Twitter | Search | |
Chris Siebenmann
That cks. Overcommitted sysadmin, photographer, bicyclist, and other multitudes. I write a lot of words for a programmer.
10,938
Tweets
220
Following
1,595
Followers
Tweets
Chris Siebenmann Aug 9
Replying to @glev2005
We haven't currently noticed any (and we are using it by default), but we haven't looked. The last time I checked the code, file level prefetch could misfire in significant ways if your ARC was overloaded or too small; it'd prefetch then throw away a lot of stuff.
Reply Retweet Like
Chris Siebenmann Aug 9
Replying to @zekjur @lathiat
Our problem is that the dependencies aren't in the current repos either, because we're trying to install a specific kernel, which that requires a zillion packages, so we'd like to only have to put one or a few top level packages on the command line.
Reply Retweet Like
Chris Siebenmann Aug 6
Replying to @lathiat
Installing a kernel needs a whole bunch of packages between specific versions, generic meta-packages, headers, modules, etc etc. I would really like not to have to fish them all out one by one and dpkg -i the lot of them.
Reply Retweet Like
Chris Siebenmann Aug 6
Replying to @lathiat
The challenge today is installing a specific older kernel that we have copies of in our Ubuntu apt mirror but that is not in the official Ubuntu repos, so is not in the upstream package metadata that our mirror serves.
Reply Retweet Like
Chris Siebenmann Aug 6
It has been '0' days since I've wound up hating Debian's choice to sign package metadata instead of packages (or perhaps 'in addition to' these days). Why? Because it makes it much more difficult to support 'install a package, satisfying dependencies from this directory of debs'.
Reply Retweet Like
Chris Siebenmann Aug 6
Replying to @thatcks
Dear Fedora: If I have to resort to bodhi to get the latest stable main repository version of packages, DNF has failed *spectacularly*. If these packages had security updates, there would be real issues here.
Reply Retweet Like
Chris Siebenmann Aug 6
Replying to @thatcks
At this point I would like to burn DNF modules to the ground as a failed idea and a failed implementation. Having the 'latest' release of something is clearly a lie.
Reply Retweet Like
Chris Siebenmann Aug 6
It turns out that Fedora 31 DNF modules, even reset ones that aren't active, are apparently still both out of date and blocking package updates. Modular ripgrep is 11.x, current main repo ripgrep is 12.1.0, but good luck getting that.
Reply Retweet Like
Chris Siebenmann Jul 31
Replying to @thatcks
Since I'm me I felt compelled to modify the Prometheus exporter I found a bit. People almost never put all the data in metrics exporters that they could, and sometimes they don't put it in formats that are easy to use in graphs, alert rules, etc.
Reply Retweet Like
Chris Siebenmann Jul 31
My UPS is now in use and of course I grabbed a Prometheus metrics exporter for it (that pulls data from the UPS's Linux software). I'm not sure I believe some of that data, but I'll see. (I suspect that the 'load percentage' is subject to imprecision that can take it to 0.)
Reply Retweet Like
Chris Siebenmann Jul 29
Sometimes work involves bolting barn doors in case of future horses, even if we hope they never show up. But they might, so better safe than sorry.
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @oclsc
You can be 'After=' or 'Before=' on a service that isn't enabled or doesn't even exist, which is handy sometimes if you're trying to create generic .service units that don't care what's installed on a particular machine but that do have ordering requirements.
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @oclsc
Requires= vs Wants=, really. After= is just ordering; Wants= tries to start the service too. Although in this case After= alone would be sufficient since Postfix is already enabled.
Reply Retweet Like
Chris Siebenmann Jul 28
It has been '0' days since I shot myself in the foot with a systemd 'Requires=' clause. As usual, the foot shooting was because if systemd restarts your required unit, yours restarts too. (In this case I had 'Requires=postfix.service' and no, not really.)
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @oclsc
The SmartRG SR505n is what I have (and am happy with), but I'm not sure it's still readily available new (I was just looking). Works fine as a bridge/modem and a wireless AP.
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @oclsc
Out of curiosity, which VDSL modem did you get? My VDSL modem is working fine so far but it's also ancient, or at least five years old apparently, and in these WFH times that may make me slightly nervous.
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @_r_o_n_e_
I want something to just change the /etc/passwd entry without insisting on doing other work (because we have other processes for it). I think FreeBSD pw(8) may work that way, although the manpage isn't clear.
Reply Retweet Like
Chris Siebenmann Jul 28
Replying to @oclsc
There are all sorts of error conditions involved in scripted modifications of /etc/passwd as text, instead of structured data. You'd better be really careful about regexps.
Reply Retweet Like
Chris Siebenmann Jul 28
I really wish Linux had a good way to change a login's primary group without running chgrp over all their files (usermod insists on always chgrp'ing, yes it's tempting to build my own version with that taken out).
Reply Retweet Like
Chris Siebenmann retweeted
Warner Losh Jul 27
So when did get reboot? How is it related to the old 'sync sync sync halt dance'? Why is the answer not as clear as one would like? Read about it in my latest history blog.
Reply Retweet Like