Scaling Jira to ~6M Versions
Atlassian, Wednesday, November 11th, 2026
Atlassian explains how it cut Jira Search query resolution time for a customer with about 6 million release versions.
Jira Search resolves JQL queries by replacing names and functions with ids before execution, and version functions such as earliestUnreleasedVersion() originally loaded all versions into memory for filtering.
For a customer with about 6 million versions this caused slow database scans and large network transfers, with the slowest resolutions taking almost 30 seconds and P99 around 18 seconds.
A cache bought time but did not fix the underlying problem.
Atlassian changed how it retrieves and processes version data, using project-scoped queries and parallel batching, and P99 resolution time dropped to 92 milliseconds.