In 2016, Stack Overflow experienced a 34-minute outage. The root cause? A regular expression (Regex) used in a part of the code that handled user input.
It took 10 minutes to pinpoint the issue, 14 minutes to write the fix, and another 10 minutes to deploy the solution and restore Stack Overflow’s availability.
What caused the outage?
^[\s\u200c]+|[\s\u200c]+$According to Stack Overflow: “While this looks straightforward—matching ‘all the spaces at the end of the string’—it can be problematic for backtracking regex engines. In this case, a malformed post included roughly 20,000 consecutive whitespace characters on a comment line.
If the string contains 20,000 space characters in a row but not at the end, the regex engine starts checking each space. After the 20,000th space, it encounters a different character but still expects a space or the end of the string.
The engine backtracks, attempting to match \s+$ starting from the second space, the third space, and so on, leading to a total of 199,990,000 checks. This was sufficient to cause a significant delay. The regex has since been replaced with a substring function.”
What Is Backtracking?
Backtracking occurs when a regular expression fails to match part of a string. While matching itself is straightforward, if a regex engine cannot find a match, it starts backtracking. This means the engine revisits previous choices and attempts different options, which can lead to performance issues.
Catastrophic Backtracking
In addition to general backtracking, there is a specific issue known as catastrophic backtracking. This happens when a regex engine spends an excessive amount of time trying different combinations to match a pattern, often leading to severe performance degradation. This is especially problematic with complex regex patterns and large input sizes.
The Stack Overflow incident serves as a reminder of the potential pitfalls of using regex. While a 34-minute outage might not seem like a major issue, what other problems has regex caused?
Cloudflare Outage (2019)
The CloudFlare outage on July 2, 2019, was another major example of how regex issues can lead to catastrophic failures.
An engineer wrote a regular expression prone to severe backtracking. This regex led to widespread CPU exhaustion, causing Cloudflare’s global CPU usage to spike to 100%.
(?:(?:\"|'|\]|\}|\\|\d|(?:nan|infinity|true|false|null|undefined|symbol|math)|\`|\-|\+)+[)]*;?((?:\s|-|~|!|{}|\|\||\+)*.*(?:.*=.*)))According to Cloudflare: “The regex was designed in a way that could cause significant backtracking. The critical part of the regex,.*(?:.*=.*)., involves a non-capturing group, which can lead to excessive CPU usage when processing certain patterns. This inefficiency in the regex caused severe performance issues, ultimately leading to a massive outage.”
CrowdStrike Kernel Issue (2024)
Recently, CrowdStrike faced a major issue due to a poorly implemented regular expression in their kernel driver, leading to widespread system crashes. This incident disrupted operations across various sectors, including businesses and government systems.
The root cause of the problem was a mismatch between the expected number of input parameters (21) and the actual number provided (20) to the Content Interpreter, which was responsible for processing regex-based Rapid Response Content. When the system received input with the 21st parameter, the Content Interpreter attempted to read beyond the allocated memory, resulting in out-of-bounds access and subsequent system crashes.
This incident shows the serious problems regex can cause in critical systems.
Should We Move Away from Regex?
I’m not claiming to know everything about regex. However, their issues highlight why we need to be careful with regex and explore alternative solutions. Is it time to phase them out?