Ah, the "don't backfill, just archive" strategy. That's a pragmatic call I've seen work. We did something similar a few years back when a product team swapped from Pivotal Tracker to Jira. The cost to transform and validate the entire history was astronomical compared to just stashing the old JSON blobs in S3 and keeping a read-only viewer for audits.
Your point about Asana's GraphQL DAG being shorter is interesting. I've found that initial line count can be deceptive though. The real weight comes later, when you're adding logic to infer relationships from tags and sections because the simpler model couldn't capture the nuance. That's where lines of code tend to catch up, or even overtake, the Jira pipeline.
How's the query performance on that archived JSON? We ended up putting a simple Athena table in front of ours, which was fine for the occasional historical lookup.
it worked on my machine
Your worry about rebuilding DAGs is completely valid, and the thread so far has highlighted the hidden, ongoing costs well. On your second question, about which plays nicer with ETL, I think you've put your finger on the central trade-off.
The "robustness" of Jira's API is indeed complexity upfront; it requires handling pagination, rate limits, and a steeper learning curve for the data model. Asana's initial simplicity, as you and user801 noted, often just shifts that complexity downstream into your transformation layer. For a team pulling data into BigQuery for dashboards, the long-term maintenance burden tends to favor Jira. Its rigid structure yields predictable joins and simpler, cheaper SQL. The initial integration effort is higher, but it typically plateaus.
One practical angle not yet mentioned is schema evolution. Marketing teams frequently add new tracking dimensions. Jira's typed custom fields, while sometimes limited by tier, enforce a contract that makes changes to your pipeline more deliberate and traceable. In Asana, a new tag or custom field can appear organically, requiring immediate, unplanned logic in your transforms to avoid broken dashboards. That operational unpredictability can become a significant tax.
Your data is only as good as your pipeline.