Cases6 cases

5 to 6 of 6

1. IL-0549 Media, attachments, Git LFS and build artifacts failed for Atlassian in us-west-1 when an old network stack was removed

   Atlassian2023

   Symptoms

   From 02:36 UTC on 15 May 2023, Atlassian products with services hosted in the us-west-1 region could not store or retrieve data assets such as media, attachments and build artifacts. Automated monitoring detected it within three minutes, and the status page reported intermittent errors.

   High error rateStale endpoint routing

   Trigger

   An internal service was migrated to a new network segment as part of an initiative to harden the partitions between network segments, and the old network stack was then removed.

   Root cause

   Configuration change

   Component

   Authoritative DNS servers and records

   Contributing factor

   Unverified assumptionUnrepresentative staging

   Detection

   Automated monitoring detected the incident within three minutes.

   Remediation

   • Rolled back the network migration change, returning Atlassian systems to a known good state

   Action items

   • Review the systems that decommission service stacks and check that customer traffic is no longer served before a stack is decommissioned
   • Add steps to the service network migration process that identify associated DNS records needing attention

   Lessons

   Atlassian said the migration used blue/green deployment to limit this kind of risk, but adequate verification was not run before the old stack was decommissioned, and moving services between network segments is rare and hard to replicate in test.

   View case
2. IL-0363 static.crates.io did not resolve for 10-15 minutes during a Terraform deployment, so Rust builds could not fetch crates

   Rust Project2023

   Symptoms

   On 25 January 2023, shortly after a deployment to the crates.io production infrastructure at 09:15 UTC, the DNS record for static.crates.io failed to resolve for an estimated 10-15 minutes. Builds failed for users because crates could not be downloaded.

   DNS resolution failures

   Trigger

   A Terraform deployment to production contained a new TLS certificate for the current CDN of static.crates.io and updated DNS records; Terraform removed the current certificate and DNS records first and recreated the records only after the new certificate had been issued.

   Root cause

   Configuration change

   Component

   TerraformAuthoritative DNS servers and recordsCDN and edge network

   Contributing factor

   No safeguard on destructive changeUnrepresentative staging

   Remediation

   • Terraform finished the deployment and created a new DNS record for static.crates.io, which ended the outage

   Action items

   • Document the process of deploying changes to production, including how to pick the Git commit and how to review the changeset
   • Deploy changes developed and tested in isolation on staging individually and in sequence to production, and add this to the documentation
   • Deploy changes to production when they are merged into the main branch, so the configuration in Git and what is deployed do not drift

   Lessons

   Deploying the TLS certificate change and the DNS record change separately would have avoided the outage. Changes tested one at a time on staging need to reach production one at a time and in order, the production deploy process needs to be written down so it can be improved, and merged changes need to be deployed promptly so Git and production do not drift apart.

   View case

By root cause1