January 13, 2026
RSS is sus; GC swears it’s fine
Deconstructing the LuaJIT Pseudo Memory Leak
Dev promises a 'no-more-crashes' fix; commenters cry AI fluff, ask who's paying
TLDR: LuaJIT-plus claims to fix apps that look clean inside but keep grabbing memory until they crash. Commenters slam the write-up as vague, demand real code and benchmarks, and even speculate on Cloudflare funding—turning a technical fix into a transparency and trust showdown.
LuaJIT-plus swoops in promising to stop apps from secretly hoarding memory, the kind that makes servers look fine inside but balloon on the outside until Kubernetes kicks them out. The pitch: reclaim memory pages so the operating system actually sees things calm down. But the crowd isn’t buying hype without receipts. One voice blasts it as “AI slop” and demands plain-English specifics: if the garbage collector doesn’t rearrange memory, you’ve got to free whole pages, not just tidy the closet. Another commenter digs into the commit history and drops a spicy “Who’s paying?” bomb, hinting Cloudflare might foot the bill for a mysterious “mtrim” patch. Cue memes about “RSS” being “Really Sticky Stuff,” jokes that the app’s trash is “clean” while the house is still full, and Kubernetes “yeeting” pods when memory spikes. The split: optimists want a real fix for those stair-step memory graphs; skeptics want code diffs, benchmarks, and a clear explanation of how LuaJIT-plus actually returns memory to the OS. Until we see the bytes, the drama is bigger than the leak—because in this thread, transparency beats buzzwords every time.
Key Points
- •High-concurrency OpenResty/LuaJIT services can show steadily increasing RSS despite healthy Lua VM GC metrics.
- •The article characterizes this behavior as a “pseudo memory leak,” not caused by retained references.
- •LuaJIT’s default allocator retains memory pages, leading to internal fragmentation and decoupling of GC from OS RSS.
- •Operational impact includes Kubernetes Pods being OOM killed when resource limits are exceeded.
- •LuaJIT-plus is presented as an enhanced runtime with proactive memory reclamation to create a predictable, “breathing” memory model.