Hacker News Reader

No more easy editing/understanding grub.conf

True, the grub2 config is horrible.

No more easy to understand /etc/init.d

It's a bit more complicated, but not too bad:

Instead of "/etc/init.d/thing restart" you type "systemctl restart thing"

Instead of chkconfig --list, you type "systemctl list-dependencies"

Writing the equivalent script with systemd is much cleaner with less hacks, particularly for launching as different users and doing locking.

No more text log files for system log (journalctl instead)

By default. It's fine once you get the hang of the new syntax:

journalctl --since=today --follow

Watch out for default XFS filesystem instead of EXT4 because it is slower in real world use for databases, etc.

Depends on the workload. Speed is only one part of it. For some benchmarks by phoronix see http://www.phoronix.com/scan.php?page=article&item=linux_315...

I'll wait for more benchmarks to be certain but this database test of the 3.10 kernel (which 7 uses) of XFS vs EXT4 is not promising:

http://openbenchmarking.org/prospect/1305166-UT-FILESYSTE20/...

A 160 GB disk is not a real comparison for enterprises. XFS really starts to perform better on disks 1Tb as well as 8 cores and above. EXT4 really starts to creak when moving to filesystems that are 16TB and above. Something that is going to be common in the 7 years that Cent-OS 7 is around.

Of course with the amount of backports of patches that any RedHat kernel has the comparison to mainline version numbers is almost useless :(

For my workload the performance difference is 15% better for XFS than EXT4 on the same 3Tb of SSD with the same workload.

RHEL pretty much is the standard for enterprise-y distros. I'm pretty confident they weighed the XFS decision very heavily, and in the end chose it for good reason.

In fact, here's an article on the SUSE blog from a year ago discussing why XFS is a great choice for enterprise:

https://www.suse.com/communities/conversations/xfs-the-file-...

systemctl: I don't think anyone cares about a change of command name / syntax, it's the removal of flexible, small, readable, plain text files that is the problem. That the underlying init system of linux no longer follows the linux philosophy boggles the mind.

I have to disagree. I've written some stupidly big init.d scripts in my time. Replacing them with:

    [Unit]
    Description=My dumb script
    
    [Service]
    Exec=/usr/local/bin/dumb-script

And no longer caring about PIDs, forking, etc. has been very refreshing.

I agree, however, that systemd feels too monolithic, and I question the wisdom behind journald.

Yes they reinvented some text files as the NT event log...

TBH the first thing I do is ship the logs off the box so journald seems a waste of space.

You don't like systemd but it's probably time to accept the fact that it has "won". Most major distros including Debian is switching to it so it will get harder and harder to avoid. Might as well make the effort to learn it now.

Technically I don't have to accept it until 2020 ;-)

(and FreeBSD has declared they will never use it, which to be fair is easy for them to declare since they cannot use it because it requires posix) (correction: not posix, cgroups and udev)

You don't really ever have to accept it. Go grab the sysvinit source and run it. Have fun managing everything yourself. Ditto for copying a simple rc system like Slackware or OpenBSD provide. For that matter, throw out everything, declare dependencies and solve the problem with prolog. The thing is, once you deviate from the norm on linux, you just have to understand things at a deeper level. Not impossible, but IMO maintaining your own custom distro gets old after awhile.

It requires Linux-specific kernel APIs like cgroups and udev, not POSIX.

Most *BSD systems are mostly or fully POSIX compliant.

Well, I did mean Linux when I said "distros". :P

You can enable text log files by installing 'rsyslog'. You only have to install the package, no further configuration is needed.

Would the server then do two writes, both to the binary log for systemd and then to rsyslog?

I suspect so, because systemd needs it's logs for internal analysis?

And you should probably make sure that you do since every existing log analysis tool requires text logs.

Also, remote syslog, which journald doesn't support.

I'm afraid you're mistaken, please don't spread misinformation:

http://www.freedesktop.org/software/systemd/man/systemd-jour...

I was going on information from Red Hat that RHEL7 can't do that. Turns out, Red Hat was right, RHEL7 can't.

systemd-journald-remote was added with systemd 212, RHEL and CentOS ship with 208, so no, journald on RHEL7/CentOS7 can not do remote logging.

I also like completely ignoring how journald breaks every log analysis tool.

touche, you are correct. I'm sure it will be added as a backported enhancement at some point, but installing rsyslog is a yum install away and solves both needs.

> No more easy editing/understanding grub.conf

Oh, that's beautiful. And if you want to get s..t done, you can't even quickly search the manpage, because for religious reasons, GNU projects use that weird (and traditionally utterly verbose) info system. But hey, GNU is for GNU's Not Unix, so we shouldn't be too surprised ...

It's not like I didn't appreciate what the GNU project has done for the free software world, but it's so annoying that they seem to be focused more on religious matters than on writing good (well-documented, easy-to-configure, non-bloated) software.

I'm doing more and more server stuff with the BSDs, simply because the docs are so well-written and if I want to understand some part of the system, it doesn't feel like there are numerous layers of abstractions hiding away the interesting parts from me.

XFS and its lack of data journaling (it only journals its own metadata) is notorious for corrupting data on power loss and kernel crashes. It performs well only under specific workloads and hardware setups.

Why choose it as the default filesystem in an enterprise-oriented distro?

XFS is a dealbreaker for me. Nothing but bad experiences with that particular FS, especially when the filecount goes up.

Note that of course you can still choose to use EXT4 when installing 7

Ah, that's just the default. I see. I was wondering what they were smoking there that makes a lot more sense. It's a very peculiar choice, EXT4 is pretty good for most use cases, when you need XFS there are few alternatives but the performance issues make it a love/hate relationship and I really would not want that on my machines by default.

One problem with colocation is that a lot of the hosting providers I work with simply do a default re-image which will use whatever the OS comes with as standard. So chances are those will default to XFS and good luck getting rid of it then. (No phyical access to the machines.)

Not to be an apologist, but you can work around the performance issues in XFS, and a good amount of them were patched up in recent kernels. Disabling write barriers, delayed logging, and an independent journal device make it as fast (or faster) than ext4. But ideally you should stick to what you know and have already tuned unless you need some feature from XFS.

So use ext4, it isn't as though ext4 isn't included :)

Thanks for posting this. That's what i was expecting.

Does any of that really improve my day?

Nope. In fact most of it makes it worse and adds friction. More cogs turning in different directions.

Will keep our devops stuff on CentOS 6.5 for at least another 4 years...

Will keep our devops stuff on CentOS 6.5 for at least another 4 years...

What if someone needs one of the newer packages provided by RHEL7?

What if I don't?

Most of our stuff is on JVMs apart from SVN/DAV/apache/LDAP and that's a 40 min job to get packages ready for from srpms from scratch. We just have our own repo and import stuff that isn't in the distro repo ourselves.

FreeBSD looking like a good option in the future.

They can use the software collections to install it.

http://developerblog.redhat.com/2013/09/12/rhscl1-ga/

> systemd replaces init.d

CentOS 6 had Upstart, right? Which is likely to become deprecated due to the fact that Ubuntu is also moving to systemd.

Sure, Red Hat are the ones pushing systemd, but some disruptive change away from Upstart was inevitable.

I think its a RC of CentOS 7 because of "POTENTIAL" word .

As excited as I am for 7.0 I highly dislike the systemd and grub changes. I'll wait for benchmarks before I decide to go to 7.0 or fully switch my stack to Debian

debian is also switching to systemd

and using grub2