Developer Cookbook¶
This cookbook consists of a set of examples of common tasks that developers may encounter when developing Girder applications.
Client cookbook¶
The following examples are for common tasks that would be performed by a Girder client application.
Authenticating to the web API¶
Clients can make authenticated web API calls by passing a secure temporary token
with their requests. Tokens are obtained via the login process; the standard
login process requires the client to make an HTTP GET
request to the
api/v1/user/authentication
route, using HTTP Basic Auth to pass the user
credentials. For example, for a user with login “john” and password “hello”,
first base-64 encode the string "john:hello"
which yields "am9objpoZWxsbw=="
.
Then take the base-64 encoded value and pass it via the Authorization
header:
Authorization: Basic am9objpoZWxsbw==
If the username and password are correct, you will receive a 200 status code and a JSON document from which you can extract the authentication token, e.g.:
{
"authToken": {
"token": "urXQSHO8aF6cLB5si0Ch0WCiblvW1m8YSFylMH9eqN1Mt9KvWUnghVDKQy545ZeA",
"expires": "2015-04-11 00:06:14.598570"
},
"message": "Login succeeded.",
"user": {
...
}
}
The authToken.token
string is the token value you should pass in subsequent API
calls, which should either be passed as the token
parameter in the query or
form parameters, or as the value of a custom HTTP header with the key Girder-Token
, e.g.
Girder-Token: urXQSHO8aF6cLB5si0Ch0WCiblvW1m8YSFylMH9eqN1Mt9KvWUnghVDKQy545ZeA
Note
When logging in, the token is also sent to the client in a Cookie header so that web-based clients can persist its value conveniently for its duration. However, for security reasons, merely passing the cookie value back is not sufficient for authentication.
Upload a file¶
If you are using the Girder javascript client library, you can simply call the upload
method of the girder.models.FileModel
. The first argument is the parent model
object (an ItemModel
or FolderModel
instance) to upload into, and the second
is a browser File
object that was selected via a file input element. You can
bind to several events of that model, as in the example below.
var fileModel = new girder.models.FileModel();
fileModel.on('g:upload.complete', function () {
// Called when the upload finishes
}).on('g:upload.chunkSent', function (info) {
// Called on each chunk being sent
}).on('g:upload.progress', function (info) {
// Called regularly with progress updates
}).on('g:upload.error', function (info) {
// Called if an upload fails partway through sending the data
}).on('g:upload.errorStarting', function (info) {
// Called if an upload fails to start
});
fileModel.upload(parentFolder, fileObject);
If you don’t feel like making your own upload interface, you can simply use
the girder.views.UploadWidget
to provide a nice GUI interface for uploading.
It will prompt the user to drag and drop or browse for files, and then shows
a current and overall progress bar and also provides controls for resuming a
failed upload.
Using the Girder upload widget in a custom app¶
Your custom javascript application can easily reuse the existing upload widget provided in the Girder javascript library if you don’t want to write your own upload view. This can save time spent duplicating functionality, since the upload widget provides current and overall progress bars, file displays, a drag-and-droppable file selection button, resume behavior in failure conditions, and customizable hooks for various stages of the upload process.
The default behavior of the upload widget is to display as a modal dialog, but many users will want to simply embed it underneath a normal DOM element flow. The look and behavior of the widget can be customized when the widget is instantiated by passing in options like so:
new girder.views.UploadWidget({
option: value,
...
});
The following options are not required, but may be used to modify the behavior of the widget:
[parent]
- If the parent object is known when instantiating this upload widget, pass the object here.[parentType=folder]
- If the parent type is known when instantiating this upload widget, pass the object here. Otherwise setnoParent: true
and set it later, prior to starting the upload.[noParent=false]
- If the parent object being uploaded into is not known at the time of widget instantiation, passnoParent: true
. Callers must ensure that the parent is set by the timeuploadNextFile()
actually gets called.[title="Upload files"]
- Title for the widget. This is highly recommended when rendering as a modal dialog. To disable rendering of the title, simply pass a falsy object.[modal=true]
- This widget normally renders as a modal dialog. Passmodal: false
to disable the modal behavior and simply render underneath a parent element.[overrideStart=false]
- Some callers will want to hook into the pressing of the start upload button and add their own logic prior to actually sending the files. To do so, setoverrideStart: true
and bind to theg:uploadStarted
event of this widget. The caller is then responsible for callinguploadNextFile()
on the widget when they have completed their actions and are ready to actually send the files.
For general documentation on embedding Girder widgets in a custom application, see the section on client development.
Server cookbook¶
The following examples refer to tasks that are executed by the Girder application server.
Creating a REST route¶
The process of creating new REST resources and routes is documented here.
The API docs of the route
method can be found
here.
Loading a resource by its ID¶
This is a fundamental element of many REST operations; they receive a parameter representing a resource’s unique ID, and want to load the corresponding resource from that ID. This behavior is known as model loading. As a brief example, if we had the ID of a folder within our REST route handler, and wanted to load its corresponding document from the database, it would look like:
self.model('folder').load(theFolderId, user=self.getCurrentUser(), level=AccessType.READ)
The load
method of each model class takes the resource’s unique ID as its
first argument (this is the _id
field in the documents). For access controlled
models like the above example, it also requires the developer to specify
which user is requesting the loading of the resource, and what access level is required
on the resource. If the ID passed in does not correspond to a record in the database,
None
is returned.
Sometimes models need to be loaded outside the context of being
requested by a specific user, and in those cases the force
flag should be used:
self.model('folder').load(theFolderId, force=True)
If you need to load a model that is in a plugin rather than a core model, pass
the plugin name as the second argument to the model
method:
self.model('cat', 'cats').load(...)
The ModelImporter class
conveniently exposes a method for retrieving instances of models that are statically
cached for efficient reuse. You can mix this class into any of your classes to
enable self.model
semantics. The ModelImporter.model
method is
static, so you can also just do the following anywhere:
ModelImporter.model('folder')...
Send a raw/streaming HTTP response body¶
For consistency, the default behavior of a REST endpoint in Girder is to take
the return value of the route handler and encode it in the format specified
by the client in the Accepts
header, usually application/json
. However,
in some cases you may want to force your endpoint to send a raw response body
back to the client. A common example would be downloading a file from the server;
we want to send just the data, not try to encode it in JSON.
If you want to send a raw response, simply make your route handler return a
generator function. In Girder, a raw response is also automatically a streaming
response, giving developers full control of the buffer size of the response
body. That is, each time you yield
data in your generator function, the
buffer will be flushed to the client. As a minimal example, the following
route handler would send 10 chunks to the client, and the full response
body would be 0123456789
.
@access.public
def rawExample(self, params):
def gen():
for i in range(10):
yield str(i)
return gen
Serving a static file¶
If you are building a plugin that needs to serve up a static file from a path
on disk, you can make use of the staticFile
utility, as in the following
example:
import os
from girder.utility.server import staticFile
def load(info):
path = os.path.join(PLUGIN_ROOT_DIR, 'static', 'index.html')
info['serverRoot'].static_route = staticFile(path)
The staticFile
utility should be assigned to the route corresponding to
where the static file should be served from.
Note
If a relative path is passed to staticFile
, it will be interpreted
relative to the current working directory, which may vary. If your static
file resides within your plugin, it is recommended to use the special
PLUGIN_ROOT_DIR
property of your server module, or the equivalent
info['pluginRootDir']
value passed to the load
method.
Sending Emails¶
Girder has a utility module that make it easy to send emails from the server. For
the sake of maintainability and reusability of the email content itself, emails are stored
as Mako templates in the
girder/mail_templates directory. By convention, email templates should
include _header.mako
above and _footer.mako
below the content. If you wish
to send an email from some point within the application, you can use the
utility functions within girder.utility.mail_utils
, as in the example
below:
from girder.utility import mail_utils
...
def my_email_sending_code():
html = mail_utils.renderTemplate('myContentTemplate.mako', {
'param1': 'foo',
'param2': 'bar'
})
mail_utils.sendEmail(to=email, subject='My mail from Girder', text=html)
If you wish to send email from within a plugin, simply create a
server/mail_templates directory within your plugin, and it will be
automatically added to the mail template search path when your plugin is loaded.
To avoid name collisions, convention dictates that mail templates within your
plugin should be prefixed by your plugin name, e.g.,
my_plugin.my_template.mako
.
If you want to send email to all of the site administrators, there is a
convenience keyword argument for that. Rather than setting the to
field,
pass toAdmins=True
.
mail_utils.sendEmail(toAdmins=True, subject='...', text='...')
Note
All emails are sent as rich text (text/html
MIME type).
Logging a Message¶
Girder application servers maintain an error log and an information log and expose a utility module for sending events to them. Any 500 error that occurs during execution of a request will automatically be logged in the error log with a full stack trace. Also, any 403 error (meaning a user who is logged in but requests access to a resource that they don’t have permission to access) will also be logged automatically. All log messages automatically include a timestamp, so there is no need to add your own.
If you want to log your own custom error or info messages outside of those default behaviors, use the following examples:
from girder import logger
try:
...
except Exception:
# Will log the most recent exception, including a traceback, request URL,
# and remote IP address. Should only be called from within an exception handler.
logger.exception('A descriptive message')
# Will log a message to the info log.
logger.info('Test')
Adding Automated Tests¶
The server side Python tests are run using unittest. All of the actual test cases are stored under tests/cases.
Adding to an Existing Test Case
If you want to add tests to an existing test case, just create a new function in the relevant TestCase class. The function name must start with test. If the existing test case has setUp or tearDown methods, be advised that those methods will be run before and after each of the test methods in the class.
Creating a New Test Case
To create an entirely new test case, create a new file in cases that ends with _test.py. To start off, put the following code in the module (with appropriate class name of course):
from .. import base
def setUpModule():
base.startServer()
def tearDownModule():
base.stopServer()
class MyTestCase(base.TestCase):
Note
If your test case does not need to communicate with the server, you do not need to call base.startServer() and base.stopServer() in the setUpModule() and tearDownModule() functions. Those functions are called once per module rather than once per test method.
Then, in the MyTestCase class, just add functions that start with test, and they will automatically be run by unittest.
Finally, you’ll need to register your test in the CMakeLists.txt file in the tests directory. Just add a line like the ones already there at the bottom. For example, if the test file you created was called thing_test.py, you would add:
add_python_test(thing)
Re-run CMake in the build directory, and then run CTest, and your test will be run.
Note
By default, add_python_test allows the test to be run in parallel
with other tests, which is normally fine since each python test has its own
assetstore space and its own mongo database, and the server is typically
mocked rather than actually binding to its port. However, some tests (such
as those that actually start the cherrypy server) should not be run concurrently
with other tests that use the same resource. If you have such a test, use the
RESOURCE_LOCKS
argument to add_python_test. If your test requires the
cherrypy server to bind to its port, declare that it locks the cherrypy
resource. If it also makes use of the database, declare that it locks the
mongo
resource. For example:
add_python_test(my_test RESOURCE_LOCKS cherrypy mongo)
Serving a custom app from the server root¶
Normally, the root node (/
) of the server will serve up the Girder web client.
Some plugins will wish to change this so that their own custom app gets served out of
the server root instead, and they may also want to move the Girder web client to
be served out of an alternative route so they can still use it in addition to
their custom front-end application.
To achieve this, you simply have to swap the existing server root with your own
and rebind the old app underneath. In your plugin’s load
method, you would
add something like the following:
info['serverRoot'], info['serverRoot'].girder = CustomAppRoot(), info['serverRoot']
This will make it so that /
serves your CustomAppRoot
, and /girder
will
serve the normal Girder web client. That also has the side effect of moving the
web API (normally /api
) as well; it would now be moved to /girder/api
, which
would require a change to the server.api_root
value in girder.local.cfg
.
If you would rather your web API remained at /api
instead of moving under
/girder/api
, you would simply have to move it underneath the new server root. To
do that, just add the following line below the previous line:
info['serverRoot'].api = info['serverRoot'].girder.api
This will now serve the api out of both /api
and /girder/api
, which
may be desirable. If you only want it to be served out of /api
and not
/girder/api
, just add a final line below that:
del info['serverRoot'].girder.api
Supporting web browser operations where custom headers cannot be set¶
Some aspects of the web browser make it infeasible to pass the usual
Girder-Token
authentication header when making a request. For example,
if using an EventSource
object for SSE, or when you must redirect the user’s
browser to a download endpoint that serves its content as an attachment. In such
cases, you may allow specific REST API routes to authenticate using the Cookie.
You should only do this if the endpoint is “read-only”, that is, in cases where
it does not make modifications to data on the server, to avoid vulnerabilities
to Cross-Site Request Forgery attacks. If your endpoint is not read-only and
you are unable to pass the Girder-Token
header to it, you can pass a token
query parameter containing the token as a last resort, but in practice this will
probably never be the case.
In order to allow cookie authentication for your route, simply set the
cookieAuth
property on your route handler function to True
. Example:
@access.public
def download(self, params):
...
download.cookieAuth = True