Vitaly Aminev @ Vancouver 🇨🇦

Eng @phantom Ex @streamlayer, Founder @Makeomatic Software Architect. Speaker

[Matteo Collina @matteocollina]

🎟️ NodeConf EU 2026 is in Bologna in 3 weeks, and tickets are running low.

Sep 29-30. The Savoia Regency, an 18th-century villa set in a 10,000m² park, 5km from the city center. Pool between sessions, Emilia-Romagna food, and two days of real Node.js depth.

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@yagiznizipli semaphoreci is pretty good, but there isnt much of a variety - containers mainly

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@jasnell @yagiznizipli @nearform @NodeConfEU @Cianomaidin 4 year old (sized for 6 yo) and 1 year old (prob sized for 18 months at the time) here :D

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@juristr @t3dotgg Wouldnt night mode / true tone / other software and/or hardware specific alrerations play a role here? Anecdotally this test seem biased - too many of my connections on twitter report similar results

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@yagiznizipli @Cloudflare @nodejs congrats!

[Matteo Collina @matteocollina]

.@NodeConfEU reunion at @ThisNextThing!

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @ThisNextThing Cant wait :)

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@yagiznizipli Ikea ones are surprisingly good based on my research

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina What are the dates? :)

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@jasnell Both kids are coming this year :)

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @_rafaelgss @yagiznizipli Crash referenced that is, not the leak, worded that poorly

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

#nodejs @matteocollina @_rafaelgss @yagiznizipli any suggestions on how to make gc aggressive? http/2 + nghttp allocates a lot of memory for each stream on http2 session and what I'm seeing is that rss/heap grows fast, but no leak, small old space helps, but still

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @_rafaelgss @yagiznizipli https://t.co/EsG3tmoaPc thats what it was apparently, heh, never realized it was an actual bug

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @jasnell thanks for the help with debugging http2 :) jemalloc did the trick - no leak, just crazy memory fragmentation, which made it look like a leak. do you think it's worth properly documenting this somewhere?

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @MicheleRivaCode startup visa or federal skilled worker programs are great

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @jasnell @_rafaelgss @yagiznizipli well, osx doesnt behave well enough, BUT, patched jemalloc@5 into node through patchelf on alpine container, and, magically rss drops down to normal levels even after 300k+ streams on several connections. no idea why fragmentation wasn't the case earlier - usage was similar

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@matteocollina @jasnell @_rafaelgss @yagiznizipli So far the best guess is there is no leak, but rather very annoying memory fragmentation. RSS keeps growing while heap stays low, so it looks like a leak, but it is not. Will try out jemalloc to see if that helps

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@MicheleRivaCode Would be cool to catch up at some point :) If you fancy some snowboarding or skiing - might be fun?

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@jasnell @matteocollina @_rafaelgss @yagiznizipli And with 1k RST streams there is no goaway/reset of the connections that follows

[Vitaly Aminev @ Vancouver 🇨🇦 @v_aminev]

@jasnell @matteocollina @_rafaelgss @yagiznizipli tcp dump, UV_EOF triggered before 258 -> node sends GOAWAY -> streams continue closing -> client sends FIN, node sends FIN. So socket is operational, yet UV_EOF was received. Should it be handled differently? I assume it happens because of loop overload (10k RST frames)