[Yanel-dev] problem re-starting Yanel containing a realm with search lock

[Josias Thöny]

Michael Wechner wrote:
> Josias Thöny wrote:
[...]
> 
>> That's why I changed the behaviour to have only one IndexWriter which 
>> is closed when the repository is shut down.
> 
> 
> where exactly did you change this? Within the configure/init and 
> shutdown methods of Yarep? Is this Yarep repo implementation specific?

It's specific to the VFS implementation, and the changes are in r36400 
and in r36520, concerning these files:
 
public/yarep/trunk/src/impl/java/org/wyona/yarep/impl/repo/vfs/VirtualFileSystemNode.java
public/yarep/trunk/src/impl/java/org/wyona/yarep/impl/repo/vfs/VirtualFileSystemOutputStream.java
public/yarep/trunk/src/impl/java/org/wyona/yarep/impl/repo/vfs/VirtualFileSystemRepository.java

> 
>> Now I realize that this doesn't work in a cluster. Doh!
> 
> 
> right, but it's very important that this works within a clustered 
> environment
> 
>> Maybe there should be a special clustered yarep implementation which 
>> performs some synchronization. I think it should not be required for 
>> all yarep implementations to be clusterable, because it can add a lot 
>> of complexity
> 
> 
> maybe we can hide the complexity
> 
>> and adds a performance penalty.
> 
> 
> You mean because initializing an IndexWriter is bad for performance? I 

I guess so, although I didn't do any profiling.


> would argue the cluster improves the peformance in the end/long run, so 
> that being able to cluster is more important than having a performance 
> gain for a single node.
> 
> Well, how can we fix this? At the moment we have the problem, that all 
> our clustered solutions do not work anymore 

I think it concerns the clustered repos which use vfs and use indexing.

> and I would like to fix this 
> as quick as possible. At the same time you are saying that one receives 
> a LockObtainFailedExceptions when multiple threads are trying to write. 
> I am sure there is some way that we can fix both problems.

I had a closer look at the lucene implementation. It seems that when 
it's trying to obtain the write lock, it actually tries multiple times 
until a timeout occurs. The default timeout is 1 second.
Maybe we can revert to the old way, increase the timeout, and 
catch&ignore the LockObtainFailedException.
I hope this should fix the problems, or WDYT?

josias

> 
> Any idea?
> 
> Thanks
> 
> Michi
> 
>>
>> Josias
>>
>> [1] 
>> http://lucene.apache.org/java/2_2_0/api/org/apache/lucene/index/IndexWriter.html 
>>
>>
>>>
>>>> Maybe with a finalize() method of the IdentityManager/PolicyManager?
>>>>
>>>> BTW, why is indexing enabled for the res-config repository of the 
>>>> yanel-website realm?
>>>
>>>
>>>
>>> no, it would make sense to disable this.
>>>
>>> Cheers
>>>
>>> Michi
>>>
>>>>
>>>> josias
>>>>
>>>>>
>>>>> Also maybe 
>>>>> http://www.manning-sandbox.com/message.jspa?messageID=39279 helps
>>>>>
>>>>> Thanks
>>>>>
>>>>> Michi
>>>>>
>>>>
>>>> _______________________________________________
>>>> Yanel-development mailing list
>>>> Yanel-development at wyona.com
>>>> http://lists.wyona.org/cgi-bin/mailman/listinfo/yanel-development
>>>
>>>
>>>
>>>
>>
>> _______________________________________________
>> Yanel-development mailing list
>> Yanel-development at wyona.com
>> http://lists.wyona.org/cgi-bin/mailman/listinfo/yanel-development
> 
> 
>