Steven L

1: Ah 🤦 good point. We could make a new buffer but that'd be kinda pointless (either a bunch of zero valued things which are the same as an empty...

@rabbbit maybe... segmented buffers in particular might be reasonable, since it can leave the larger objects un-claimed in the pool until a large response begins (we don't stream anything, so...

Well 🤔 After sleeping on it a bit, _would_ it be essentially solved if we sharded by size (this seems important regardless) with a reasonable cap (64KB like `encoding/json`?) and...

> You mean, we don't return anything from the pool, until someone returned data to the pool? I probably not fully follow you here. I mean something like this sequence:...

If the changes prove helpful: I might suggest something similar go into Zap, as we occasionally see fairly high memory use in its pool too, and it follows the same...

Well. Happened again, and this time the outbound-unary proto bufferpool was involved too: Our inbounds were doing the same thing: But we also had this outbound stack taking up a...

Excellent find, thank you! We've been attempting to figure out a(?) leak for a while, but all the code it has been hitting internally have been *incredibly* complicated and nearly...

Ah. This one is basically known, and kinda sorta "intended". What's happening is: - 0s: parent starts with an execution timeout of 1m - 0s: child starts with an execution...

>I was hoping to catch a TimedoutError and then perform the cleanup in a detached workflow context. You can, you just can't do _anything_ after your execution timeout - it's...

Working on this! I'm not entirely sure how to verify it on both arm-without-rosetta (it's not really uninstallable) or osx-on-intel (I don't have one), but #6581 drops the `arch ...`...