Uploaded June 2026 | Updated September 2026, 3 weeks ago
How Canva Built Simple, Auditable, and Maintainable Data Retention
Dr. Phillip Ward, Canva
When data retention fails, companies face regulatory scrutiny and customer distrust. At Canva's scale, manual compliance audits are not viable and the engineering effort required for building and ongoing maintenance are unsustainable. When we refreshed our internal data policies, we saw an opportunity for change. We re-architected retention from the ground up to coordinate and evidence deleting data.
We delivered a new retention framework that required few lines of code per service, and scales across the thousands of storage instances at Canva. The new design prioritises direct measurement of compliance, and minimal engineering effort. The approach decoupled policy from implementation, and isolated domain-specific data knowledge to individual services. Despite it's simplicity, it supports daily audits, logging, chaining, and zero-code policy changes.
In this talk, I will discuss the design choices that underpin this novel approach to retention.
This work includes fantastic contributions from Xushen Ma, Jessica Hu, Saeed Attar, and many other colleagues at Canva.
View the full PEPR '26 program at usenix.org/conference/pepr26/program
How Canva Built Simple, Auditable, and Maintainable Data Retention
Dr. Phillip Ward, Canva
When data retention fails, companies face regulatory scrutiny and customer distrust. At Canva's scale, manual compliance audits are not viable and the engineering effort required for building and ongoing maintenance are unsustainable. When we refreshed our internal data policies, we saw an opportunity for change. We re-architected retention from the ground up to coordinate and evidence deleting data.
We delivered a new retention framework that required few lines of code per service, and scales across the thousands of storage instances at Canva. The new design prioritises direct measurement of compliance, and minimal engineering effort. The approach decoupled policy from implementation, and isolated domain-specific data knowledge to individual services. Despite it's simplicity, it supports daily audits, logging, chaining, and zero-code policy changes.
In this talk, I will discuss the design choices that underpin this novel approach to retention.
This work includes fantastic contributions from Xushen Ma, Jessica Hu, Saeed Attar, and many other colleagues at Canva.
View the full PEPR '26 program at usenix.org/conference/pepr26/program
