3 ms·
Ops person here. When you set up a node, you set up all its networking, its services, its packages, its configuration, etc. It doesn't matter what your technolo
by peterwwillis 7y ago
Ops person here. When you set up a node, you set up all its networking, its services, its packages, its configuration, etc. It doesn't matter what your technology is - you set it up to do whatever it needs, and then hopefully you never have to re-do that again. Systemd's hydra of services affected a lot of that, so we had to learn all the weird gotchas to get newer distros to work properly.
And it's not just a little different, it's more complex. This increases the number of bugs, the time to test and roll out changes, integration work, etc. So the process is more complex. And the operation is also more complex, meaning more training for anyone who's used to how things worked for 30 years. Doing minor ops work now requires looking at 5 different manuals, requiring extra software, adding configuration management where it wasn't before, changing support instructions, healthchecks/automation tasks, migrations to newer disk images while supporting legacy functionality, etc.
And here's the part that really kills me. Alternatives to all the parts of systemd already existed; if you really needed a better system, you could pick up any similar software and make it work! But then distributions start rolling this out en-masse for all new releases, and we don't have a choice to opt-in. It was a breaking change that did nothing for us that we didn't already do for ourselves, even requiring more work than before.
Do our systems boot faster now? My EC2 instances still seem to take forever to spin up, and physical machines still take ~5 minutes to POST, so I'm going to say no. Can we do new things that we couldn't before? No. Are our systems more stable now? No. Are they less complex? No. So while some people talk up the theoretical benefits, I as an ops person have not seen them. But I still have to look up man pages to remember the three different filesystem trees systemd keeps files in, or the wonky configuration files, or what 5 extra commands I need to get something done.
If you're going to make breaking changes, at least have it make my life easier after I get used to it. Systemd has not done that. And it seems to just keep getting heavier.
- pard68 7y agoOur bottle neck is often pid 1 for boot time. We don't do mem check on boot so that saves us 1 second * Gb_of_RAM in boot, which is why most BIOS take forever on a server. We also don't run anything in PCI slots so we have the PCI checker disabled too (all our servers are Dells they have some software you can license to config the BIOS). I'm new to the game, I only vaguely remember a time when Ubuntu wasn't on systemd. I don't have 30 years of experience with any init system, but I do have 10+ years with systemd. I didn't use another init until I got this job and had to learn sysv. So I'm coming from the exact opposite, it sounds like, place of you. To me, everything you said is 100% true, but about sysv and its line. I also don't provision too many servers these days because with the exception of a few things that were deemed "to important" all our servers have automated build scripts to go from 0 to prod.