How a 3-Line Script Deleted 883 Atlassian Sites
Oct 29, 2026
On April 5, 2022, a three-line script ran for 23 minutes inside Atlassian’s cloud. When it finished, 883 customer sites were permanently deleted. For 775 companies, Jira, Confluence, Opsgenie, and Statuspage went dark — some for 14 days.
This is the story of a miscommunicated identifier, a peer review that didn’t catch the difference between a site ID and an app ID, a permanent-delete flag, and an API that accepted both without warning. It is also a story about disaster recovery built for infrastructure failure — not for “a script just deleted entire customer sites by ID.” And it ends with a lesson about intent validation: no dry run, no confirmation, no one asking, “What are we actually deleting?”
Chapters
00:00 — 883 sites gone in 23 minutes
00:35 — The setup: retiring Insights
01:05 — Failure 1: site IDs sent instead of app IDs
01:30 — Failure 2: permanent delete mode
01:55 — Failure 3: the API accepted both ID types
02:20 — The deletion window: 07:38–08:01 UTC
02:35 — Services affected: Jira, Confluence, Access, Opsgenie, Statuspage
02:50 — They deleted their own contact list
03:10 — Slow communication and underestimated impact
03:30 — Restoration: manual, site by site, 14 days
04:00 — Aftermath: soft delete, blocked bulk deletes, hardened API
04:25 — The real lesson: intent validation and dry runs
Key takeaway: Every individual step worked as designed. The missing control was not technical — it was asking what you are actually deleting before you delete it.
Show More Show Less 