly.
>>>
>>> A test case like that would help clarify the use case and if it can
>>> reproduce, would be a great start to getting a quick resolution.
>>>
>>> There is a test that does transacted send received across stores, but
>>> does not check for cleanup. It may be easy to extend it in that way.
>>>
>>> in activemq-core
>>> org.apache.activemq.store.StorePerDestinationTest#testTransactedSendReceiveAcrossStores
>>>
>>>
>>> On 23 October 2012 14:20, Gilles Harloux <gilles.harloux@gmail.com> wrote:
>>>> I found that link while you were answering, it seems <g>. I also found
>>>> https://issues.apache.org/jira/browse/AMQ-3866 and it looks a lot like
>>>> what I experience.
>>>>
>>>> I tried to delete the index, start a broker and debug as explained in
>>>> that ticket. Breakpointing at MessageDatabase#process, line 904
>>>> (version 5.7.0) tells me it seems there is XA in the mix (data is a
>>>> KahaAddMessageCommand instance, with
>>>> data.f_transactionInfo.f_localTransactionId == null and
>>>> data.f_transactionInfo.f_xaTransactionId != null). I guess it's due to
>>>> the use of mKahaDB, where I am transacting across both stores.
>>>>
>>>> In that ticket, you asked for the relevant data file. I can share mine
>>>> if there's any interest -- I just have to find a convenient way to
>>>> provide you with a 32 Mb file.
>>>>
>>>> On Tue, Oct 23, 2012 at 12:24 PM, Gary Tully <gary.tully@gmail.com>
wrote:
>>>>> have a read of http://activemq.apache.org/why-do-kahadb-log-files-remain-after-cleanup.html
>>>>>
>>>>> The logging referenced there will show you what destinations are
>>>>> holding on to references to the journal data files.
>>>>>
>>>>> w.r.t the usage %, the journal size increases in chunks of data file
>>>>> size, so a new journal data file can push the usage over the limit.
>>>>>
>>>>> On 23 October 2012 09:15, Gilles Harloux <gilles.harloux@gmail.com>
wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I have a system embedding a broker with KahaDB as a store. I am trying
>>>>>> to get a feel about disaster recovery behavior. So what I am basically
>>>>>> doing is randomly kill & restart the process. I see a condition
(after
>>>>>> killing & restarting multiple times) where the messages get consumed,
>>>>>> but kahaDB journal files (db-*.log) don't get reclaimed. As I set
up a
>>>>>> storeUsage limit, it ends up blocking the system.
>>>>>>
>>>>>> I say the messages get consumed, because I can see my application
>>>>>> logging what messages it gets to handle; Also, JMX tells me that
the
>>>>>> broker's TotalMesageCount is zero, while the StorePercentUsage is
>>>>>> above 100 (depending on parameters such as message size & rate,
I saw
>>>>>> anything from 112 to 293 percent usage).
>>>>>>
>>>>>> So, two questions:
>>>>>>  - How is it possible (for any reasons) to get a usage percentage
>>>>>> above 100? (In other situations, I saw it happen with memory too).
>>>>>>  - Why is it I can't get kahaDB to reclaim seemingly unused journal
files?
>>>>>>
>>>>>> TIA,
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> http://redhat.com
>>>>> http://blog.garytully.com
>>>
>>>
>>>
>>> --
>>> http://redhat.com
>>> http://blog.garytully.com
>
>
>
> --
> http://redhat.com
> http://blog.garytully.com

Mime
View raw message