Alex Valiushko

It has been discussed previously I believe, with the suggestion to implement `PeekAll` that returns a slice. For context, go stdlib merges headers since per HTTP spec `K: v1\r\nK: v2`...

> We may want to either generate an erlang_ls.config file My bad! I should've clarified, the schema is meant for the client config. Here's a snippet from nvim docs: ```...

**Jinja** Docs (v3): https://jinja.palletsprojects.com/en/3.1.x/ Implementations (WIP, v2): - https://github.com/dbt-labs/tree-sitter-jinja2 - https://github.com/theHamsta/tree-sitter-jinja2 Sample Vim plugins (likely v2): - https://github.com/lepture/vim-jinja - https://github.com/glench/vim-jinja2-syntax

And a note from @thuandb (queue depth mistmatch because of a different env): > It is confirmed that IO is the bottleneck causing high memory utilization during queue sync. >...

My test was run from a somewhat stale master, at e2edb1b. In prod this was observed on 3.8.22 and 3.8.27.

Updated description to remove the mention of allocs: I haven't actually looked at allocations yet. Memory may sway because of paging.

So in total, two reports about this still happening in CMQv2? @johanrhodin Looking forward to reproduction steps as well.

@mkuratczyk your branch shows significant improvement across all my tests. I can still see the spike, but it's much harder to cross into alarm now.

Not too rigorous, but I've been running various sync scenarios over the last couple days: cluster member loss, DLQ, lazy and not, various message sizes. Tests show similar or better...

Fantastic addition. Any chance to backport it into earlier versions?