Monitoring a Live Experiment¶
There are a number of ways that you can monitor a live experiment:
Command line tools¶
dallinger summary --app {#id}
, where {#id}
is the id (w...
) of
the application.
This will print a summary showing the number of participants with each status code, as well as the overall yield:
status | count
----------------
1 | 26
101 | 80
103 | 43
104 | 2
Yield: 64.00%
The Dashboard¶
The Dallinger experiment server provides a dashboard view for experiment
administrators to monitor running experiments. The dasboard can be found at
/dashboard
, and requires login credentials that are provided by the
commandline output when launching an experiment using dallinger debug
,
dallinger sandbox
, or dallinger deploy
.
When running under dallinger debug
a browser window should open with the
dashboard already logged in. The dashboard username and password can also be
found in the dashboard_user
and dashboard_password
configuration
parameters in the deployed config.txt
configuration file. By default the
user is named admin
and the password is generated randomly, but the user
name and password can be specified using configuration files.
Customizing the Dashboard¶
You can add custom tabs to the Dallinger Dashboard by adding and registering
new Flask routes on the dashboard
Blueprint, and resgistering the view as a dashboard_tab
. For example in your experiment.py
you
could add the following code to add a “My Experiment” tab to the dashboard:
from dallinger.experiment_server.dashboard import dashboard, dashboard_tabs
@dashboard.route("my-experiment")
def my_experiment():
return "Hello, World. This is some information about My Experiment"
dashboard_tabs.insert("My Experiment", "my-experiment")
The dashboard also supports nested tab/menus using the DashboardTab
object:
from dallinger.experiment_server.dashboard import dashboard_tabs, DashboardTab
def child_tabs():
return [DashboardTab('Child1', 'child1'), DashboardTab('Child2', 'child2')]
complex_tab = DashboardTab('Title', 'route_name', child_tabs)
dashboard_tabs.insert_tab(complex_tab)
The dashboard_tabs
object supports the following methods for managing the
available tabs on your experiment’s dashboard:
-
class
dallinger.experiment_server.dashboard.
DashboardTabs
(tabs)[source]¶ -
insert
(title, route_name, position=None)[source]¶ Creates a new dashboard tab and inserts it (optionally at a specific position)
Parameters: - title (str) – Title string to appear in the dashboard HTML
- route_name (str) – The registered route name (optionally prefixed with
dashboard.
) - position (int, optional) – The 0-based index where the tab should be inserted. By default tabs will be appended to the end.
-
insert_tab
(tab, position=None)[source]¶ Insert a new dashboard tab (optionally at a specific position)
Parameters: - tab (DashboardTab) – DashboardTab instance
- position (int, optional) – The 0-based index where the tab should be inserted. By default tabs will be appended to the end.
-
insert_before_route
(title, route_name, before_route)[source]¶ Creates a new dashboard tab and inserts it before an existing tab by route name
Parameters: - title (str) – Title string to appear in the dashboard HTML
- route_name (str) – The registered route name (optionally prefixed with
dashboard.
) - before_route (str) – The route name to insert this tab before.
Raises: ValueError – When
before_route
is not found in registered tabs
-
insert_tab_before_route
(tab, before_route)[source]¶ Insert a new dashboard tab before an existing tab by route name
Parameters: - tab (DashboardTab) – DashboardTab instance
- before_route (str) – The route name to insert this tab before.
Raises: ValueError – When
before_route
is not found in registered tabs
-
insert_after_route
(title, route_name, after_route)[source]¶ Creates a new dashboard tab and inserts it after an existing tab by route name
Parameters: - title (str) – Title string to appear in the dashboard HTML
- route_name (str) – The registered route name (optionally prefixed with
dashboard.
) - after_route (str) – The route name to insert this tab after.
Raises: ValueError – When
after_route
is not found in registered tabs
-
insert_tab_after_route
(tab, after_route)[source]¶ Insert a new dashboard tab after an existing tab by route name
Parameters: - tab (DashboardTab) – DashboardTab instance
- after_route (str) – The route name to insert this tab after.
Raises: ValueError – When
after_route
is not found in registered tabs
-
The DashboardTab
object used by the various insert_tab*
methods provide the following API:
-
class
dallinger.experiment_server.dashboard.
DashboardTab
(title, route_name, children_function=None, params=None)[source]¶ -
__init__
(title, route_name, children_function=None, params=None)[source]¶ Creates a new dashboard tab
Parameters: - title (str) – Title string to appear in the dashboard HTML
- route_name (str) – The registered route name (optionally prefixed with
dashboard.
) - children_function – A callable that returns an iterable of
DashboardTab
to be displayed as children of this tab - params – A mapping of url query string parameters used when generating the route url.
-
The dashboard monitoring view can be extended by adding panes to the sidebar or
extending the existing panes. This can be done customizing the
monitoring_panels
and/or
monitoring_statistics
methods of
your experiment class. Additionally, you can customize the display of the selected
nodes customizing the node_visualization_html
method, or the visualization_html
property on your
model class.
The dashboard database view can be customized by customizing the
json_data
method on your model classes to
add/modify data provided by each model to the dashboard views, or by modifying
the DataTables data returned by the
table_data
method in your
Experiment
class.
-
class
dallinger.experiment.
Experiment
(session=None)[source] Define the structure of an experiment.
-
monitoring_panels
(**kw)[source]¶ Provides monitoring dashboard sidebar panels.
Parameters: **kw – arguments passed in from the request Returns: An OrderedDict()
mapping panel titles to HTML strings to render in the dashboard sidebar.
-
monitoring_statistics
(**kw)[source]¶ The default data used for the monitoring panels
Parameters: **kw – arguments passed in from the request Returns: An OrderedDict()
mapping panel titles to data structures describing the experiment state.
-
node_visualization_html
(object_type, obj_id)[source]¶ Returns a string with custom HTML visualization for a given object referenced by the object base type and id.
Parameters: - object_type (str) – The base object class name, e.g.
Network
,Node
,Info
,Participant
, etc. - id (int) – The
id
of the object
Returns: A valid HTML string to be inserted into the monitoring dashboard
- object_type (str) – The base object class name, e.g.
-
table_data
(**kw)[source]¶ Generates DataTablesJS data and configuration for the experiment. The data is compiled from the models’
__json__
methods, and can be customized by either overriding this method or using thejson_data
method on the model to return additional serializable data.Parameters: **kw – arguments passed in from the request. The model_type
parameter takes astr
or iterable and queries all objects of those types, ordered byid
.Returns: Returns a dict
with DataTablesJS data and configuration, filters using arbitrary keyword arguments. Should containdata
andcolumns
keys at least, withcolumns
containing data for all fields on all returned objects.
-
dashboard_database_actions
()[source] Returns a sequence of custom actions for the database dashboard. Each action must have a
title
and aname
corresponding to a method on the experiment class.The named methods should take a single
data
argument which will be a list of dicts representing the datatables rendering of a Dallinger model object. The named methods should return adict
containing a"message"
which will be displayed in the dashboard.Returns a single action referencing the
dashboard_fail()
method by default.
-
You may also add new actions to the dashboard database view by adding additional
title
and name
pairs to the
dashboard_database_actions()
output along
with corresponding methods that process submitted data. The
dashboard_fail()
method is an example of
such an action.
Papertrail¶
You can use Papertrail to view and search the live logs of your experiment. You can access the logs either through the Heroku dashboard’s Resources panel (https://dashboard.heroku.com/apps/{#id}/resources), where {#id} is the id of your experiment, or directly through Papertrail.com (https://papertrailapp.com/systems/{#id}/events).
Setting up alerts¶
You can set up Papertrail to send error notifications to Slack or another communications platform.
- Take a deep breath.
- Open the Papertrail logs.
- Search for the term
error
. - To the right of the search bar, you will see a button titled “+ Save Search”. Click it. Name the search “Errors”. Then click “Save & Setup an Alert”, which is to the right of “Save Search”.
- You will be directed to a page with a list of services that you can use to set up an alert.
- Click, e.g., Slack.
- Choose the desired frequency of alert. We recommend the minimum, 1 minute.
- Under the heading “Slack details”, open (in a new tab or window) the link new Papertrail integration.
- This will bring you to a Slack page where you will choose a channel to post to. You may need to log in.
- Select the desired channel.
- Click “Add Papertrail Integration”.
- You will be brought to a page with more information about the integration.
- Scroll down to Step 3 to get the Webhook URL. It should look
something like
https://hooks.slack.com/services/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
. - Copy this link to your clipboard.
- Change anything else you want and then scroll to the bottom and click “Save integration”.
- Go back to Papertrail page that you left in Step 7.
- Paste the copied URL into the input text box labeled “Integration’s Webhook URL” under the “Slack Details” heading.
- Click “Create Alert” on the same page.
- Victory.