2014 Taverna tutorial Advanced Taverna

Advanced Taverna
Stian Soiland-Reyes and Christian Brenninkmeijer
University of Manchester
materials by Katy Wolstencroft, Aleksandra Pawlik, Alan Williams
http://orcid.org/0000-0001-9842-9718
Bonn University, 2014-09-01 / 2014-09-03
This work is licensed under a http://www.taverna.org.uk/
Creative Commons Attribution 3.0 Unported License

Advanced Exercises
 The Taverna engine can also help you control the data flow
through your workflows. It allows you to manage iterations
and loops, add your own scripts and tools, and make your
workflows more robust
 The following exercises give you a brief introduction to some
of these features
 Like in the previous tutorial workflows in this practical use
small data-sets and are designed to run in a few minutes. In
the real world, you would be using larger data sets and
workflows would typically run for longer

List handling - cross or dot product
As you may have already seen, Taverna can automatically
iterate over sets of data, calling a service multiple times for
each value in the input list.
When 2 sets of iterated data are combined (one to each input
port), Taverna needs extra information about how they
should be combined. You can have:
A cross product – combining every item from port A with
every item from port B - all against all
A dot product – only combining item 1 from port A with item
1 from port B, item 2 with item 2, and so on – line against
line

List handling – example workflow
 Download and open the workflow “Demonstration of
configurable iteration” from
http://www.myexperiment.org/workflows/4332
 Or see “Run this workflow in Taverna” on myExperiment, and copy the
link into File -> Open Workflow Location
 Read the workflow metadata to find out what the workflow
does (by looking at the ‘Details’)
 Run the workflow and look at the results
 Click on individual services to inspect the intermediate values
and multiple invocations for:
 AnimalsList, ColourAnimals, ShapeAnimals
 Alternatively, add additional workflow output ports from AnimalsList
and ColourAnimals, and rerun.

List handling - configuration
 Go back to the Design view
 Select the ColourAnimals service by clicking on it
 Select the Details tab in the workflow explorer, open List
handling and click on Configure,
 or right-click on ColourAnimals, select Configure running…
then List handling…
 Click on Dot product in the pop-up window. This allows you to
switch to cross product (see the next slide)

List handling– configuring - 1

List handling – configuring - 2
 Click on Dot Product
 Click Change to Cross
Product on the right
 Click OK
 Run the workflow again

List handling - difference
 What is the difference between the results of the two runs?
What does it mean to specify dot or cross product?
NOTE: The iteration strategies are very important. Setting cross
product instead of dot when you have 2000x2000 data items
can cause large and unnecessary increases in computation!

List handling - workflow
e.g. red, green,
blue, yellow
How does Taverna
combine them?
e.g. cat, donkey,
koala

List handling - Cross product
Red
Green
Blue
Yellow
Cat
Donkey
Koala
Red cat, red donkey, red koala
Green cat, green donkey, green koala
Blue cat, blue donkey, blue koala
Yellow cat, yellow donkey, yellow koala

List handling - Dot product
Red
Green
Blue
Yellow
Cat
Donkey
Koala
Red cat
Green donkey
Blue koala
There is no yellow animal because the list lengths don’t match!

List handling - summary
 The default in Taverna is cross product
 Be careful! All against all in large iterations give very big
numbers!
 For more complex list handling, e.g. combination of 3 or more
ports, see
http://dev.mygrid.org.uk/wiki/display/tav250/List+handling

Looping asynchronous services
 Find the workflow “EBI_InterproScan_broken” in the workshop
pack on myExperiment
 InterproScan analyses a given protein sequence (or set of
sequences) for functional motifs and domains
 This workflow is asynchronous. This means that when you
submit data to the ‘runInterproScan’ service, it will return a
jobID and place your job in a queue (this is very useful if your
job will take a long time!)
 The ‘Status’ nested workflow will query your job ID to find out
if it is complete

Looping
The default behaviour in a workflow is to call each service
only once for each item of data – so what if your job has not
finished when ‘Status’ workflow asks?
 Download and run the workflow, using the default protein
sequence and your own email address
 Almost every time, the workflow will fail because the results
are not available before the workflow reaches the
‘get_results’ service – the ‘status’ output is still RUNNING

Looping
 This is where looping is useful. Taverna can keep running the
Status service until it reports that the job is done.
 Go back to the Design view
 Select the Status nested workflow
 Select the Details tab in the workflow explorer, open
Advanced and click on Add looping,
 or right-click on Status, select Configure running… then
Looping…
 (Example on next slide)

Looping
 Use the drop-down boxes in the looping window to set
getStatus_output_status is not equal to RUNNING

Looping
 Save the workflow and run it again
 This time, the workflow will run until the ‘Status’ nested
workflow reports that it is either DONE, or it has an ERROR.
 You will see results for text, but you will still get an error for
‘xml’. This is because there is one more configuration to change
– we also need Control Links to delay the exectution of
getXmlResult.

Control Links
 Normally a service in a workflow will run as
soon as all its input ports are available –
even if graphically it may be “further down”
 A control link specifies that there is a
dependency on another service even if there
is no direct or indirect data flowing between
them.
 In a way the data still flows, but internally on the
called service, outside the workflow
 A control link is shown as a line with a white
circle at the end. In our workflow this means
that getTextResult will not run until the
Status nested workflow is finished

Control Links
 We will add control links to fix the ‘xml’ output
 Switch to the Design view
 Right-click on getXmlResult and select Run after from the drop
down menu.
 getXmlResults is moved down in the diagram, showing the
new control link
 Set it to Run after -> Status
 Save and run the workflow
 Now you will see that getXmlResults and getTextResults take a
bit longer before they run
 This time, results are available for both xml and text

Retries: Making your Workflow Robust
 Web services can sometimes fail due to network connectivity
 If you are iterating over lots of data items, this is more likely to
cause problems because Taverna will be making lots of
network connections.
 You can guard against these temporary interruptions by adding
retries to your workflow
 As an example, we’ll use two local services to emulate
iteration and occasional failures.
 Click a File -> New workflow

 In the Service panel,
 Select the service
Create Lots Of Strings under Available Services -> Local
services -> test
 Add it to the workflow by dragging
it into the workflow diagram
 Also add Sometimes Fails

 Add an output port and connect the service as on the picture
below
 Run the workflow as it is and count the number of failed
iterations. (Tip: Change view values to view errors)
 Run the workflow again. Is the number the same?
 Inspect the intermediate values at Sometimes_fails.

 Now, select the Sometimes_Fails service and select the Details
tab in the workflow explorer panel
 Click on Advanced and Configure for Retry
 In the pop-up box, change it so that it retries each service
iteration 2 times
 Run the workflow again – how many failures do you get this
time? Did you notice the slow down due to retries?
 Change the workflow to retry 5 times – does it work every
time now?

 In network communication, a common strategy for handling
errors is to incrementally wait longer and longer before a
retry – improving chance of recovery.
 In Taverna Retries this can be set by modifying “Delay
increase factor” and “Maximum delay2.
 The settings on the right would retry
after delays of:
1. 1.0 s
2. 1.5 s (1.0 s * 1.5)
3. 2.3 s (1.5 s * 1.5)
4. 3.4 s (2.3 s * 1.5)
5. 5.0 s (3.4 s * 1.5 = 5.1s) – above max 5.0 s

Parallel Service Invocation
 If Taverna is iterating over lots of independent input data, you
can often improve the efficiency of the workflow by running
those iterated jobs in parallel
 Run the Retry workflow again and time how long it takes
 Go back to the Design window, right-click on the
‘sometimes_fails’ service, and select ‘configure running’
 This time select ‘Parallel jobs’ and change the maximum
number to 20
 Run the workflow again
 Does it run faster?

Parallel Service Invocation :
Use with Caution
 Setting parallel jobs usually makes your workflows run faster
(at a cost of more memory/cpu usage)
 Be careful if you are using remote services. Sometimes they have
policies for the number of concurrent jobs individuals should run (e.g.
The EBI ask that you do not submit more than 25 at once).
 If you exceed the limits, your service invocations may be blocked by the
provider. In extreme cases, the provider may block your whole
institution!
 Some remote services don’t handle parallel calls well, as it could cause
concurrency issues server side – e.g. overwriting internal files.
 A good number of concurrent jobs can be anything between 3
and 20 – trial and error is as important as checking the service
documentation.

2014 Taverna tutorial Advanced Taverna

Recommended

More Related Content

Similar to 2014 Taverna tutorial Advanced Taverna (20)

More from myGrid team (14)

Recently uploaded (20)

2014 Taverna tutorial Advanced Taverna