Kirill Müller 🇺🇦

We would like to guide our efforts towards improving duckplyr, focusing on the features with the most impact. For this, duckplyr now has opt-in telemetry: https://duckdblabs.github.io/duckplyr/reference/fallback.html .

This has to be enabled explicitly, and is fairly easy to opt out of entirely. We make an effort to anonymize the data before collecting it or uploading it.

I’m aware this is a sensitive topic, and I’m curious to hear your thoughts. Would you opt in, opt out, or upload manually? https://masto.machlis.com/@smach/112055384564988705

Fallback to dplyr — fallback

The duckplyr package aims at providing a fully compatible drop-in replacement for dplyr. To achieve this, only a carefully selected subset of dplyr's operations, R functions, and R data types are implemented. Whenever duckplyr encounters an incompatibility, it falls back to dplyr. To assist future development, the fallback situations can be logged to the console or to a local file and uploaded for analysis. By default, duckplyr will not log or upload anything. The functions and environment variables on this page control the process. fallback_sitrep() prints the current settings for fallback logging and uploading, the number of reports ready for upload, and the location of the logs. fallback_review() prints the available reports for review to the console. fallback_upload() uploads the available reports to a central server for analysis. The server is hosted on AWS and the reports are stored in a private S3 bucket. Only authorized personnel have access to the reports. fallback_purge() deletes some or all available reports.

Mar 8, 2024 at 8:48amWeb
Show threadBarry RowlingsonMar 8, 2024
@kirill R needs a telemetry package so you don't have to roll your own, and people then know where to go looking for telemetry settings.
Show threadMichael SumnerMar 8, 2024
@geospacedman @kirill I'm building telemetryverse, it's 26 loc so far
Show threadKirill Müller 🇺🇦Mar 8, 2024
@geospacedman Antoine Fabri has https://github.com/moodymudskipper/bigbrothr, but it needs more attention. Could that be an R Consortium project?
GitHub - moodymudskipper/bigbrothr: Provide Automated Feedback to Package Maintainers on the usage of their package.

Provide Automated Feedback to Package Maintainers on the usage of their package. - moodymudskipper/bigbrothr

GitHub
Show threadKirill Müller 🇺🇦Mar 8, 2024
@geospacedman And I’m still curious: would you opt in or opt out?
Show threadBarry RowlingsonMar 8, 2024
@kirill I'd opt out and add the IP address to my blocklist :)
Show threadKirill Müller 🇺🇦Mar 8, 2024

@geospacedman That’s definitely a valid choice, I hope it’s easy enough with duckplyr.

And I also hope it’s easy enough to do it the other way.

Show threadBarry RowlingsonMar 8, 2024
@kirill My main duckdb use is on a Secure TRE and telemetry would get blocked anyway. Takes 24 hours to get approval for data egress!
Show threadKirill Müller 🇺🇦Mar 8, 2024
@geospacedman I have split data collection and data upload, both can be configured separately. Do you have concerns regarding the data collection per se, or only regarding the upload?
Show threadBarry RowlingsonMar 8, 2024
@kirill Data collection (locally) is just "logging" by another name, right (does it use an extant logging package?). No problem with logging. I never want to involve remote servers with my data analysis in any way whatsoever though (and yes I understand this is the default for your system).
Show threadKirill Müller 🇺🇦Mar 8, 2024
@geospacedman Just to be super clear: both logging/data collection and uploads are opt-in. We write to a standard location, but it’s ndjson without a logging package. You could opt in for logging, review the logs and share with us what you think is good to share. How does that feel?
Show threadOlivier LeroyMar 8, 2024
@kirill Would a small survey could also solved that?
Show threadMilesMcBainMar 8, 2024

@defuneste @kirill this!

I appreciate you have done this in a very sensitive way, and I congratulate you for that. Unfortunately if you imagine telemetry is going to balance a ‘vocal minority’ in feature priority discussions I think you’ll be disappointed. I would not think to voluntarily enable telemetry unless I had some feature I was campaigning for. So I predict your vocal minority and users with telemetry voluntarily enabled will be the same people.

Show threadMilesMcBainMar 8, 2024
@defuneste @kirill trying to solicit opinions via a survey is much more likely to generate the kind of output you want IMHO.
Show threadvitaly_drukerMar 8, 2024
@kirill explicit opt-in would be a must for me. I think I'd only consider this if i heard about how the telemetry helped you - e.g. you write up something about how it specifically helped package development.
Show threadKirill Müller 🇺🇦Mar 8, 2024
@vitaly_druker Yes, it might have been unclear. It’s off by default, except for a demo and a nudge once per session. See https://github.com/duckdblabs/duckplyr?tab=readme-ov-file#telemetry (link to section in README). Does that help?
GitHub - duckdblabs/duckplyr: A drop-in replacement for dplyr, powered by DuckDB for performance.

A drop-in replacement for dplyr, powered by DuckDB for performance. - duckdblabs/duckplyr

GitHub
Show threadvitaly_drukerMar 8, 2024
@kirill yeah that sounds good, I was speaking more generally about this. I'm personally not a fan of the nudges (even if per session) I didn't see it yet but is there an option to keep it silent?
Show threadKirill Müller 🇺🇦Mar 8, 2024
@vitaly_druker Yes, the nudge once per session is what you’d need to opt out of, but that’s just setting an environment variable.