Canvas Data 2/DAP Client Library Production Readiness

[WSUDavidFink]

My institution has historically used Canvas Data to access and extract data that in turn is loaded and used in our data warehouse. In response to the announcement of Canvas Data 2 (CD2), we started an implementation of new data flows which access CD2 through the DAP Client Library (CLI) provided by Instructure. However, we haven't been unable to reliably extract the data we need through Canvas Data 2 using the provided CLI. 

The issues we've seen are primarily generic HTTP 400 errors (with equally generic error messaging), which are raised by the CLI when attempting to download snapshots of data tables requested through CD2. However, this also creates a problem in which we have not been able to fully validate the integrity and accuracy of the data returned from CD2 against "Canvas Data 1" (CD1) data we are currently using in Production (as is recommended by Instructure when integrating CD2 into ETL). We're told the download issues are a result of the amount of data our institution has and that the engineering team was already working on some fix to ensure larger datasets can be downloaded successfully, but were provided no timeline for when this fix may be available. We understand there are other institutions experiencing this issue as well. As an additional factor, the DAP Client Library is still listed as being in the beta stage of development despite CD2 reaching GA eight months ago. 

With the deprecation date for CD1 being next month, there is growing concern at my institution from both technical and business stakeholders that Canvas Data 2/the DAP Client Library is not ready for Production use. Is Instructure still committed to the December 31st deprecation date? If so, what options do institutions have to ensure uninterrupted access to reliable, complete, and accurate Canvas data? Finally, will the DAP Client Library be promoted to a Production-ready state in time for the deprecation date?

[mclark19]

We are in relatively good shape with our CD2 processes, though we have yet to go a full week without some kind of error, usually caused by the authentication token timing out (which leads to a 401 error) while trying to get updated status for a job. We've built in some restart logic, so we are usually successful on the second try, but I would feel more confident if those were fewer and further between than they have been. (Things seem a little better this week, so fingers crossed.) I just sense that there are capacity issues that haven't been fully resolved yet, which makes me a bit nervous about CD1 being retired.

Perhaps someone from the Instructure side could weigh in on improvements that have been made on their end in the recent past and what expectations are for stability moving forward.

[dsweeney2]

+1 for token timeouts here. We've also had issues with missing data, where the first file of multiple parts is empty when retrieving snapshots.

[panda_bot]

---

Is Instructure still committed to the December 31st deprecation date? If so, what options do institutions have to ensure uninterrupted access to reliable, complete, and accurate Canvas data? Finally, will the DAP Client Library be promoted to a Production-ready state in time for the deprecation date?

---

I'm also curious about this as my team continues to run into issue with CD2. Any chance we can have the Instructure team chime in? 

[reynlds]

Has Instructure unleashed @panda_bot AI on to us in a "posting" capacity all of a sudden??? 

[audra_agnelly]

I'm increasingly concerned about the stability of this product for the December 31st sunset. We are running CD2, CD1, and building some of our own tables via APIs. We are excited about the data latency improvements of CD2 and are planning to move away from the API calls to pull certain data, but we would have been in a major institutional bind several weeks ago had we been relying on CD2 as the source of data. We had a few weeks where everything was working smoothly and this morning we're stuck in this 'job status: running' loop and none of our tables are downloading.

[BobODell]

@WSUDavidFink  Thanks for sharing your experience with us. As a product lead for Instructure data and analytics I wanted to help address your concerns and share a little more about what's happening and what's coming. We are 100% committed to Canvas Data 2 and also to deprecating the legacy Canvas Data 1 at the end of this year. Today Canvas Data 2 is used by hundreds of customers, running millions of queries per month, processing billions of transactional events each and every day, and delivering better latency with a richer data set.

With that said, we recognize there have been a few issues that have come up. Fixing all issues that are raised through our support team is our #1 priority and several of those fixes are being rolled out within the next week. Another issue we realize some customers have encountered is with the CLI and Client Library. We are also working hard to improve the performance and any identified issues and are planning to provide continuous updates in the coming weeks. This improvement and focus won't stop with these and we are working hard to continue addressing any additional issues that are identified as quickly as possible. Please continue to submit issues to our support team as thats the fastest way for us to address your concerns. 

[WSUDavidFink]

@BobODell, thank you for the response.

For clarity for you and the community here, my institution opened a support ticket about the key issue blocking us from being able to adopt Canvas Data 2 four weeks ago, but we've been disappointed that no one from Instructure has been able to communicate a specific timeline for when the issue preventing us and other large institutions from adopting CD2 will be resolved. With four weeks until the CD1 deprecation date and no communication on when CD2 will be fully functional, there is significant concern that we'll lose access to Canvas data (causing a data outage), which is made worse by the fact that my team still has no timeline to pass along to our users and data consumers for how long a potential outage may be. This is made worse by the fact that losing access to the CD1 data we've been using will make it significantly more difficult for us to validate data coming through CD2 to start using it in Production.

Regarding the unanswered questions I posed in my original post, can you speak to a) what options institutions experiencing the same/similar issues have to avoid data outages and b) whether we can expect the DAP Client Library to be considered out of beta and Production-ready by December 31st, 2023?

I appreciate any guidance you can provide to me as well as others experiencing similar issues.