uima-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Burn Lewis <burnle...@gmail.com>
Subject Re: DUCC - Work Item Queue Time Management
Date Thu, 24 Sep 2015 12:58:48 GMT
For DUCC 2.x the jobclasspath.properties file defines the JD & JP
classpaths for JUST the ducc code in the JD & JP.  The user code in the JD
(your collection reader) and in the JP (your annotator pipeline) uses ONLY
the classpath you provide plus one ducc jar.

Adding UIMA and application jars to the jobclasspath.properties file should
not help your user code (it does help in 1.x which uses a combined
ducc+user classpath.)

So the major change for DUCC 2.0 is that you must specify a complete
classpath for your application.

~Burn

On Thu, Sep 24, 2015 at 7:59 AM, Lou DeGenaro <lou.degenaro@gmail.com>
wrote:

> Reshu,
>
> Absent some extraordinary circumstance, you should not be touching
> jobclasspath.properties file.
>
> Specify your classpath requirement using --classpath when you submit your
> job or register your service.  This is where you'd add UIMA jars, for
> example.
>
> Lou.
>
> On Tue, Sep 22, 2015 at 12:38 AM, reshu.agarwal <reshu.agarwal@orkash.com>
> wrote:
>
> > Hi,
> >
> > Thanks for replying. I have downloaded the latest code from github and
> > build it. Now, the problem of "Missing the -Dducc.deploy.JdURL property"
> > was resolved.
> >
> > *Lewis:* I used the provided one resources/jobdriver.classpath's file of
> > DUCC 2.0.0 and have to do the same in DUCC 2.1.0.'s
> > resources/jobclasspath.properties file.
> >
> > I added the required lib folder for my job to this file. Now the file  be
> > like:
> >
> > ducc.jobdriver.classpath = \
> >   ${DUCC_HOME}/lib/uima-ducc/*:\
> >   ${DUCC_HOME}/apache-uima/lib/uima-core.jar:\
> >   ${DUCC_HOME}/lib/apache-log4j/*:\
> >   ${DUCC_HOME}/webserver/lib/*:\
> >   ${DUCC_HOME}/lib/http-client/*:\
> >   ${DUCC_HOME}/apache-uima/apache-activemq/lib/*:\
> >   ${DUCC_HOME}/lib/apache-camel/*:\
> >   ${DUCC_HOME}/lib/apache-commons/*:\
> >   ${DUCC_HOME}/lib/google-gson/*:\
> >   ${DUCC_HOME}/lib/springframework/*:/
> > ___/home/ducc/Uima_pipeline/lib/*<---------(I changed here for my job.)_
> >
> > ducc.jobprocess.classpath = \
> >   ${DUCC_HOME}/lib/uima-ducc/*:\
> >   ${DUCC_HOME}/apache-uima/lib/uima-core.jar:\
> >   ${DUCC_HOME}/lib/apache-log4j/*:\
> >   ${DUCC_HOME}/webserver/lib/*:\
> >   ${DUCC_HOME}/lib/http-client/*:\
> >   ${DUCC_HOME}/apache-uima/apache-activemq/lib/*:\
> >   ${DUCC_HOME}/apache-uima/apache-activemq/lib/optional/*:\
> >   ${DUCC_HOME}/lib/apache-camel/*:\
> >   ${DUCC_HOME}/lib/apache-commons/*:\
> >   ${DUCC_HOME}/lib/springframework/*
> >
> > This changes works in DUCC 2.1.0 version and my job completed
> > successfully. But this is not a solution as these all jars will add to
> each
> > job even if not necessary. This lib folder contains third party jar as
> well
> > as UIMA and UIMA AS jars.
> >
> >
> > On 09/22/2015 01:56 AM, Burn Lewis wrote:
> >
> >> re your original problem of a missing UIMA class:
> >>
> >> It should not be necessary to modify resources/jobdriver.classpath ...
> >> were
> >> you using the one provided with 2.0 or do you have a locally modified
> one?
> >> Please let us know what changes to the 2.0 one you had to make.
> >>
> >> You should just add the required UIMA jars to the classpath you provide
> >> when you submit the job.  If you provide a deployment descriptor you'll
> >> need to supply all the UIMA-AS jars, e.g.
> >>
> >>
> ${DUCC_HOME}/apache-uima/lib/*:${DUCC_HOME}/apache-uima/apache-activemq/lib/*:${DUCC_HOME}/apache-uima/apache-activemq/lib/optional/*
> >>
> >> otherwise you probably need only 1 jar, e.g.
> >> *${DUCC_HOME}/apache-uima/lib/uima-core.jar*
> >>
> >> Note that in these examples I've used the UIMA jars that are included
> with
> >> DUCC, but in general it would be better if you used your own copy of
> UIMA,
> >> at whatever level is best for your application.
> >>
> >> In DUCC 1.x the DUCC jars and their dependencies were added to the
> user's
> >> classpath, but this often caused problems when DUCC code and user code
> >> used
> >> different versions of a 3rd party jar, so in DUCC 2.0 we use a different
> >> classloader for DUCC & user code, and add only one DUCC jar to the
> user's
> >> classpath.
> >>
> >> ~Burn
> >>
> >>
> >> On Mon, Sep 21, 2015 at 9:18 AM, Jaroslaw Cwiklik <cwiklik@us.ibm.com>
> >> wrote:
> >>
> >> Reshu, if you have maven and svn installed on your machine you can
> >>> checkout the latest code from the svn:
> >>>
> >>> svn co https://svn.apache.org/repos/asf/uima/sandbox/uima-ducc/trunk/
> .
> >>>
> >>> and  build it with: mvn clean install
> >>> You'll get a new ducc tarball in target dir
> >>>
> >>>   Jerry Cwiklik
> >>> IBM Watson RTP North Carolina
> >>> UIMA Extensions
> >>> 4205 S MIAMI BLVD
> >>> DURHAM , NC , 27703-9141
> >>> United States
> >>> Building: 502  |  Floor: 02  |  Office: M210
> >>> Tel: 919-254-6641  TL:444-6641
> >>> Email: cwiklik@us.ibm.com
> >>>
> >>> [image: Inactive hide details for Lou DeGenaro ---09/21/2015 08:44:06
> >>> AM---Reshu, This is a bug in DUCC 2.0.0. See https://issues.apac]Lou
> >>> DeGenaro ---09/21/2015 08:44:06 AM---Reshu, This is a bug in DUCC
> 2.0.0.
> >>> See https://issues.apache.org/jira/browse/UIMA
> >>>
> >>> From: Lou DeGenaro <lou.degenaro@gmail.com>
> >>> To: user@uima.apache.org
> >>> Date: 09/21/2015 08:44 AM
> >>> Subject: Re: DUCC - Work Item Queue Time Management
> >>> ------------------------------
> >>>
> >>>
> >>>
> >>> Reshu,
> >>>
> >>> This is a bug in DUCC 2.0.0.  See
> >>> https://issues.apache.org/jira/browse/UIMA
> >>> -4576?jql=project%20%3D%20UIMA.
> >>>
> >>> Presently, you would need download the current DUCC source and build a
> >>> new
> >>> tarball to get the fix.
> >>>
> >>> In the mean time, I'll investigate how interim DUCC releases (tarballs)
> >>> are
> >>> posted to the Apache website.
> >>>
> >>> Lou.
> >>>
> >>> On Mon, Sep 21, 2015 at 7:25 AM, reshu.agarwal <
> reshu.agarwal@orkash.com
> >>> >
> >>> wrote:
> >>>
> >>> Hi,
> >>>>
> >>>> As you said:    "In DUCC 2.0 you must explicitly supply UIMA in the
> >>>> classpath of your submission. This was not the case in DUCC 1.x where
> >>>>
> >>> UIMA
> >>>
> >>>> was added by DUCC under the covers."
> >>>>
> >>>> I defined the same but still facing the error. In JD initialization,
I
> >>>> defined the java class library path required in classpath parameter
in
> >>>>
> >>> job
> >>>
> >>>> specification. But it was showing error until I added the same in
> >>>> resources/jobdriver.classpath. After this It was initialized and then
> >>>> started showing error of "Missing the -Dducc.deploy.JdURL property".
> >>>>
> >>>> I was getting java.lang.RuntimeException: Missing the
> >>>> -Dducc.deploy.JdURL
> >>>> property even in 1.job. Why this error is coming?
> >>>>
> >>>> Thanks in Advance.
> >>>>
> >>>> Reshu.
> >>>>
> >>>>
> >>>> On 09/18/2015 02:47 PM, Lou DeGenaro wrote:
> >>>>
> >>>> Reshu,
> >>>>>
> >>>>> In DUCC 2.0 you must explicitly supply UIMA in the classpath of
your
> >>>>> submission.  This was not the case in DUCC 1.x where UIMA was added
> by
> >>>>> DUCC
> >>>>> under the covers.
> >>>>>
> >>>>> In fact this gives you more flexibility in that you are no loner
tied
> >>>>> to
> >>>>> using a particular version of UIMA.
> >>>>>
> >>>>> Lou.
> >>>>>
> >>>>> On Fri, Sep 18, 2015 at 12:24 AM, reshu.agarwal <
> >>>>>
> >>>> reshu.agarwal@orkash.com
> >>>
> >>>> wrote:
> >>>>>
> >>>>> Jerry,
> >>>>>
> >>>>>> I have tried DUCC 2.0.0 to run same job on it. I don't know
why but
> >>>>>>
> >>>>> same
> >>>
> >>>> job descriptor didn't work. It showed some exception at initialization
> >>>>>> time
> >>>>>> which was not in case of 1.1.0.
> >>>>>>
> >>>>>> Is there any changes regarding job descriptor or service descriptor?
> >>>>>>
> >>>>> The
> >>>
> >>>> both did not work in my case for DUCC 2.0.0 but for DUCC 1.0.0 and
> DUCC
> >>>>>> 1.1.0.
> >>>>>>
> >>>>>> In Service descriptor it shows some spring Framework's class
not
> found
> >>>>>> exception. See below:
> >>>>>>
> >>>>>> *java.lang.NoClassDefFoundError:
> >>>>>> org/springframework/context/ApplicationListener*
> >>>>>>
> >>>>>> Thanks in advance.
> >>>>>>
> >>>>>> Reshu.
> >>>>>>
> >>>>>>
> >>>>>> On 09/17/2015 08:15 PM, Jaroslaw Cwiklik wrote:
> >>>>>>
> >>>>>> Hi, can you try Ducc 2.0.0? It was recently released into Apache.
> One
> >>>>>>
> >>>>> of
> >>>
> >>>> the key changes was to remove queues as means of transport between JD
> >>>>>>> (Job
> >>>>>>> Driver) and JP (Job Process). Instead, each JP uses HTTP
to
> request a
> >>>>>>> Work
> >>>>>>> Item (CAS) from a JD.
> >>>>>>>
> >>>>>>> DUCC 1.1.0 has a concept of a WI timeout which I think is
24 hours
> by
> >>>>>>> default. A timer is started in a JD when each WI is dispatched
to a
> >>>>>>>
> >>>>>> JP.
> >>>
> >>>> If
> >>>>>>> the WI does not come back for whatever reason, the timer
pops and a
> >>>>>>> JD
> >>>>>>> will
> >>>>>>> attempt to retry that WI.
> >>>>>>>
> >>>>>>> To debug your problem with DUCC 1.1.0 I suggest attaching
JMX
> console
> >>>>>>> to a
> >>>>>>> running JP to see where its threads are. Before doing this,
check
> JP
> >>>>>>> logs
> >>>>>>> to see if there is an exception.
> >>>>>>>
> >>>>>>> Jerry
> >>>>>>>
> >>>>>>>
> >>>>>>>
> >>>>>>> On Thu, Sep 17, 2015 at 4:32 AM, reshu.agarwal <
> >>>>>>> reshu.agarwal@orkash.com>
> >>>>>>> wrote:
> >>>>>>>
> >>>>>>> My DUCC version is 1.1.0.
> >>>>>>>
> >>>>>>> On 09/17/2015 11:35 AM, reshu.agarwal wrote:
> >>>>>>>>
> >>>>>>>> Hi,
> >>>>>>>>
> >>>>>>>> I am facing a problem in DUCC that some documents were
shown in
> >>>>>>>>>
> >>>>>>>> queue
> >>>
> >>>> but
> >>>>>>>>> did not get processed. In Job, work item list shows
a particular
> >>>>>>>>>
> >>>>>>>> work
> >>>
> >>>> item's status "queued" and queueing time is "4115 seconds".
> >>>>>>>>>
> >>>>>>>>> I want to set queueing time of work item not more
then 1 minute.
> >>>>>>>>>
> >>>>>>>> What
> >>>
> >>>> is
> >>>>>>>>> the reason for the same? Is there any method to
solve this? How
> can
> >>>>>>>>>
> >>>>>>>> I
> >>>
> >>>> set
> >>>>>>>>> maximum queueing time for work item?
> >>>>>>>>>
> >>>>>>>>> Thanks in advance.
> >>>>>>>>>
> >>>>>>>>> Reshu.
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message