Usage Statistics

General

When you bring multiple Shiny apps to your end users, it may be interesting to track usage of the different applications over time. It may a.o. help to understand your user base and to prioritize maintenance work for the different applications.

ShinyProxy has a general model on board that allows one to store Shiny app events and for different storage back-ends, different plug-ins are available. Currently, ShinyProxy ships with plug-ins that can store events in the following back-ends:

Alternative plug-ins can easily be developed and are on the roadmap.

Starting from ShinyProxy 2.5.0 an extra plug-in was added that supports Micrometer.io. Micrometer itself is a modular library that allows to support many popular monitoring system. Our focus is to export Prometheus Metrics, however, any Micrometer supported monitoring system is supported.

For Kubernetes, a complete monitoring stack for ShinyProxy can be found on GitHub. This provides metrics using Prometheus, log aggregation using Loki and dashboards using Grafana. Everything has been prepared to easily deploy this on your infrastructure.

InfluxDB

Installation

InfluxDB installation is very simple: on an Ubuntu 16.04 LTS system one just needs to download and install the package using something along

wget https://dl.influxdata.com/influxdb/releases/influxdb_1.0.2_amd64.deb'
sudo dpkg -i influxdb_1.0.2_amd64.deb

To start the database one can use

sudo service influxdb start

and querying its status can be done with

sudo service influxdb status

Database creation

In order to store the usage statistics, a specific database needs to be created on the InfluxDB instance. In this example we will give it the name shinyproxy_usagestats.

To do so one can use the CLI client:

influx

then, at the database prompt, enter

CREATE DATABASE shinyproxy_usagestats
quit

Configuration

Once the database is created, we can configure ShinyProxy to post all usage events to the shinyproxy_usagestats database. It is sufficient to add a single property in the application.yml file under the shiny / proxy section:

proxy:

  #  ...

  usage-stats-url: http://localhost:8086/write?db=shinyproxy_usagestats

As can be seen, InfluxDB has the advantage that the storage of usage events can be done on a separate server and should not necessarily happen on localhost i.e. the machine on which the ShinyProxy server itself is running.

Note:

  • If authentication on the shinyproxy_usagestats database is required, one can use the usage-stats-username and usage-stats-password properties.
  • More detailed information on configuring ShinyProxy can be found here

SQL Based Databases (JDBC)

Installation

Please refer to the official documentation of your database system for installation. For example, on an Ubuntu 16.04 system MonetDB can be installed from the official Debian package repository that is available here.

Configuration

It is sufficient to only add a JDBC url in the usage-stats-url field in the proxy section of the application.yml file:

proxy:

  # ...

  # for MonetDB
  usage-stats-url: jdbc:monetdb://localhost:50000/usage_stats

  # for MySQL/MariaDB
  usage-stats-url: jdbc:mysql://localhost/shinyproxy

  # for SQLite
  usage-stats-url: jdbc:sqlite:shinyproxy.sql

  # for PostgreSQL
  usage-stats-url: jdbc:postgresql://localhost/postgres

  # for MS SQL Server:
  usage-stats-url: jdbc:sqlserver://localhost;databaseName=shinyproxy

Similarly to InfluxDB, the events can be stored on a separate server and event logging should not necessarily happen on localhost i.e. the machine on which the ShinyProxy server itself is running.

Note:

  • The default login and password that will be used to log into the database are monetdb and monetdb respectively. If other credentials have been set on the database, one can enter the usage-stats-username and usage-stats-password properties below the usage-stats-url field.
  • ShinyProxy automatically creates a table to store the events.
  • The default table used is event, this can be changed by configuring the proxy.usage-stats-table-name property.
  • More detailed information on configuring ShinyProxy can be found here.

Micrometer (Prometheus)

This section describes on how to set up the Micrometer monitoring backend using Prometheus. It assumes that ShinyProxy, Prometheus and Grafana (see later) are all running on localhost.

Configuration

Start by changing your ShinyProxy configuration so that it includes the following configuration:

proxy:

#  ...

    usage-stats-url: micrometer

management:
  prometheus:
    metrics:
      export:
        enabled: true

#  ...

The Prometheus metrics are now available at the localhost:9090/actuator/prometheus endpoint. Note that you should not include any context-path in this URL.

Setup Prometheus

The next step is to ensure that Prometheus scrapes your ShinyProxy server. For demonstrating purposes, it is enough to download it and extract it on your server. The next step is to adapt your configuration (by editing the included prometheus.yml file), so that Prometheus scrapes the ShinyProxy server:

scrape_configs:
  - job_name: 'shinyproxy'
    metrics_path: '/actuator/prometheus'
    static_configs:
      # note: this is the port of ShinyProxy Actuator service, not the port of Prometheus which is by default also 9090
      - targets: [ 'localhost:9090' ]
        labels:
          namespace: local
          shinyproxy_realm: local

Now you can start Prometheus. Make sure to run it on a different port than 9090, since that port is already being used by ShinyProxy:

./prometheus --web.listen-address="0.0.0.0:7070"

Now go to your browser and open http://localhost:7070/targets, this page should contain one target called shinyproxy with the UP state.

Setup Grafana

Now that Prometheus scrapes the metrics, we need a way to visualize these metrics. An excellent tool for this is Grafana. Start by downloading Grafana, again you can use the standalone binaries and simple extract them. Then start Grafana using:

./bin/grafana-server

Next, access Grafana on http://localhost:3000 and login using admin:admin. Next add Prometheus as datasource, remember that Prometheus is running at http://localhost:7070.

The final step is to import the dashboard. Download the JSON file (right-click and choose save as) and import it into Grafana. If everything is correctly setup, you should see the graph changing when you log in, start an app etc.

In order to provide more accurate graphs in Grafana, the scrape interval used by Prometheus should correspond to the one specified in the configuration of Grafana. The dashboard makes use of this option in order to perform certain calculations. This settings can be configured at the datasource settings in Grafana. See the Grafana docs.

Note: with the release of ShinyProxy 2.6.0 and 3.0.0 the dashboard has been updated and is incompatible with previous releases.

Exposed Metrics

All metrics exposed by ShinyProxy are tagged with the shinyproxy_instance tag, containing the instance id of the ShinyProxy server.

ShinyProxy keeps track of the following actions:

  • User Login (userLogins)
  • User Logout (also by expiring session) (userLogouts)
  • Authentication failure authFailed
  • Application start (appStarts, contains a spec_id tag)
  • Application stop (appStops, contains a spec_id tag)
  • Duration of application startup (startupTime, contains a spec_id tag)
  • Duration of application usage (usageTime, contains a spec_id tag)
  • Time between the start of the application process and when the app is reachable by ShinyProxy. For example, in the case of the Shiny app, this is the time the R process takes to startup and be reachable on its webserver. (applicationStartupTime, contains a spec_id tag)
  • Time needed for the container to get assigned to a node on Kubernetes. Includes the time to launch a new node if the cluster needs to scale-up. (containerScheduleTime, contains a spec_id tag)
  • Duration of pulling the image (only available on Docker and Kubernetes, imagePullTime, contains a spec_id tag)
  • Duration of container startup (before the process itself starts) (containerStartupTime, contains a spec_id tag)
  • Application start failures (startFailed, contains a spec_id tag)
  • Application crashes (appCrashes, contains a spec_id tag)
  • Absolute number of running apps (absolute_apps_running, contains a spec_id tag)
  • Absolute number of logged-in users (absolute_users_logged_in)
  • Absolute number of active users. A user is counted as active if they were active in the last minute. (absolute_users_active)

Screenshot

grafana-dashboard

Metric name prefix

It is a good practice to differentiate the metrics generated by ShinyProxy with any other metric stored in Prometheus. Of course, Prometheus attaches some tags (a.k.a labels) to all metrics that can be used for this purpose. Nevertheless, it can be useful to prefix all metrics exposed by ShinyProxy. This can be done by setting the proxy.usage-stats-micrometer-prefix option to a self-chosen string.

For example, the following code will prefix every metric exposed by ShinyProxy with shinyproxy_. It is not required to end the prefix with _ or any other separator symbol. The reason is that ShinyProxy automatically determines the correct separator for the used metric backend.

proxy:
  usage-stats-micrometer-prefix: shinyproxy

Using multiple statistics backends

In some cases it can be useful to log statistics to multiple databases. For example, logging infrastructure related metrics to Prometheus (i.e. observability) and keeping an audit trail in a SQL database. This is possible ( since ShinyProxy 3.0.0) using the proxy.usage-stats property. This property takes a list of configuration blocks. Each block should contain at least the parameter url which is explained above. Each block can also have a username and password property if needed for the statistics backend. For example, to log to Micrometer (for Prometheus) and PostgreSQL:

proxy:
  usage-stats:
    - url: micrometer
    - url: jdbc:postgresql://localhost/postgres
      username: postgres
      password: mysecretpassword
      table-name: my-custom-table-name # optional