cool! :) and I know too well how it works with this getting carried away stuff ;) by the way, wouldn't you by any chance fancy getting carried away on some kind of udev events instrumentation/logging? ;P Back to the matter - in the meantime, I've actually also advanced with the stuff a bit, so I'll try to confront my solution with yours when I get back home later today, and reply with my thoughts on the PR (or send a competing PR ;P so that stuff can again grind to a halt in discussions and bikeshedding, eh :/ )
Still, just a few quick notes now, replying to what you wrote above in the email: - as to the graph, I've found out that systemd-bootchart is kinda broken in this regard; I've got a patch, I'll explain more in comments to the PR, and also obviously will try to send it upstream. In short words, the tool currently works correctly in this regard only if CONFIG_SCHED_DEBUG=y too. - as to the kernel option(s): in my personal opinion, I think it could be OK if the flags were applied to the kernel *only* if bootchart was enabled as NixOS *module*, i.e. xyz.systemd.bootchart.enable = true; or something like that. IIUC, this would have to put burden of recompiling the kernel (something like 1h+) on user (no Hydra support), so I think there should probably be a big warning/note about that in the module's/option's meta information? Cheers! :) /Mateusz. On Wed, Jul 13, 2016 at 2:51 AM, Nikolay Amiantov <[email protected]> wrote: > I've looked into possibility of plugging this, got a little carried away > and in the end implemented most of it [1]. I'm sorry, I don't generally > like taking fun implementing things away from people (helping is better > since it provokes learning)! > > However, this is not finished yet -- generated SVGs are suspicious > because they contain no processes for me. It would be cool if you'd > finish testing this because my motivation has worn off (initially I > wanted to measure impact of different splash screen service > configurations on boot times). > > However I'm happy to help, please ask questions regarding this code in > the PR! I've tried to make it as straightforward as possible. > > Note that this requires a controversial change in our standard kernel > config, which means both you'd need to recompile your kernel and we'd > need to get a green light from people responsible for kernel config > decisions to get it mainlined. > > 1: https://github.com/NixOS/nixpkgs/pull/16902 > > On 07/12/2016 07:09 PM, Mateusz Czaplinski wrote: >> I'm mostly interested in analyzing what happens during systemd. I seem >> to understand systemd-bootchart should provide somewhat more detailed >> analysis of what's going on - e.g. like in >> https://wiki.apertis.org/Docs/System_startup#systemd-bootchart. I find >> the basic "systemd-analyze plot" somewhat too opaque for the noob I >> am... to tell the truth, I'd love to collect as much information as >> possible from wherever possible. E.g. some udev-related stuff looks >> highly suspicious in my plot (I have long red bar for my disk); but I >> couldn't find yet if/how it is possible to log the timing of udev >> events during boot... So, was hoping bootchart could maybe help me at >> least in some way...? >> >> I'd be very grateful for any hints on how to plug systemd-bootstrap at >> any stage/point up to systemd start. I'm OK with patching nixpkgs (and >> `nixos-rebuild -I nixpkgs=~/nixpkgs switch`); anything is ok, though >> I'd sure be extra grateful for "style/approach" guidance which could >> result in a pull request that could be useful/make life easier for >> future users. >> >> /Mateusz. > > -- > Nikolay. _______________________________________________ nix-dev mailing list [email protected] http://lists.science.uu.nl/mailman/listinfo/nix-dev
