oodt-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Mattmann <chris.mattm...@gmail.com>
Subject Re: workflow control question
Date Wed, 23 Sep 2015 15:33:36 GMT
Hey Val,

I will try and answer this today.


Chris Mattmann

-----Original Message-----
From: "Mallder, Valerie" <Valerie.Mallder@jhuapl.edu>
Reply-To: <dev@oodt.apache.org>
Date: Wednesday, September 23, 2015 at 8:15 AM
To: "dev@oodt.apache.org" <dev@oodt.apache.org>
Subject: RE: workflow control question

>Hi Bruce,
>Thank you for your well thought out response. Now I feel bad because my
>question is not nearly as big as your answer! :( I should have been a
>little bit more clear with my question. I am looking for a way within
>OODT to prevent a second instance of a workflow from starting before the
>first instance of the same workflow has finished.  I have been looking at
>the examples for using WorkflowConditions to control the operation of a
>workflow, but there are no specific examples that do what I would like to
>do. So, if anyone has an example of doing this kind of thing, please let
>me know. Otherwise I will have to grow my own.  I am currently building a
>custom WorkflowCondition and from within that condition class I will try
>to see if I can query the workflow manager to get information about the
>last running workflow.
>Valerie A. Mallder
>New Horizons Deputy Mission System Engineer
>Johns Hopkins University/Applied Physics Laboratory
>> -----Original Message-----
>> From: Bruce Barkstrom [mailto:brbarkstrom@gmail.com]
>> Sent: Tuesday, September 22, 2015 8:29 AM
>> To: dev@oodt.apache.org
>> Subject: Re: workflow control question
>> The usual approach to this kind of problem is to use techniques from
>> programming that involve scheduling.  I'm most familiar with Ada, where
>>there's a
>> long history of work in this area.
>> A classic text is
>> Klein, M. H., et al., 1993: A Practitioner's Handbook for Real-Time
>> Analysis: Guide to Rate Monotonic Analysis for Real-Time Systems,
>> Boston, MA
>> These scheduling problems are usually divided into soft problems, where
>> consequences of missing the schedule are not catastrophic and hard
>> where missing the schedule causes a system failure that is capable of
>> people.  The analysis in this reference suggests that there are two
>>kinds of
>> approaches to scheduling that can be guaranteed to work: Rate Monotonic
>> Earliest Deadline First (EDF) which allows you to take up about 70% of
>> production capacity and scheduling with homogeneous processes which
>> you to move to nearly 100% of capacity.
>> You can think of these as the difference between the traffic flow of an
>> highway and a railroad.  In the former, each car has some average
>> between itself and the other vehicles, but the car can move around
>>within that
>> average distance.
>> In the latter, the distance between cars is pretty close to fixed.
>> Two more recent works are
>> Burns, A. and Wellings, A., 2007: Concurrent and Real-Time Programming
>>in Ada,
>> Cambridge Univ. Press, Cambridge, UK
>> and
>> McCormick, J. W., Singhoff, F., and Hugues, J., 2011: Building Parallel,
>> Embedded, and Real-Time Applications with Ada, Cambridge Univ. Press,
>> Cambridge, UK
>> Both of these works cover various approaches to building a production
>> environment.  The concerns include deadlock, resource starvation, and
>> component failures.  In cases where the system uses priorities to help
>>derive the
>> schedule, you can also have priority inversion.
>> The scheduling problem has a pretty large literature since it shows up
>>not just in
>> the IT environment, but also in any organization that has to deal with
>> scarce resources.  You might also want to take a look at the work by
>> Lamport:
>> <http://research.microsoft.com/en-us/um/people/lamport/pubs/pubs.html>
>> Lamport has an analysis tool known as TLA+ that has been used for formal
>> analysis of scheduling requirements.  This tool is available online.
>> You can go to the TLA Home Page
>> <http://research.microsoft.com/en-us/um/people/lamport/tla/tla.html>
>> and download it from there.
>> As you might expect, this kind of problem is not trivial - and even
>> people make design mistakes.
>> I don't have an easy solution to suggest to you.  To do this kind of
>>work properly,
>> you'll need to conduct an analysis based on the environment you'll be
>>working in.
>> Also, as Lamport explains, you have to worry about the basic scheduling
>>issues -
>> and then you need to deal with scheduling in the presence of unreliable
>> components.  The difference between professional scheduling analysis
>>and simple
>> analysis is in whether the consequences of failure can kill people or
>>just simply
>> manually restarting the system and then figuring out what got corrupted.
>> Bruce B
>> On Mon, Sep 21, 2015 at 5:27 PM, Mallder, Valerie <
>> wrote:
>> > Hi All,
>> >
>> > What is the easiest way to prevent an improper start of workflow?
>> >
>> > I have a cron job that sends an event (i.e. once an hour) to my
>> > workflow manager telling workflow manager to start a workflow. But,
>> > the workflow could take a long time to run depending on how many files
>> > are available to be processed at that time. If the workflow takes
>> > longer than an hour to complete, the cron job is going to send another
>> > event to workflow manager telling it to start the workflow again. But
>> > I don't want it to start the workflow again if the previous workflow
>> > hasn't completed yet. It's perfectly OK for workflow manager to ignore
>> > that second request to start the workflow again and just wait for the
>> > next event to be sent by the cron job.
>> >
>> > I don't want to reinvent the wheel. Has anyone already done something
>> > this?  I've looked into the workflow preconditions, and I created a
>> > WorkflowStatusCondition class to use as a precondition. But, I can't
>> > tell if it is possible to check the status of the first workflow
>> > instance from within a WorkflowConditionInstance object in a second
>> instance.
>> >
>> > Does anyone know how I would do that?
>> >
>> > Val
>> >
>> >
>> > Valerie A. Mallder
>> >
>> > New Horizons Deputy Mission System Engineer The Johns Hopkins
>> > University/Applied Physics Laboratory
>> > 11100 Johns Hopkins Rd (MS 23-282), Laurel, MD 20723
>> > 240-228-7846 (Office) 410-504-2233 (Blackberry)
>> >
>> >

View raw message