Cloudflare, the content delivery and cybersecurity provider used by approximately 20% of all websites, says it has resolved the issue that caused almost a third of its sites to go down early on Friday morning. Affected sites included bank sites, the career social media platform LinkedIn and the virtual meeting service Zoom.

San Francisco-based Cloudflare said that up to 28% of its sites went down on Dec. 5 at 3:47 a.m. ET, after the company made changes to its systems, but said that it had fixed the problem within 25 minutes. 

Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.

The problem occurred when Cloudflare increased the buffer designed to prevent a critical vulnerability. To facilitate the update, the company says it turned off a testing tool, and this then caused the system to return an error. Reversing the changes fixed the issue.

This is the second Cloudflare outage in recent weeks. An unrelated change on Nov. 18 brought down the company’s services for multiple hours. 

Cloudflare is not the only web company to experience problems recently. An Amazon Web Services outage in October affected millions of visitors to popular destinations, including Reddit, Snapchat, Fortnite, Roblox, Venmo, and Amazon itself.