Re: Re: Delayed Assignments in multi-server environment?

This is a multi-part message in MIME format...

------------=_1289226387-30467-7833
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Hi,

yes, if you have shark running in many VMs (clustering scenario), caches 
MUST be turned-off, and your problem is definitely caused by the caches 
being turned on.
If you have several VMs running, and load balancing, it should 
compensate the efficiency of the caches, but anyway, have it tested for 
the performance. The clustering is (beyond avoidance of single point of 
failure) used for the scalability of the system, so under the heavy 
load, it should perform better than a single VM with the caches (except 
if you have a really powerful 64 bit machine, with a lot of RAM memory 
to cache everything :-) ).

Greetings,
Sasa.

On 08-Nov-10 15:07, Kofford, C Todd wrote:
> Hi Sasa, thanks for the quick reply and the info regarding the new location of the project(s)&  forum(S).
>
> Hopefully you can answer just one more question pertaing to my situation.
>
> If I have a single shark-based JSP webapp, hitting a single shark DB, deployed to multiple instances of tomcat (spread out over several servers each with it's own JVM) which are all sitting behind a load balancer, could cacheing be causing the this "delay" issue where other users (on different tomcats/servers/JVMs) cannot see new assignments? If so, how much does the cache impact performance if I have to turn it off?
>
> My assignment manager class is basically a clone of the standard assignment manager that comes with shark, with a few modifications to lookup workflow variables in the assignments which dynamically detemine the participant. I don't think that there is anything special in this code that would contribute to a delay in assignments.
>
> In addition, I have not seen this "delay" issue in my TEST or QA environments which each only have a single tomcat/server/JVM.
>
> Thanks again for the help!
>
>
> Todd Kofford
> [email protected]
> University of Kansas - IT
>
> -----Original Message-----
> From: Sasa Bojanic [mailto:[email protected]]
> Sent: Monday, November 08, 2010 4:07 AM
> To: [email protected]
> Subject: [shark] Re: Delayed Assignments in multi-server environment?
>
> Hi Todd,
>
> First of all, to tell everyone from the OW2 Shark's mailing list that we moved Shark and JaWE projects to SourceForge (http://sourceforge.net/projects/sharkwf and http://sourceforge.net/projects/jawe). Soon, there will be 3.3-1 release of Shark/JaWE, and this will be the last release based on XPDL1. The next releases will conform to XPDL2 (at a beginning only the automatic transformation of XPDL1 files to XPDL2 files according to new schema, and in the future we will support more and more things from XPDL2, including BPMN for JaWE).
> There is already a forum where you can post the questions for JaWE on SourceForge, and soon there will be the one for Shark as well, so in the future please post your questions to these forums.
>
> Now, back to your question. If you are using Shark embedded into the single application on the server (like your JSP application), and no other application is using shark on top of the same database, caches should not be the issue. It could be an issue with your implementation of AssignmentManager interface, do you have some special implementation that delays creation of assignments based on some configuration or properties from XPDL?
>
> On the other hand, if you are using more than one application on top of Shark (using the same shark database), then the caches has to be turned-off (this is the case when using both JSP and SwingAdmin application in parallel).
>
> To turn-off the caches, in the configuration files for both applications cache entries should be commented (for CacheManagerClassName entry) or set to zero (for DODS' cache entries).
>
> #CacheManagerClassName=org.enhydra.shark.caching.LRUCacheMgr
>
> DatabaseManager.defaults.cache.maxCacheSize=0
> DatabaseManager.defaults.cache.maxSimpleCacheSize=0
> DatabaseManager.defaults.cache.maxComplexCacheSize=0
> DatabaseManager.defaults.cache.maxMultiJoinCacheSize=0
>
>
> Greetings,
> Sasa.
>
> On 05-Nov-10 20:33, Todd Kofford wrote:
>> I am using Together Workflow Server 3.1-2 in a multi-server
>> environment with a JSP client front end similar to the JSP client
>> example. I really hope that someone can help me with this, because
>> it's a production problem that I need to resolve pretty quickly.
>>
>> I am having users report that sometimes their workflow assignments are
>> being delayed for up to a day. For example user, Joe, creates a new
>> workflow process, accepts the activity and then completes the
>> assignment. In our XPDL workflow, the process is supposed to generate
>> an activity for user Nancy next for approval. However, Nancy doesn't
>> see her new assignment until the next day, even though she tries
>> multiple times to refresh her list of assignments (i.e.,
>> sConn.getResourceObject().get_sequence_work_item(0);)
>>
>> Has anyone seen this before? I also see this when work is done in
>> either the JSP client webapp or admin swing app, that the other
>> application cannot see what was done until I restart the admin app or
>> tomcat for the JSP client webapp. Is this a cache issue?
>>
>> Thanks in advance (if anyone responds)
>>
>> Todd Kofford
>> [email protected]


------------=_1289226387-30467-7833
Content-Type: text/plain; charset="UTF-8"; name="message-footer.txt"
Content-Disposition: inline; filename="message-footer.txt"
Content-Transfer-Encoding: quoted-printable


--
You receive this message as a subscriber of the [email protected] mailing lis=
t.
To unsubscribe: mailto:[email protected]
For general help: mailto:[email protected]?subject=3Dhelp
OW2 mailing lists service home page: http://www.ow2.org/wws

------------=_1289226387-30467-7833--