Miguel, I've created new jira to track this:
On Wed, Mar 9, 2016 at 12:59 AM, Miguel Alvarez <[email protected]> wrote: > Thanks! That helps. > > -----Original Message----- > From: Eddie Epstein [mailto:[email protected]] > Sent: March 7, 2016 6:53 > To: [email protected] > Subject: Re: UIMA-AS: How to configure processing timeouts? > > Miguel, > > A number of possibilities. > > If the UIMA-AS process is scaling vertically across multiple threads, and > the long document processing is not invalidating the computation internals, > then periodically test document processing time in the application logic > and stop processing the document by throwing an > AnalysisEngineProcessException. > > If there are multiple UIMA-AS services, consider running batch jobs with > DUCC. On a processing timeout DUCC will automatically kill and restart the > affected JobProcess (which in general contains multiple pipeline instances, > each running in a different thread), and resubmit other documents that were > co-resident in the JP. > > If the processing timeout and/or OOM conditions are being caused by large > document, use a CasMultiplier to split up large documents and run them in > pieces and use another CasMultiplier to reassemble the pieces. Core UIMA > has some sample code that does this. > > In general a JVM does not survive an OOM unharmed. > > Eddie > > > On Sun, Mar 6, 2016 at 7:22 PM, Miguel Alvarez <[email protected]> > wrote: > > > Hi Eddie, > > > > Thanks for the prompt reply, and the clarification. Now I know : ) > > > > At the moment I am just exploring options, and I am not sure the best > > option for me is to terminate the UIMA-AS service. Sometimes the > > processing of a document takes very long (due to the logic the engines > > have), and I would like to skip the processing of those documents > > altogether but without having to terminate the service (basically just > > cancel the processing of that document but maintain the service > > running so it can process the next one). I am not sure if UIMA-AS > supports anything like this. > > > > And on a similar note, in very rare occasions our engines will run out > > of memory. Is there a way for an UIMA-AS service to recover from an OOM > error? > > > > Thanks again for your help. > > > > Sincerely, > > Miguel > > > > -----Original Message----- > > From: Eddie Epstein [mailto:[email protected]] > > Sent: March 6, 2016 13:57 > > To: [email protected] > > Subject: Re: UIMA-AS: How to configure processing timeouts? > > > > Hi, > > > > I think UIMA-AS documentation is missing important information on the > > processCas timeout, which is that this timeout only works for remote > > delegates. Arbitrary in-process code cannot be interrupted without > > destroying the JVM. Since the input CAS sent to remote delegates is > > saved locally, the error handling allows that delegate to be skipped > > and processing on the CAS to continue. > > > > Would it be useful for you in this situation, timeout for an > > in-process delegate, to simply have the UIMA-AS service terminate? > > > > Eddie > > > > > > On Sun, Mar 6, 2016 at 4:36 PM, Miguel Alvarez <[email protected]> > > wrote: > > > > > Hi, > > > > > > > > > > > > When using UIMA-AS I would like to have a timeout that would stop > > > the processing of a CAS document in case it is taking too long. When > > > I read the documentation it seems like this should be possible, but > > > for some reason it doesn't seem to be working for me, probably > > > because I am configuring something wrong. I am using version 2.6.0 > > > and Java 7 > > > > > > > > > > > > The UIMA-AS deployment descriptor contains a top level engine that > > > is an aggregate engine descriptor which contains only one primitive > > > AE using a "fixed flow". Find below the descriptor I am using (in > > > this case the processing of a document should timeout after 1 > > > second, right?), and I have tried multiple combinations of the > > > settings, but for some reason I am not able to make the processing of > a CAS timeout. > > > > > > > > > > > > I have also tried the timeout on the client side, and that one works > > fine. > > > > > > > > > > > > What am I doing wrong? Does the delegate need to be a remote service > > > for these timeouts to work? Does the delegate need to be pointing to > > > another aggregate engine that wraps the primitive engine? > > > > > > > > > > > > Thanks, > > > > > > Miguel > > > > > > > > > > > > <?xml version="1.0" > > > encoding="UTF-8"?><analysisEngineDeploymentDescription > > > xmlns="http://uima.apache.org/resourceSpecifier";> > > > > > > <name>UIMA-AS MyEngine</name> > > > > > > <description>Processes the document.</description> > > > > > > <version/> > > > > > > <vendor/> > > > > > > <deployment protocol="jms" provider="activemq"> > > > > > > <casPool numberOfCASes="1" initialFsHeapSize="2000000"/> > > > > > > <service> > > > > > > <inputQueue endpoint="MyService" brokerURL="${defaultBrokerURL}" > > > prefetch="1"/> > > > > > > <topDescriptor> > > > > > > <import location="MyAggregate.xml"/> > > > > > > </topDescriptor> > > > > > > <analysisEngine async="true"> > > > > > > <delegates> > > > > > > <analysisEngine key="MyPrimitiveEngine" async="false"> > > > > > > <scaleout numberOfInstances="2"/> > > > > > > <casMultiplier poolSize="1" initialFsHeapSize="2000000" > > > processParentLast="false"/> > > > > > > <asyncAggregateErrorConfiguration> > > > > > > <getMetadataErrors maxRetries="0" timeout="0" > > > errorAction="terminate"/> > > > > > > <processCasErrors maxRetries="5" timeout="1000" > > > continueOnRetryFailure="false" thresholdCount="10" > > thresholdWindow="10000" > > > thresholdAction="terminate"/> > > > > > > <collectionProcessCompleteErrors timeout="0" > > > additionalErrorAction="terminate"/> > > > > > > </asyncAggregateErrorConfiguration> > > > > > > </analysisEngine> > > > > > > </delegates> > > > > > > <asyncPrimitiveErrorConfiguration> > > > > > > <processCasErrors thresholdCount="0" thresholdWindow="0" > > > thresholdAction="terminate"/> > > > > > > <collectionProcessCompleteErrors timeout="0" > > > additionalErrorAction="terminate"/> > > > > > > </asyncPrimitiveErrorConfiguration> > > > > > > </analysisEngine> > > > > > > </service> > > > > > > </deployment> > > > > > > </analysisEngineDeploymentDescription> > > > > > > > > > > > > > > > > > >
