You are browsing a read-only backup copy of Wikitech. The live site can be found at wikitech.wikimedia.org
Analytics/Systems/Superset: Difference between revisions
|(3 intermediate revisions by the same user not shown)|
|Line 104:||Line 104:|
# from the -
# deploy only to an-tool1005, without logging in the ops's sal
# deploy only to an-tool1005, without logging in the ops's sal
|Line 138:||Line 135:|
=== Deploy to production ===
=== Deploy to production ===
# ssh to
# ssh to and set the working directory
== How to ==
== How to ==
Latest revision as of 18:49, 18 May 2022
Superset is an Apache incubator project, originally started at AirBnB. It enables visualizations and dashboards built from various analytics data sources. WMF's Superset instance can be found at https://superset.wikimedia.org. Like Turnilo, it provides access to various Druid tables, as well as data in Hive (and elsewhere) via Presto.
To access Superset, you need
nda LDAP access. For more details, see Analytics/Data access#LDAP access. Once done, you can log in using your Wikitech username and password.
- The "Druid Datasources" list shows ingested tables that are available for querying. As of October 2018, this includes e.g. daily and hourly pageviews data (the daily version is only updates once a month, but goes further back), a sampled excerpt of webrequest data, unique devices, and a few select EventLogging schemas. If a recently created Druid datasource is not yet visible in the list, try clicking "Scan New Datasouces".
- NULL values don't show up properly in the values selection dropdown list for filters (i.e. one can't use that dropdown to exclude NULL values from a chart or limit it to NULL values). But one can use the regex option instead: Type in ".+" (without the quotes), and accept the offer to create that as an option.
- By default, always use predefined SUM metrics when available. When choosing a metric then picking the SUM aggregation function, the aggregation is managed by superset and uses the
floatSumoperator. This operator uses 32 bits floats instead of 64 bits longs or double, leading to inaccuracies. Usually predefined
SUM(...)metrics are available and should be used, as they are manually defined using
longSum64 bits operators.
- If you build a chart based on a table with structs, you won't be able to access the fields of the struct because Superset recognizes the struct as a single string column. The workaround is to add a computed column with
struct.fieldas the SQL expression.
- Superset expects time columns to be in SQL timestamp string format (
2021-01-01 00:00:00) and has trouble with columns in ISO 8601 string format (
2021-01-01T00:00:00Z). To fix this, it's best to create a computed column that casts the time to the right format using an SQL expression like (
CAST(TO_ISO8601_TIMESTAMP(dt) AS VARCHAR).
Advanced time ranges
Superset accepts a lot of different things in advanced time ranges boxes, including SQL time functions and any human-language times accepted by the Parsedatetime library.Unfortunately, neither Superset nor Parsedatetime provides any documentation of this, so there's no actual list of all the possibilities. But when you put things in, Superset does at least give you immediate feedback about whether they're valid and what they're interpreted as. You can also use a variety of helpful date functions such as
Here are some useful examples:
|desired range||start time value||end time value|
|the last 26 full Monday-Sunday weeks
(useful for a weekly graph covering the last half year)
|up to yesterday
(useful when working with hourly data at daily granularity and today is incomplete)
Druid datasources vs Druid tables
We should stop adding datasources to the "Druid Datasources" list if possible in favor of Druid tables (backed by Druid SQLAlchemy mappings). In order to add a Druid table, do the following:
- Go to Sources -> Tables
- Hit the + sign to add one
Druid Analytics SQLas database,
druidas schema and the Druid datasource name as table name (like
Migrate a chart to Druid tables
It is easy to spot if a chart is not using a Druid table (but an old Druid datasource) simply checking the chart list:
In the above example the "pageviews_daily" datasource is not prepended with "druid." (like druid.pageviews_daily) so it is an old Druid datasource. If you want to move the chart to the Druid table, hit the edit button for it and switch the datasource name (it will likely change some default settings so make sure that the chart is correctly visualized) and then save it. You dashboards will be updated automatically!
If the druid.your-datasource-name is not present yet, please check the above section, since it may need to be defined via Druid -> Tables menu'.
Superset allows user to query data via SQL using a dedicated tool called SQL Lab: https://superset.wikimedia.org/superset/sqllab. Multiple databases are available to query:
presto_analytics_hiveto explore Data Lake data using Presto
Druid Analytics SQLto explore Druid data cubes (note: the full power of SQL is not available for Druid data)
Sync the staging database with the production one
When testing a new release in the staging environment it is nice to get the same dashboards as in production, since two different databases are used and they get out of sync very quickly (nobody updates dashboards in staging). The procedure is the following:
# The two databases live on the same mariadb instance ssh an-coord1001.eqiad.wmnet # Moving to /srv since it is on a separate partition with more free space cd /srv # Dump the production db sudo sh -c 'mysqldump superset_production > superset_production_$(date +%s).sql' # Get the filename of the production dump ls -l # Connect to mysql and drop the staging db sudo mysql # Good practice to avoid specific commands to be replicated via binlog. set session sql_log_bin=0; # Drop and re-create the staging database drop database superset_staging; create database superset_staging DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci; exit # Load the production database into the staging one. Change the filename according # to the dump previously taken! sudo mysql superset_staging < superset_production_111111.sql # Then remember that some specific production things need to be changed once superset # runs. For example, in databases -> presto-analytics (in the superset ui) the kerberos principal # used is saved in the config and changes between production and staging (the hostname is different).
To upgrade, first follow the instructions in the analytics/superset/deploy README to update the deploy repository. Once deployed, activate the superset virtualenv, add /etc/superset to PYTHONPATH (to allow superset to pick up configuration) and follow the Superset upgrade instructions (minus the pip install superset --upgrade part). This should be something like:
sudo su superset source /srv/deployment/analytics/superset/venv/bin/activate export PYTHONPATH=/etc/superset superset db upgrade superset init
Deploy to staging
This assumes you have a change open for the analytics/superset/deploy repository.
# ssh to deploy1002 and set the working directory ssh deploy1002.eqiad.wmnet cd /srv/deployment/analytics/superset/deploy # Checkout your change from gerrit (use "Download patch" in the top-right menu) git fetch https://gerrit.wikimedia.org/r/analytics/superset/deploy refs/changes/83/789683/1 && git checkout FETCH_HEAD # deploy only to an-tool1005, without logging in the ops's sal scap deploy --no-log-message -f -l an-tool1005.eqiad.wmnet "Test deployment for something important"
Test staging via ssh tunnel
# Create a ssh tunnel and then test it via localhost:8080 on the browser ssh -L 8080:an-tool1005.eqiad.wmnet:80 an-tool1005.eqiad.wmnet
Upon accessing http://localhost:8080, you will be presented with a basic auth challenge that allows you to authenticate via LDAP.
Test as different user on staging
In order to test features such as permissions, you may want to impersonate a different user.
First, add an extension to your browser that allows you to set custom headers, such as https://addons.mozilla.org/en-US/firefox/addon/modify-header-value/ for Firefox.
Then, set the following headers for localhost:8080:
X-Cas-Uid: <user you want to impersonate> X-Forwarded-Proto: http
Connect directly to the superset server by opening an SSH tunnel on an-tool1005 port 9080:
ssh -NL 8080:an-tool1005.eqiad.wmnet:9080 an-tool1005.eqiad.wmnet
Open http://localhost:8080 in your browser, and you should be logged in as the user specified by X-Remote-User. You can see who you are logged in as by viewing your profile from the Settings menu.
If you change the header to be a different user, logout via the Settings menu to change to be the new user.
Deploy to production
# ssh to deployment.eqiad.wmnet and set the working directory ssh deployment.eqiad.wmnet razzi@deploy1002:~$ cd /srv/deployment/analytics/superset/deploy $ scap deploy -l an-tool1010.eqiad.wmnet "Deployment for something important"
Don't forget to run migrations after you deploy (see Analytics/Systems/Superset#Upgrading).
Test a different role's permissions
During a couple of superset upgrades, there have been permissions removed from the Alpha role, which causes our users to see errors, sometimes, frustratingly, "unknown error" (at least by looking in the network tab you can see 401 unauthorized status).
It's theoretically possible to create test users, I don't have a method to do so currently so instead I modify my user to have the role I want to test. I do so in the superset app shell; other database operations could be done manually this way.
Log in to the superset instance (in this example I am connected to production, an-tool1010), switch to the superset user, set up the python environment, and start the shell:
razzi@an-tool1010:~$ sudo su superset superset@an-tool1010:/home/razzi$ source /srv/deployment/analytics/superset/venv/bin/activate (venv) superset@an-tool1010:/home/razzi$ export PYTHONPATH=/etc/superset (venv) superset@an-tool1010:/home/razzi$ superset shell Loaded your LOCAL configuration at [/etc/superset/superset_config.py] ... >>> app <SupersetApp 'superset.app'>
The database connection is accessible under
>>> session = app.extensions['sqlalchemy'].db.session >>> from flask_appbuilder.security.sqla.models import User >>> me = session.query(User).filter_by(username='razzi').first() >>> me razzi - >>> me.roles [Alpha] # You can modify roles as follows: >>> Role = me.roles.__class__ >>> session.query(Role).all() [Admin, Alpha, Gamma, granter, Public, sql_lab] >>> [admin, alpha, *roles] = session.query(Role).all() >>> me.roles.append(alpha) >>> me.roles.remove(admin) >>> session.add(me) >>> session.commit()
At this point I'd have alpha and not admin roles (this can be confirmed by going to your superset profile) and I can test what other Alpha users see.
systemctl status superset.service
systemctl restart superset