Locked History Actions

Events/GCC2013/TrainingDay/API

Training Day

The Galaxy API

The Galaxy API is a way of interacting with Galaxy data without using the web-based, user interface.

A more thorough explanation can be found at Learn/API along with some general steps on how to begin using the API. This page may duplicate some of that data in order to keep information relevant to the workshop in one place.

Full documentation for each API function and module is posted at ReadTheDocs.org.

This Wiki page is the outline of a 2013 GCC workshop that attempts to teach the conventions of Galaxy's API while building an example script.


What's an API and why would I care?

An API (Application Programming Interface) is the syntax defined by a program for how it can be controlled by another.

A web (REST or REpresentational State Transfer) API is just an another API where the syntax, instead of function calls and their arguments, is made up of a series of URLs and HTTP requests. The server providing the web API becomes the program and the URLs become the commands - all through the medium of the HTTP protocol.

Imagine if your home had a web API. You might be able to turn on the lights simply by typing myhouse.net/lights/dining-room/on, myhouse.net/fireplace/on?setting=low, etc. Each URL entered would tell this external system (your home) to perform a single action.

You certainly can do these things yourself by manually turning light switches on and lighting fires - but the key advantages to an API are that you can:

  1. Iteratively accomplish tasks: `for every book in the bedroom, return the book to the bookshelf (ok - that probably won't be possible any time soon - but wouldn't it be nice)`
  2. Scheduled or responsive tasks even when you're not around:

    when I'm on vacation, feed the cat everyday at 8am and bring in the mail at 2pm

  3. Compose complex tasks/scripts from simple tasks to remove the tedium of common scenarios:

    wake me up at 6:00am, turn on the lights, and start making some frighteningly strong coffee -> ohgodmorning.py

In more Galaxy specific terms, the web API allows you to use Galaxy's capabilities programmatically: running a workflow over each file in a directory, moving a dataset to a library only if it passes some QA threshold, upload a fastq file when the sequencer finishes writing it, or combine any or all of the above into another script.

Additionally, and perhaps more importantly: even when your scripts run, the features that make Galaxy great are still applied to your data:

It's also worth noting that all the work you did via the API is still accessible and modifiable when you return to the UI.



How to Access and Use the API

Anything that can communicate over HTTP can use the Galaxy web API.

Some programming or scripting languages have their own libraries to do this:

There are several scripts and excellent programming libraries designed to help with the Galaxy API:

Data returned from (and in some cases passed to) the Galaxy API is often described or formatted in JSON (JavaScript Object Notation) - a compact and flexible way of describing relatively complex data structures. Conversion to the language you may be programming in is automatic in the case of the scripts/api directory, BioBlend, and !Blend4j and relatively easy for python, and a non-issue for JavaScript.


The concepts of resources, verbs, and the REST model

A common way to construct an API command is to view it as a simple (imperative) sentence: <verb> <a resource>

From that then, it follows that each web API URL (command) is composed of three parts:

  1. an HTTP method - the verb
  2. a URL path - the resource
  3. and any additional arguments or parameters the API command may need



A Sample Use

We'll slowly build a fairly complex script using the API to:

The final script should not be considered a 'good' script, however, because a good script would take into account:

There are 10 steps total to building the script. Each step (step_1.py, step_2.py, ...) has its own progressively longer version of the final script, adding new sections and new functionality.

As each step introduces new functionality in the API, they will themselves include new, short modules that center around some resource and the API methods that can be applied to them. For example, when writing step_2.py, we'll create several history scripts (histories_1.py, histories_2.py) that each add more to what we can do with histories. step_2.py will then import these smaller modules to combine them into our final script.

Also note: that almost every 'resource' script can function on its own, providing some functionality (such as querying or creating).



The Set Up

The API Key

Your API key is unique identifier that is sent to the API to link you with your Galaxy account and checks your access to data. It is effectively your username and password. Treat it with the same caution and care you would your password.

For the purposes of simplifying some of our scripts, we're going to add our API key as an environment variable in our terminal session.

Keeping any form of unencrypted authentication can be dangerous and that includes your API key. If you use the following technique, place the sourceed environment variable file in a secure location on your local filesystem.

Let's load the API key into our terminal session:

   1 source apiworkshop-setup.sh

And check what environment variables it's added. First, the API key:

   1 echo $GALAXY_API_KEY

then the base url of where our Galaxy installation is served:

   1 echo $GALAXY_BASE_URL

These will both be needed in every call to the API.

Structure

We'll be using python and scripts/api/common.py as a layer on top of urllib2. setup.py will load our key and base URL for every API call.

Much of the functionality of the resource scripts (such as users_1.py, histories_1.py, etc.) is available already in BioBlend or Blend4j - we won't use them here so we can get a closer look on the internals and conventions of the API.



The Steps


1. Checking your key with the user API

We'll start with the scripts: step_1.py and users_1.py.

These are simple scripts that only make a connection, call GET api/users, and return information about us - the current user.

Of course, normally we already know this information but this allows us to test whether our server, url, and key are working properly.

Taking a look inside users_1.py we can see we're simply calling a function in our main clause: get_users. It takes our key (since it's our form of authentication here) and the root URL, builds on that URL with the api/ + resource and then calls common.get with those parameters (which does some of the heavy/tedious lifting here).

All API method calls will follow this pattern: build a URL from a base URL and a resource then pass it to an HTTP method in common.py with the key authentication.

   1 ./users_1.py

Here's the data the users_1.py script should return (if all went well):

   1 [ { u'email': u'apiworkshop@example.com',
   2     u'id': u'1cd8e2f6b131e891',
   3     u'model_class': u'User',
   4     u'quota_percent': None,
   5     u'url': u'/api/users/1cd8e2f6b131e891'}]

(Note: your ID may be different)

Note the syntax of what's returned: a list with one element, the element is in python dictionary form, and the strings are unicode strings. Most resource data returned from the API will be in dictionary form (after having been converted from JSON objects in common.py). It's a list because we used the index API method (we'll cover that in step 2) and there's only one element because we're only allowed to access our own user data. Unicode strings will be a common sight during the workshop for both dictionary keys and values. They're a special string for encoding many human languages and writing systems but for all intents and purposes here they can be thought of as normal strings.

Now - try the first iteration of our main script, step_1.py.

   1 ./step_1.py

You should see the same data returned as users_1.py - we're not doing much yet.


2. Looking at your histories with the history API

Here we'll learn how to query all our histories and query a specific history.

Scripts: step_2.py, histories_1.py, histories_2.py


histories_1.py

histories_1.py allows us to get (summary) data on all our histories using the GET method and the index API method.

Calling histories_1.py from the command line should show us a single, 'Unnamed history' complete with Id, name, and other attributes. Note that, even though there is only one history, the form of the data returned is a list:

   1 [ { u'deleted': False,
   2     u'id': u'c24141d7e4e77705',
   3     u'model_class': u'History',
   4     u'name': u'Unnamed History',
   5     u'published': False,
   6     u'tags': [],
   7     u'url': u'/api/histories/c24141d7e4e77705'}]

histories_1.py is an index API method. These are generally used to return lists of often summary information about a group of resources (e.g. all my histories, or the results of a search query such as 'all histories whose names begin with X').


histories_2.py

histories_2.py allows us to get more detailed info about a specific history also using the GET command.

You must call histories_2.py from the command line and give it a single Id as a parameter - in this case, copy the id attribute from the history returned from histories_1.py (without quotes) and paste it after the histories_2.py script name:

   1 ./histories_2.py 1cd8e2f6b131e891

(Again: the ID may be different - make sure you paste the one from histories_1.py)

In histories_2.get_history, we're building the url and including an ID in it to get data about a specific history. Many resources are stored in the database with an ID and adding them to the url allows us to specify which resource we want to use. Even though were using the same HTTP method/verb as index here (GET), because we've added the id to the path, the API now recognizes we want to do a show: get specific information on one history. The way you build a URL (and what/how you add extra parameters) can make each of the four basic HTTP methods do very different things.

You should get a single python dictionary containing quite a bit of information about your current history. Although you have no datasets in it yet, notice the attributes which would contain the counts and ids of datasets keyed by their potential states.

Note that histories_2.py is the same file as histories_1.py with a new function added in. The resource scripts will all follow this pattern. The 'last' resource script will always contain the functionality of the ones 'before' it, just as each step_N will add new functionality to the overall, final script. (Warning: the effects of calling the resource scripts on the command line will change however from number to number - it will often be a demonstration of the most recently added functionality.)

Again:

Many resources in the Galaxy API will have index and show methods for reading and retrieving information.


Errors!

Now that we have something we can pass an argument to (and therefore break), let's break it! Try entering this:

   1 histories_2.py bler

You should see a stack trace. You should already be familiar with a stack trace but, if you're not, note the two most important pieces of info it provides: an error string (urllib2.HTTPError: HTTP Error 500: Internal Server Error) and a location in a script file. There is difference here from a normal error in a local script: although we made a mistake the stack trace location won't help us much because the relevant code is on the server. We can only use the error string to let us know what happened.

In many places in the Galaxy API 'HTTP' errors and exceptions will be thrown, both with (hopefully) descriptive messages and also with an HTTP status code. Generally when you make an error using the API, Galaxy will (or should) return something in the 400's (meaning user error). In this case, even though it was clearly our fault, Galaxy returned a 500 (Internal Server Error) meaning something went wrong on it's side.

(We're striving to make this more consistent - but when using the API you should be aware of this pitfall)


3. Creating a history with the history API

Here we'll learn how to create a new history which will (automatically) become our 'working' history.

Scripts: step_3.py, histories_3.py


histories_3.py

We've added the function create_history in this version of histories. This will build a URL to api/histories and call that URL with the HTTP method POST (common.post). Note that we don't need to send an ID as this is a new history and Galaxy will assign the ID itself (the new ID will be returned to us so we can use to further manipulate the history later). This is a common pattern with create.

Note also that create_history allows us to immediately name the history by passing in a name argument. The name is passed to common.post bundled in a dictionary. This is also a common pattern with POST (and PUT) and, although these additional arguments are often optional in practice, there are cases were they're required. It's generally best to check the API documentation or the code to find out.

histories_3.py is also set up to create a new history from the command line now but since we're also doing that in step_3.py, we'll just call that:

   1 ./step_3.py

The output:

   1 created history! Step 3
   2 new history:
   3 { u'annotation': u'',
   4   u'contents_url': u'/api/histories/ee3742ed6d8b4a0c/contents',
   5   u'deleted': False,
   6   u'genome_build': None,
   7   u'id': u'ee3742ed6d8b4a0c',
   8   u'model_class': u'History',
   9   u'name': u'Step 3',
  10   ...

(The output has been truncated here with '...')

According to the script we created a new history with a name of 'Step 3'.

We can double check this a couple of different ways - we could refresh Galaxy in our browser and this history should be our current, 'working' history. Or we could call either histories_1.py (which will have a list of our histories that should contain the new one) or step_2.py (which will give details about our current history).

Since each step_N.py script will build on the previous, we'll be creating a new history for each step and each new history will be named 'Step N'. This will allow us to 'start fresh' with each step and not worry about 'polluting' any one history. You can even call ./step_3.py (or any of the step scripts) multiple times without fear of ruining any following steps.


4. Checking the contents of a history with the history contents API

Histories can be thought of as containers for the datasets we produce over time. The python class for those datasets (that are associated with a history and a user) is HistoryDatasetAssociation (HDA).

Here we'll learn how to query the datasets within a history using a contents style API URL. This is very similar in functionality to the right-hand history panel in the Galaxy web UI - in fact that panel uses a version of this API call to build its HTML.

Scripts: step_4.py, hdas_1.py

hdas_1.py will do an index of the hdas if you call it from the command line. Examine hdas_1.get_hdas. In some ways it's very similar to histories.get_history (the show) but we've added /contents to the URL path.

If we want to get the HDAs from a history we need to tell the API two things: 1) the specific history we want the HDAs from and 2) that we're interested in the 'contents' and not the information on the history itself (which we would get from show). This is a common pattern for the API when dealing with containers, we need to specify the containers and tell the API to do something with the contents.

./step_4.py does everything that steps 1 to 3 do and then queries the HDAs in the new history:

   1 ./step_4.py

The output:

   1 created history! Step 4
   2 HDAs:
   3 []

Note the empty list for the HDAs. Since this is an index of the HDAs we know it will return a list, but the emptiness is a bit anti-climatic. This is a new history, however, so it is the correct return value. If it had HDAs to return, that would have been a pretty big bug!


5. Uploading a file to a new HDA with the history contents API

Here we'll start moving a bit faster: we'll upload a file from our 'local' machine into a new history and get some details on the HDA that Galaxy creates for it.

Scripts: step_5.py, tools_1.py, hdas_2.py

The file will be myIlluminaRun.solexa.fastq and contain some solexa fastq reads. As an aside: this file is from the history uploaded by Anton Nekrutenko at Galaxy Main server's published histories: https://main.g2.bx.psu.edu/u/aun1/h/quickie-14. It's the initial data used in the screencast quickie: http://screencast.g2.bx.psu.edu/quickie_13_fastq_basic/flow.html

The tools_1.py contains one function: upload_hda. The Galaxy developers would like to have every tool be available to run through the API but this is still a work in progress. In this case, we use another module/script upload_to_history.py to remove some of the complexity and it itself uses the requests module to handle moving the file into the POST data. Be sure to check out those modules for more details on what's involved.

The takeaway here is that, to run a tool we combine four things:

  1. the POST HTTP method (or create again)

  2. the resource URL api/tools

  3. the ID of the tool to use: upload1

  4. and any parameters the tool needs: the file, a name, some extra options and directives

   1 ./step_5.py

The output:

   1 created history! Step 5
   2 uploaded hda! myIlluminaRun.solexa.fastq
   3 HDAs now:
   4 [ { u'id': u'9d052533bb57bd3e',
   5     u'name': u'myIlluminaRun.solexa.fastq',
   6     u'type': u'file',
   7     u'url': u'/api/histories/9752b387803d3e1e/contents/9d052533bb57bd3e'}]
   8 Uploaded:
   9 { u'accessible': True,
  10   u'api_type': u'file',
  11   u'data_type': u'auto',
  12   u'deleted': False,
  13   u'display_apps': [],
  14   u'display_types': [],
  15   u'download_url': u'/api/histories/9752b387803d3e1e/contents/9d052533bb57bd3e/display',
  16   u'file_ext': u'auto',
  17   u'file_size': 0,
  18   u'genome_build': u'?',
  19   u'hda_ldda': u'hda',
  20   u'hid': 1,
  21   u'history_id': u'9752b387803d3e1e',
  22   u'id': u'9d052533bb57bd3e',
  23   ...

There's our uploaded file in Galaxy.


Asynchronous

Note the state attribute of the uploaded file:

   1 u'state': u'queued',

This means that the job that uploads the HDA isn't actually finished when our script is. In fact, since many things happen asynchronously (not always one-after-another) - Galaxy's infrastructure, the API, and calls to it are designed to return data and continue even if a particular function or command is not 'finished' yet. Essentially, what upload_hda does is tell Galaxy to bring in the file, return some data immediately, and then move on to the next command in the script.

This is an advantage in many situations and a disadvantage in others.

An advantage is that you don't have to wait for Galaxy's jobs to finish before sending more commands: you could call the upload_hda 100 times for 100 files within a minute and all those jobs will queue while you go get a beer. Galaxy will create those 100 jobs in the queue while you do other things.

If you have to use the data from an asynchronous operation like running a tool or a workflow, it becomes (a slight) disadvantage and you need to wait for something to finish. Handling this properly is something we'll cover in step 6.


6. Uploading a file and waiting until the API says it's 'ok'

Here we'll handle the asynchronous nature of Galaxy tool running and jobs in order to be sure the data is uploaded and ready before we start using it.

Scripts: step_6.py

Most of our changes will be 'local' this time in the step_6.py version of our final script. We'll use a while loop and the state information from the HDA show api to wait for the upload to finish.

The psuedo code for this loop might look like this:

Get the state of the HDA from the API
While the state isn't 'ok':
  Wait for four seconds
  Check the state again using the API

We'll also output some of that information while our script is running. This is always a good idea especially during development so you know what's going on.

   1 ./step_6.py

The output:

   1 created history! Step 6
   2 uploaded hda! myIlluminaRun.solexa.fastq
   3          uploaded_hda_state: queued
   4          (waiting 4 seconds...)
   5 Uploaded:
   6 { u'accessible': True,
   7   u'api_type': u'file',
   8   u'data_type': u'fastq',
   9   ...

(Note: you may see more waiting when you run the script)

Now we know that when the loop completes in our script the dataset is finished uploading and any operations we do after the loop will have access to the dataset's data.


7. Running a workflow on an uploaded HDA using the workflows API

Now we'll run a workflow on our uploaded file and again wait for all of it to finish. We'll be moving faster here.

Scripts: step_7.py, workflows_1.py


Workflows

Workflows are complicated structures. For this workshop, the needed workflow here is already loaded and available to the workshop user (but it is also available at TODO). All we need to do is 'invoke' it from the api on the uploaded HDA. This is another example of POST + resource URL + parameters to create. It's also an example of composing complex scripts - this time using the server side instead of the client.

Note: there are some rough ways to change the parameters used in workflows dynamically through the API. Be sure to check out scripts/api/workflow_execute and scripts/api/workflow_execute_parameters for examples.

   1 ./step_7.py

The output:

   1 created history! Step 7
   2 uploaded hda! myIlluminaRun.solexa.fastq
   3      uploaded_hda_state: queued
   4      (waiting 4 seconds...)
   5 found workflow! Joined Solexa QC
   6 running Joined Solexa QC workflow...
   7 workflow started!
   8 groomed.fastq
   9      state: running
  10      (waiting 4 seconds...)
  11      ok
  12 filtered at > 17
  13      state: running
  14      (waiting 4 seconds...)
  15      state: running
  16      (waiting 4 seconds...)
  17      state: running
  18      (waiting 4 seconds...)
  19      state: running
  20      (waiting 4 seconds...)
  21      ok
  22 statistics
  23      state: running
  24      (waiting 4 seconds...)
  25      ok
  26 forward reads
  27      ok
  28 reverse reads
  29      ok
  30 statistics plot
  31      ok
  32 workflow complete!
  33 Output:
  34 { u'history': u'df7a1f0c02a5b08e',
  35   u'outputs': [ u'f597429621d6eb2b',
  36                 u'1cd8e2f6b131e891',
  37                 u'ebfb8f50c6abde6d',
  38                 u'33b43b4e7093c91f',
  39                 u'a799d38679e985db',
  40                 u'5969b1f7201f12ae']}

Note the final output: these are the HDAs created by the workflow. (We can also list them using hdas_2.py.)


8. Checking the data of an HDA using the datasets API

Now we'll introduce the possibility of a conditional in our API script: specifically, we'll look inside the dataset data of one of the HDAs created by our workflow. This will give us an opportunity to do something different depending on what we find.

Scripts: step_8.py, datasets_3.py


The datasets API

It's all fine and good to operate on the information from resources, but if we consider the actual 'currency' of Galaxy to be the data inside datasets we'll need to get at that information as well.

Note that we've skipped talking about datasets_1 and datasets_2. These would be the index and show methods for the datasets resource. In many ways, the datasets resource is a simplification of the HDA resource (and it's sibling the LibraryDatasetDatasetAssociation or LDDA resource - which we'll cover a bit). Datasets is an interface to both HDAs and LDDAs that often (effectively) 'strip' or ignore the information about the datasets' containers.

Let's try datasets_1.py, tho, to see something important:

   1 ./datasets_1.py

The output:

   1 Traceback (most recent call last):
   2 ...
   3 urllib2.HTTPError: HTTP Error 501: Not Implemented

(Again, the '...' was added to simplify the output)

HTTP Error 501: Not Implemented is a server error (with status code 501) meaning this functionality hasn't been added yet (and may never be).

In this case, you're seeing it because the developers don't have an implementation of index for datasets (what might that be? A list of all datasets for the current user? All published datasets? An interface to a dataset search?) In any event, that functionality isn't available throught the API at this time. Keep this error in mind when exploring the API.

datasets_2.py in many respects mimics the functionality of hdas_2.show and returns simple data and metadata for the dataset.

However, the same API method from datasets_2 (`GET + resource_url + ID), when passed some optional parameters allows us to get data from within the dataset. This is done through the Galaxy construct of data providers.


Data providers

Data providers allow us to get specific data from within a dataset's file contents. In this case (datasets_3.get_column_data), we're using the ColumnDataProvider which provides columns from the Tabular dataset statistics.

Data provider are also a work in progress and currently only used in visualizations, but work is underway to make them easier to use and more powerful in general.

See the datasets_3.get_dataset_column for an example of how to use the datasets API to get raw data from a file.

Let's now run step 8:

   1 ./step_8.py

The output:

   1 uploaded hda! myIlluminaRun.solexa.fastq
   2      uploaded_hda_state: queued
   3      (waiting 4 seconds...)
   4 found workflow! Joined Solexa QC
   5 running Joined Solexa QC workflow...
   6 workflow started!
   7 ...
   8 workflow complete!
   9 get_dataset_column, full_url: http://localhost:8081/api/datasets/3f5830403180d620?key=a63a28c5d5575a8ca78e97e01a73f901&data_type=raw_data&provider=column&columns=[5]
  10 mean: 38.8721137988 median: 19.776113249

So now we've gotten some quality information about our workflow run and fastq reads. As step_8.py mentions, we could use that information to conditionally run a seperate script, provide some warning/email to the user, etc.

(Also note the print out of the get_dataset_column full url: see step_8.py and datasets_3.py for an explanation of why but, in the simplest terms, its an illustration of how some API calls transmit extra parameters.)


9. Renaming and annotating an HDA using the history contents API

We'll now introduce a third HTTP method: PUT (DELETE won't be covered in this workshop but know that it's available for many resources) and use it to: update (change) the name of our final, desired fastq HDAs and annotate them with the quality information from step 8.

Scripts: step_9.py, hdas_3.py

PUT/update methods are meant to alter existing resources - much like the edit attributes page of the Galaxy UI. Their URLs are constructed very similarly to any GET method where we specify resources, containers, and contained objects by id. We pass the data we want to change in the form of a dictionary and that gets packaged into the HTTP request much like POST data does.

hdas_3.py adds the function update_hda which allows us to change the state and information contained in the HDA (but not the file contents). We can change the name, annotation, misc_info (the small amount of text that describes what the HDA's file contains), the genome build (or dbkey) and whether this HDA is visible and/or deleted. Setting deleted to true does not mean the file and HDA are purged from the filesystem and database only they are 'marked as deleted' and any admin scripts that run later can purge them. The closest analogy is your desktop OS's trash bin - in this way Galaxy allows for a reversible step before the actual file/database removal.

Many update/PUT API methods will only allow the direct change of a limited set of attributes. It's generally best to check the documentation or source to find out what has been enabled in an update method.

The update method of the HDA API is designed to accept the same dictionary that's returned from the HDA show method. This allows us to capture the dictionary from that method, change what we need, and send the same dictionary to the update method. It also allows a partial dictionary with just the keys and new values we want to change.

In either case, the update method returns a dictionary of what has successfully been changed.

Note: unfortunately, at the time this was written the HDA annotations were not passed back in the HDA show method. Although we do change the annotations of the read files in step_9.py, you'll need to check for the changed annotations in your browser - by navigating to the most recent 'Step 9' history and opening the annotations of the forward and reverse reads.

Step 9 will, however, show us the changed names of the HDAs:

   1 ./step_9.py

The output:

   1 created history! Step 9
   2 uploaded hda! myIlluminaRun.solexa.fastq
   3      uploaded_hda_state: queued
   4      (waiting 4 seconds...)
   5 found workflow! Joined Solexa QC
   6 running Joined Solexa QC workflow...
   7 workflow started!
   8 ...
   9 workflow complete!
  10 get_dataset_column, full_url: ...
  11 mean: 38.8721137988 median: 19.776113249
  12 found fwd/rev reads: forward reads reverse reads
  13 Forward: myIlluminaRun.fwd.fastqsanger
  14 Reverse: myIlluminaRun.rev.fastqsanger


10. Copying an HDA to a library using the library contents API

For our final step, we'll take our quality controlled and annotated read files and copy them into a shared Library for everyone in our lab to use.

Scripts: step_10.py, libraries_1.py, lddas_1.py

The API is a great way for administrators and 'power-users' (that's you now) to create and manage shared resources quickly and easily. This makes life easier for people at your lab that are more comfortable with the UI (or those that simply 'explore' the data more there) by creating accessible, shared data and makes your life easier by automating all or parts of the frequent bioinformatic requests you may get.


libraries_1.py and lddas_1.py

Libraries are unique containers (when compared to histories) in that they can contain other containers. These nested containers are folders and add a layer of complexity to our last task.

For the purposes of simplifying this workshop, we'll only copy our reads to the root folder of an existing library, but you should be aware that the Library/LDDA API is capable of creating and modifying both libraries and any amount of folders within them. Libraries (and their contents) can also be imported or exported through the API, effectively making it a good candidate of data communication or migration between Galaxy instances.

Step 9 will, however, show us the changed names of the HDAs:

   1 ./step_10.py

The output:

   1 created history! Step 10
   2 uploaded hda! myIlluminaRun.solexa.fastq
   3      uploaded_hda_state: queued
   4      (waiting 4 seconds...)
   5 found workflow! Joined Solexa QC
   6 running Joined Solexa QC workflow...
   7 workflow started!
   8 ...
   9 workflow complete!
  10 get_dataset_column, full_url: http://localhost:8081/api/datasets/ba751ee0539fff04?key=a63a28c5d5575a8ca78e97e01a73f901&data_type=raw_data&provider=column&columns=[5]
  11 mean: 38.8721137988 median: 19.776113249
  12 found fwd/rev reads: forward reads reverse reads
  13 changed read HDA names and annotations: myIlluminaRun.fwd.fastqsanger myIlluminaRun.rev.fastqsanger
  14 found reads library: Reads Library
  15 found root folder: /
  16 copying HDAs to library: Reads Library
  17      forward read: myIlluminaRun.fwd.fastqsanger
  18      reverse read: myIlluminaRun.rev.fastqsanger
  19 ...
  20 Forward reads in library:
  21 { u'data_type': u'fastqsanger',
  22   u'deleted': False,
  23   u'file_name': u'/home/gcc2013/Desktop/Training_Day_Workshops/API/galaxy-central/database/files/000/dataset_26.dat',
  24   u'file_size': 1093696,
  25   u'genome_build': u'?',
  26   u'hda_ldda': u'ldda',
  27   ...
  28 Reverse reads in library:
  29 { u'data_type': u'fastqsanger',
  30   u'deleted': False,
  31   u'file_name': u'/home/gcc2013/Desktop/Training_Day_Workshops/API/galaxy-central/database/files/000/dataset_27.dat',
  32   u'file_size': 1093696,
  33   u'genome_build': u'?',
  34   u'hda_ldda': u'ldda',
  35   ...

And ... we're done!



In the Future

We are always adding new functionality to the API and progressively ensuring that everything that can be done in the UI can be done in the API - because of that, please, consider the API a 'work in progress' and check the Galaxy Development Briefs for new features or fixes.


Versioning of the API

As things change in Galaxy and we make improvements to the API, there will be times when the data that's passed to an API method, the data that's returned, or the fundamental effects of an API method should change.

The fact that people have and will continue to write scripts that depend on the API functioning in a particular way complicates this issue and the developers know that a certain amount of backwards compatibility should be maintained.

Although the API should still be considered a 'work in progress', there are plans to have two versions of the API. The first, would be the 'stable' api and the second: the 'development' version. The development version may not always behave in the same manner as the stable (hopefully it will be an improvement) and shouldn't be relied upon as a source for data or control. The 'stable' version is meant to be used and consumed and should provide the most stable, secure, and robust system for scripting the API.

As the development version nears a point of completion and reliability itself, the previous stable version will be replaced with it and a new development version will be started.


Tool Running

We know that running tools via the API would be a big win for everyone - so it's a priority on our list. Before that happens some amount of core changes need to take place to make this as easy and flexible to use as possible.



More Resources and Thanks

Questions? Contact the Organizers.