[IDEA] Usage statistics for OIE - opt-in, anonymous; What will the community support?

[jonbartels]

We have metrics on OIE downloads. We do not have good information about how many running installations of OIE there are or what features are being used. This lack of data impacts the project. Having data about the installed systems will help us win support from corporate sponsors. Having data about features that are being used (or not used) will help us prioritize work.

This telemetry will be opt-in It will should never be on by default. Forcing users to register or even share anonymous statistics is wrong.

This telemetry will be non-intrusive failures to collect or send telemetry will not impact the operation of the application. Opting out of sending telemetry will have no affect on features.

I request community feedback about:

• is "telemetry" the right word? We're not capturing metrics, we're capturing information about the number of installations
• what data can we collect to identify unique installations and protect privacy? IPs? Server ID?
• what compliance issues do we need to consider? Is a GDPR guru available to give us advice
• what data is OK to collect?
• what data should we publish? Raw data? Aggregate stats?
• how often do we collect data? daily? weekly?
• when/how do we prompt for collection? I do not want to spam users with another popup. I hate those and my GH history shows that.
• what other open-source projects have good data collection practices that we can learn from? The Linux Foundation has a policy writeup

My starting idea is to collect the following information:

• OIE version
• Java version
• OS
• a sufficiently anonymous identifier for the instance - This could be done with fingerprinting, I know NextGen uses oshi. It could be done by IP or server ID. See above questions about what is OK to use
• a count of channels
• a count of deployed channels
• a count of types of connectors used in channels and deployed channels (eg 5 channels w/ HTTP, 6 channels w/ MLLP, 2 channels w/ SOAP but not deployed)
• a statistic about common settings - examples is pruning enabled, how many connectors use non-default queue settings, do TCP connectors use raw TCP or MLLP, what file writer types are in use
• This information will also be shown to the user for their instance(s). This is useful data to know about ones own instance of OIE
• This will likely be implemented as a plugin

Finally - I know the collection of usage stats is not well tolerated by the open-source community and not by privacy focused healthcare orgs. I hope my past comments and contributions to OIE and contributions to the user privacy plugin show my commitment to privacy and responsible use of data.

[mgaffigan]

100%. It's hard to make decisions in a project without knowledge of your user base. If the concern is misuse or differential knowledge, then make them public.

[david-labs-ca]

You just don't want how many OIE installs are there. But you also want which OS, what Java is being used and stuff like that. Then you can come to a conclusion which flavor is creating problem and for which Java version. I think there was a discussion on that sometime ago on discord.
Technically, we can collect how OIE is being used - like statistics. But what are we going to do with that data?

[StefanScholte]

personally I think making the stats public would make it more tolerable to collect the usage stats. Because then one can see what has been collected and how it can be used.

[jonbartels]

@StefanScholte For an initial release, should we publish only the data publicly or should we also publish our analysis of the data?

Have you seen other open-source projects that publish this kind of data in a way that you like? I would rather learn from another repo than invent this! haha

[StefanScholte]

@jonbartels Well personally I would be curious about the analysis of the data. for example, which setup is most preferable (OIE version, OS version, java version) . which type of connectors are being used the most etc.
I think posting the collected data is not a good idea because that is information that could potentially be misused.

Sorry I haven't seen another open-source project posting this.

[mgaffigan]

Data that can be misused should not be collected by an open source project.

Suggested anonymous stats (not anonymized, but anonymous from the start):

• OS
• JDK
• App version
• App defined feature use flags (true/false)

I do not suggest collecting usage volume or region beyond perhaps country level.

Access is via public influxdb+grafana or some other time series db.

[mgaffigan]

Example of a privacy page done well: https://karabiner-elements.pqrs.org/docs/privacy/