August 5, 2026
Proxy war: slow code, hot comments
Google's HTTP/2 codec slows Envoy
Google’s new web traffic code made servers sweat, and commenters were not impressed
TLDR: A Google-backed change in Envoy made servers use noticeably more processing power, and tests showed the slowdown was real enough that the default was later reversed. Commenters turned it into a bigger debate about bad tech writing, speed versus purity, and whether anyone can tell if an article was written by a person anymore.
A routine software update was supposed to be boring. Instead, it turned into the kind of workplace panic that makes charts go up and everyone suddenly starts speaking in very serious tones. The team behind the post says one customer’s proxy servers started using about 20% more CPU power after an Envoy upgrade, and after digging, they traced the drama to a Google-made replacement for the old HTTP/2 engine — basically the part that helps websites and apps pass requests around quickly. The result? Slower performance, more strain, and a rollback later on when the newer default got swapped back out.
But honestly, the real fireworks were in the community reactions. One commenter came in swinging with a brutally dry review of the write-up itself, saying it looked interesting but only if “a human bothered to write it,” which is about as close as tech forums get to throwing a martini in someone’s face. Another summed up the bigger philosophical fight: rules and correctness are nice, but if they make everything slower, people will absolutely notice. And then there was the now-familiar modern internet side quest: “So tired of these AI generated articles.” Ouch.
So the mood was a mix of performance panic, engineering I-told-you-so energy, and readers turning the comments into a mini trial about whether tech writing itself has become unreadable sludge. In other words: a classic week online, where the bug report is only half the story and the comment section steals the show.
Key Points
- •A customer’s Envoy upgrade increased proxy CPU usage by roughly 20%, and version bisection linked the change to Envoy v1.34’s default switch from nghttp2 to oghttp2.
- •The article says users had reported 15–25% latency regressions after the switch, and Envoy v1.37.0 later restored nghttp2 as the default.
- •Benchmarks across Intel Sapphire Rapids, AMD Zen 4, AWS Graviton4, and Google Axion showed nghttp2 outperforming oghttp2 by 15–25% RPS/core on header-heavy proxied traffic and 7–19% under heavy connection churn.
- •The investigation found that both codecs spend most of their time on HTTP/2 header decompression rather than other proxy tasks.
- •The article identifies HPACK processing, especially Huffman decoding of literal header values on fresh connections and unique headers, as a major hotspot in proxy workloads.