mike's questions

We've got haproxy running in non-daemon mode, with DJB's supervise (a.k.a. daemontools) script keeping an eye on it.

However, as our site traffic has grown, its single thread has reached 100% CPU. We'd like to start using haproxy's nbproc option, to get it using multiple cores. However, this requires haproxy to be in "daemon mode".

Is there a way to make haproxy work under supervise while its daemon mode is on? If so, will it properly respond to "svc -d" commands?

And if there is no way, have you guys found it to be reliable enough that we won't need supervise anymore?

I usually start Postgres with pg_ctl -D /db start

How can I make it start under "nice", so it runs at a lower priority?

Back in the olden days, runlevel 3 was text mode and runlevel 5 was GUI mode. So if you wanted Linux to boot up in text mode, you would just set 3 as the initial runlevel and you're done.

Nowadays it seems that things are more complicated, and since I don't want to break anything, I ask:

What's the right way to get Ubuntu 9.10 to always boot up in multiuser text mode?

Here's my /etc/dhcp3/dhclient.conf:

request subnet-mask, broadcast-address, time-offset, routers,
        domain-name, domain-name-servers, domain-search, host-name,
        netbios-name-servers, netbios-scope, interface-mtu;
timeout 60;
alias {
  interface "eth0";
  fixed-address 192.168.1.222;
}
lease {
  interface "eth0";
  fixed-address 192.168.1.222;
  option subnet-mask 255.255.255.0;
  option broadcast-address 255.255.255.255;
  option routers 192.168.1.254;
  option domain-name-servers 192.168.1.254;
}

When I run "dhclient eth0", I get this:

There is already a pid file /var/run/dhclient.pid with pid 6511
killed old client process, removed PID file
Internet Systems Consortium DHCP Client V3.1.1
Copyright 2004-2008 Internet Systems Consortium.
All rights reserved.
For info, please visit http://www.isc.org/sw/dhcp/

wmaster0: unknown hardware address type 801
wmaster0: unknown hardware address type 801
Listening on LPF/eth0/00:1c:25:97:82:20
Sending on   LPF/eth0/00:1c:25:97:82:20
Sending on   Socket/fallback
DHCPREQUEST of 192.168.1.27 on eth0 to 255.255.255.255 port 67
DHCPACK of 192.168.1.27 from 192.168.1.254
bound to 192.168.1.27 -- renewal in 1468 seconds.

I used strace to make sure that dhclient really is reading that conf file. Why isn't it paying attention to my "fixed-address 192.168.1.222" line? Why is it doing a DHCPREQUEST for 192.168.1.27 instead?

I've got a logfile in the standard syslog format. It looks like this, except with hundreds of lines per second:

Jan 11 07:48:46 blahblahblah...
Jan 11 07:49:00 blahblahblah...
Jan 11 07:50:13 blahblahblah...
Jan 11 07:51:22 blahblahblah...
Jan 11 07:58:04 blahblahblah...

It doesn't roll at exactly midnight, but it'll never have more than two days in it.

I often have to extract a timeslice from this file. I'd like to write a general-purpose script for this, that I can call like:

$ timegrep 22:30-02:00 /logs/something.log

...and have it pull out the lines from 22:30, onward across the midnight boundary, until 2am the next day.

There are a few caveats:

• I don't want to have to bother typing the date(s) on the command line, just the times. The program should be smart enough to figure them out.
• The log date format doesn't include the year, so it should guess based on the current year, but nonetheless do the right thing around New Year's Day.
• I want it to be fast -- it should use the fact that the lines are in order to seek around in the file and use a binary search.

Before I spend a bunch of time writing this, does it already exist?

It seems that my logfiles generated by syslog are getting rotated twice each day. I believe this is due to both /etc/cron.daily/logrotate and /etc/cron.daily/sysklogd doing overlapping work.

Here are the relevant lines from the sysklogd script:

logs=$(syslogd-listfiles)

test -n "$logs" || exit 0

for LOG in $logs
do
   if [ -s $LOG ]; then
      savelog -g adm -m 640 -u ${USER} -c 7 $LOG >/dev/null
   fi
done

What are the best common practices for dealing with this? Should files generated by syslog not be listed in logrotate? What if I want to use some of logrotate's more advanced features, like postrotate scripts -- should I try to make /etc/cron.daily/sysklogd do it, or should I comment out those lines and let logrotate do it?

I prefer to install my own version of Firefox. How can I tell apt to never attempt to upgrade it, even if i type "apt-get update ; apt-get upgrade" later?

Edit: Okay, so dpkg -l firefox prints this:

ii  firefox   3.0.3+nobinonly-0ubuntu2   meta package for the popular mozilla web browser

So I added this to my prefs:

Package: firefox
Pin: version 3.*
Pin-Priority: 1000

But apt-get upgrade still prints:

The following packages have been kept back:
    (nothing related to firefox)
The following packages will be upgraded:
    (blah blah blah)
    firefox firefox-3.0 firefox-3.0-branding firefox-3.0-gnome-support 
    firefox-gnome-support
    (blah blah blah)

I also tried making the package "*firefox*", but that didn't work either.

I like Ubuntu's NetworkManager -- it does a great job keeping track of all my WiFi parameters and passwords. However, when I actually plug in an Ethernet cable, the frustration begins.

When hooked up to my work LAN, I'd like to force a particular domain search path. But I can't just edit resolv.conf, since NetworkManager will always overwrite it. And I can't change the settings on the DHCP server.

As a total hack, I set up a script to run whenever the network is brought up. It looks at resolv.conf, and if it's the one from the work LAN, it overwrites it. However, this doesn't work well enough -- it seems that periodically resolv.conf will get overwritten by something that doesn't trigger my script.

Is there a less hacky way to do this? Or even a more hacky way, provided that it actually works?

Possible Duplicate:
Can I nohup/screen an already-started process?

On Unix (specifically, Linux), I've started a job in a regular ssh->bash session. I'd like to leave work soon, but I now realize that the job is going to take several hours.

If I had just started this job in screen, I could detach and go home. But I didn't.

Is there any way to disconnect the job from its ssh session, so I can shut down my computer, (dropping the TCP connection and killing the ssh session), and yet have the program keep running?

I don't care about its output -- in fact, I redirected stdout to a file. I just want it to run to completion.

I have a database (it's Postgres, if that matters) and I would like to take two of its tables, export them, and then import them into another database. How do I do that?

Edited to add Google bait: dump load

We have a role account at work that has a pretty big crontab. Its MAILTO is pointed at a shared address, so that a number of us get notified if something fails.

I'd like to add an entry to this crontab, but I only want myself to be notified if something goes wrong. Is there a way to change MAILTO for this one entry, or otherwise accomplish my goal?

Let's say I have a quad-core box and four identical processes, each with ten threads. Is it possible, in Linux, to say that Process A is only allowed to run on CPU 0, Process B is only allowed to run on CPU 1, etc?

It's well-known that you should never fsck a mounted partition. I can understand how this could easily lead to corruption if the filesystem is written to by fsck (e.g., the -a option is used), but why can't read-only checks be run on mounted disks?

How do I determine the block size of an ext3 partition on Linux?

As I understand it, each file on a Unix-like operating system has an inode number (which can be viewed with "ls -i"), and each inode is a list of disk blocks that contain the actual data of a file.

Is there a Linux command which takes a filename as its argument and prints out the list of disk blocks that that file's inode points to?

P.S. The filesystem in question is ext3.

When I click on a *.json or *.patch file in Firefox, it asks me what external application I want to view it in.

However, I don't want to use an external application. It's plain text -- I just want to view it in-browser.

There doesn't seem to be anything in the UI for this.