@bortzmeyer 10% de la sagesse du monde est hébergé sur des pages free.fr

halfa boosted

The solution is always more elegant when you understand the problem

Thanks you for coming all the way speaking! That makes sense.. I'll need to experiment a bit. So far I looked at stuff with pmap -x but I'm not sure that the reported metrics are the same as the accounting done by cgroups. For context we are playing with an experimental k8s truly-single-binary build and, as the binary (~100MB) is used as init, I was wondering what would be the impact on the pod cgroup forge.tedomum.net/acides/hepto

halfa boosted
halfa boosted

"Is it okay if I log this once a minute from the agent?"

"You tell me if it's worth it - that's about 4.6 billion times a day across our fleet."

halfa boosted

When someone tells you to have a laser focus, please remember not to stare directly into the laser as that may cause a permanent lack of focus.

halfa boosted
halfa boosted

Everyone at #FOSDEM is touting their "single binary", I guess what we need now is a binary dating site

Small question for @cdown : how is shared memory managed within cgroups? If I have a big executable (let's say 100MB, because cool kids like static these days) that is executed from different cgroups, is that memory accounted for in each cgroups? Are the pages cached globally by the kernel? Do each process have it's own cache?

Show thread

I'm hyped for the next one, Chris Down is a good speaker and his cgroup talks are usually nice summaries for what is happening in the kernel

7 years of cgroup v2: the future of Linux resource control fosdem.org/2023/schedule/event

Replacing taggz images by squashfs (bonus points for the screen slides)

halfa boosted

Why resolving two names in a GUI program is hard ;
A summary of available name resolution APIs on Linux and why a new one is needed by Petr Menšík

getaddrinfo is blocking. getdns and adns are DNS-specific. systemd-resolved is systemd.

GUI applications want to wait for events, not to have blocking calls.

Show thread
halfa boosted

So, the parser is very fast but also very specialized. If IETF adds things to the zone file syntax, the parser is doomed.

github.com/NLnetLabs/simdzone

Show thread
halfa boosted

First step to speed parsing: drop Lex and Yacc (too slow and too general for DNS zone file parsing).

Next step, preventing the CPU pipeline to stall (bubbles - NOP - stall the flow.)

Show thread
Show older
Mastodon

Instance généraliste majoritairement francophone.