| Summary: | sys-apps/openrc patch (for 0.34.11) to decrease openrc starting time | ||
|---|---|---|---|
| Product: | Gentoo Hosted Projects | Reporter: | nobody <noreply> |
| Component: | OpenRC | Assignee: | OpenRC Team <openrc> |
| Status: | RESOLVED FIXED | ||
| Severity: | enhancement | ||
| Priority: | Normal | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| See Also: | https://github.com/openrc/openrc/issues/85 | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Attachments: | caching patch | ||
|
Description
nobody
2018-03-23 10:49:12 UTC
Created attachment 524962 [details, diff]
caching patch
In case it's not obvious, the patch change the bet to (time cpu cycle + disk access to generate deptree > time to copy the file and reload the deptree from the cache) AntP point me to /etc/init.d/savecache that is suppose to do that, and /lib/rc/sh/init.sh that restore /lib/rc/cache content on boot. Actually, yes, it's all there, and working, i made tests and if you let savecache saving the deptree file, init.sh will restore it when booting, and openrc will re-use it. Result is then the same as my propose patch, except implementation is trivial now as we only need to tell savecache to also backup deptree :) diff -u savecache.old savecache --- savecache.old 2018-03-25 17:58:19.564654706 +0200 +++ savecache 2018-03-25 17:57:52.284995748 +0200 @@ -49,7 +49,7 @@ fi ebegin "Saving dependency cache" local rc=0 save= - for x in shutdowntime softlevel rc.log; do + for x in deptree shutdowntime softlevel rc.log; do [ -e "$RC_SVCDIR/$x" ] && save="$save $RC_SVCDIR/$x" done if [ -n "$save" ]; then I made my tests for savecache changes with openrc-0.35.5 My concern about caching dependencies, and the whole savecache service for that matter, is it assumes that / is writable. This is not always the case, that's the primary reason I generate the dependency tree on bootup instead of caching it on shutdown. (In reply to William Hubbs from comment #5) > My concern about caching dependencies, and the whole savecache service > for that matter, is it assumes that / is writable. > > This is not always the case, that's the primary reason I generate the > dependency > tree on bootup instead of caching it on shutdown. If this is the only concern you could test for it or make this only optional. But if we provide a change like this we have to communicate that from now on, user have to call "rc-update --update" on their own after doing changes to /etc/init.d or /etc/conf.d. They can no longer assume that we will pick up the changes. (In reply to William Hubbs from comment #5) > My concern about caching dependencies, and the whole savecache service > for that matter, is it assumes that / is writable. > > This is not always the case, that's the primary reason I generate the > dependency > tree on bootup instead of caching it on shutdown. There's no / writable issue at all. If / is RO on shutdown, then you cannot write the cache file, no change If / is RO on boot, you will still be able to copy it from /lib/rc/cache location to the "always writable" /run/openrc If / is RO on shutdown an a "dirty/bad" cache is already there, it's more a kind of racing issue (because / was RW before) and openrc has a dirty and rebuild detection system and will build a new one So the real major issue is if someone shutdown / in RW, and later keep shutting down / in RO and latest cache file couldn't be write in /lib/rc/cache as it will always make openrc try use it, mark it dirty, and build a new one (which is worst of the worst case performance). (In reply to Thomas Deutschmann from comment #6) > But if we provide a change like this we have to communicate that from now > on, user have to call "rc-update --update" on their own after doing changes > to /etc/init.d or /etc/conf.d. They can no longer assume that we will pick > up the changes. Even that, while of course it would be "sad" if they aren't aware of it, actually not updating it would just not enable the feature which just gave 100% of what openrc is doing today :) I agree users should be warn to rc-update to benefit from it, but there's no real issue if someone doesn't. See the issue I added to see-also for more info on why we need to always regenerate the deptree on boot. I have added a patch to 0.36 to save the cache again. However, we always have to regenerate the cache when we boot because of things like the related issue in see also. https://github.com/openrc/openrc/commit/f0ad6473 |
