Cidrick posted:

Is anyone aware of a way to view the utilization of a remote NFS mount without actually mounting it? showmount only appears to show you what exported volumes are available to be mounted, and things like nfsstat and nfsiostat give all sorts of interesting metrics that don't really help me. A good old "df" will show it, but it requires mounting, which requires root.

Context: I'm trying to find a way for our monitoring environment to query a big NFS appliance that Doesn't Play Nicely With Others so we don't have the normal way of monitoring this stuff (SNMP, ssh, etc) that I would typically use.

Cidrick posted:

Yeah, I haven't actually looked at it, but maybe it's possible to set up a service account and use curl to POST a login and then scrape screen output or something.

I was hoping there was a way via an RPC command or something to remotely query info about an NFS export that I just didn't know about :|

ExcessBLarg! posted:

The main difference there is that when compiling, you're writing out a bunch of object files to stable storage. So yeah, you're introducing a considerable amount of iowait, since the filesystem will require that at least the journal transactions are written out between each file. Putting the build directory on a tmpfs will speed things up considerably, but at the cost of losing the entire build directory in a power event--which is generally OK.

However, with read-only workloads on a file-system mounted as noatime or relatime, there's nothing to write out, and generally the first time a file is read it's maintained in the page cache unless you're RAM starved. I suppose there's some seek patterns that are terrible with mechanism disks such that a sequential prefetch makes the first-time loading faster, but in most cases it's not worth doing.

dennyk posted:

Longinus00 posted:

If you have some other programming experience then don't feel like you have to do everything in shell script. Perl/python/etc are perfectly fine languages to do stuff like this if they're available.

Megaman posted:

Anyone?

Megaman posted:

Yeah, that was just a reply typo. There is no fdisk binary at this level of the installer, the disk is at /dev/sda and the system sees /dev/sda1, but i cannot mount with mount /dev/sda1 /mnt/ or anywhere else. It claims no such file or directory, whatever that means.

evol262 posted:

Nope. Raw disk is fine

evol262 posted:

I guess by this, a little clarification:

The idea that Oracle can support Linux (and basically RHEL) better than the engineers who write it is laughable, but they tell people this, then trolled our knowledgebase looking for support articles when customers they poached called them for support.

Patches come in a bundle available separately so Oracle actually has to backport their own stuff instead of trying to tweak around patches for backported fixes in the kernel SRPM. Incidentally, this is also why their "Unbreakable" kernel is no longer compatible -- they'd actually have to put in engineering effort to get it working on the EL kernel and shipping upstream is easier for them.

evol262 posted:

We don't license. You pay for a subscription, and some entitlements cost more.

But this is totally irrelevant flamebait anyway. It's possible if you run 4 socket systems on 2008 with no CALs, I guess. And don't open any support cases. Or if you get preferential treatment from Microsoft. Or if you have a lot more RHEL servers than Windows. Or you're paying for entitlements for hpc or clustering or whatever. Or it's on system z. There are a lot of variables here.

But a plain RHEL subscription is less than a 2012 standard license. If you're comparing RHEL+entitlements to 2012 standard, it's misleading. But it's not even funny how much cheaper it is than Datacenter, even with entitlements.

And especially if any of these are true and you're not getting discounted pricing. And it basically hasn't changed in years (because RHOS or RHEV or JBoss or whatever makes it up -- RHEL itself is less than 50% of our revenue now). It's hardly more expensive now than it was when RHEL5 was current.

There's also a TCO argument, wherein support isn't free on Windows, many requisite applications cost a lot (and are free on Linux), needing less admins, blah blah. Windows is improving, but TCO has been pretty indisputable for years.

This should never be true in an apples-to-apples comparison. You're leaving something out. Probably a lot.

Oracle is also not much cheaper. You get updates for free (no subscription required), but the question becomes what your business operating is worth. Do you want to save a trivial amount of money (comparatively) knowing that you may run into an issue which takes Oracle forever to help you with, if they even have the expertise? Indemnification and support is our business model. Not subscriptions.

If your time is worth nothing, you'll never use support, and your in-house admin team is good enough to solve almost every problem, OEL may be cheaper. But that's a hard sell.

evol262 posted:

I mean, I guarantee you don't have more RHCEs than we do. It's free for us, and we require sales, support, consulting, et al to get them, plus a significant portion of engineering does it for the hell of it. I'd guess we have more than 3k RHCEs. But I get the point. We never used support at any company I was at either. We bought RHEL because it was someone to point the finger at.

Though I can say that a lot of very large airlines and financials open cases frequently. One of the benefits of being a customer is making your business needs our business needs. MRG and SELinux (among others) came about through this.

Still, my point was that, on TCO, support isn't free with Windows, and it's debatably worthless with OEL if you need it.

There actually wasn't a hike, though. It's $200 more/year than it was 8 years ago, and it's still less than a 2012/2008 standard license. Something may have changed in your company's environment, but our prices have been stable. Could be some product you're running on RHEL, though (Tivoli or whatever)

evol262 posted:

I'm not really invested in the sales figures and don't even run RHEL anywhere, so I think CentOS is a good option.

Honestly, I wouldn't even tell emc. They'll ask gotcha questions to other people (rpm -q redhat-release-notes or cat /etc/redhat-release or whatever), but you can sorta skim those questions if you know (not that I'm advocating this ). Otherwise, if your support agreement allows it, use OEL on those boxes.

I'm reasonably sure that you can use OEL as long as you don't try to manage it with satellite or anything, but ask your TAM. Plus EMC will support that. It even calls itself Redhat in redhat-release, I think.

evol262 posted:

With what certificates?

RHCSA->RHCE->specialized certs (get 5 of these and be an rhca). Difficulty depends on track

evol262 posted:

I haven't looked in a while, but the rhcsa replaced the rhct. I'd be really surprised if an rhce had to take it, even if you never have, since it's a lesser cert.

Openstack will need studying and labbing just for neutron, really...

reading posted:

How do I use "$ ssh -X <user@machine> <text editing program> <file to edit>" ? I can't do a web search for it because nothing comes up for "ssh -x remote execution" and so forth.

I want to use a text editor on my desktop to edit some text files on my beaglebone, because it sucks using the terminal to copy and paste between different documents. I'd rather use gedit or SublimeText.

Misogynist posted:

I've got a super-weird problem with my Chef server and a NAT gateway that I'm trying to wrap my brain around.

Communications between client and server occur over HTTPS on port 443. Server is on a completely separate network; traffic must traverse the Internet to reach it. My Google Compute Engine systems that are directly connected to the Internet function just fine. Systems on Google Compute Engine that are connected over a NAT gateway send a bunch of successful REST API calls, then send one specific REST API call, receive an RST from the server, drop the connection, and blow up with an EOFError in Chef/OpenSSL/Ruby.

The NAT server on GCE is itself managed through Chef and works fine. It's also using the exact same iptables NAT configuration on AWS, where it works fine. The server's logs indicate that it's serving a 200 response.

Any ideas what might be going on?

Misogynist posted:

Crossposting from the virtualization megathread because this is niche enough where it would probably get a bite in here:

I have a need to stuff like 6 or 8 GPUs in a server and share them out to VMs, where they'll both be used for video and OpenCL/CUDA tasks. Is the GPU passthrough stuff in KVM or Xen mature enough to handle this use case, and if so, which one is likely to be better at it?

Nystral posted:

Is this like saying Oracle Linux or RHEL is not plain Linux or is there something deeper going on here? Can you explain?

Liam Emsa posted:

Following the instructions here to get GLC:

code:

    sudo add-apt-repository ppa:arand/ppa
    sudo apt-get update
    sudo apt-get install glc

And yet:

code:

:~$ sudo apt-get install glc
Reading package lists... Done
Building dependency tree       
Reading state information... Done
E: Unable to locate package glc
:~$ 

Liam Emsa posted:

So... I guess it doesn't exist anymore or something?

mod sassinator posted:

Sorry I haven't used RDP much on Linux--that's just what I remember with how RDP worked on Windows.

FordPRefectLL posted:

I am trying to figure out how crontab works for a school project.

* * * * * /usr/bin/cat /var/log/messages >> /home/tux/Documents/long_log_file

That works.

0 22 * * * /usr/bin/cat /var/log/messages >> /home/tux/Documents/long_log_file

Does not work. Can anyone help me understand why?

FordPRefectLL posted:

I am on CentOS 7 and I probably configured something incorrectly or hosed something up along the way when I was installing but the date up in the top right in the GUI was showing Phoenix time but when I checked date it was showing New York time and that's what all the cron jobs were trying to run off of.

FordPRefectLL posted:

Working on another class project. Setting up Squirrelmail with Dovacot and Postfix. I'm doing everything locally, I have no actual domain or whatever. Would I just input mail.localhost and localhost for all the various addresses?

I just need to be able to log in to Squirrelmail and show it's configured.

FordPRefectLL posted:

Alright. That's how I had it configured. When I try to log in through Squirrelmail though, I get:

Error connecting to IMAP server: mail.example.com.
0 : php_network_getaddresses: getaddrinfo failed: Name or service not known

FordPRefectLL posted:

I did that and I am getting the same error.

FordPRefectLL posted:

FordPRefectLL posted:

I added mail.example.com to the hosts file and now everything works. I could kiss you right now, jaegerx.

fuf posted:

My debian server keeps trying to send out spam because of hacked websites that had malicious php added to them. I turned off sendmail so the emails aren't actually getting sent (I hope) but they keep queuing up endlessly in /var/spool/mqueue-client.

Is there anyway I can try and track down where the emails are coming from? Like which specific PHP file is calling mail()? I'm pretty sure I've removed all the malicious php code from the site files, but they keep on coming...

Liam Emsa posted:

Yeah found it with nmap, thanks.

reading posted:

How can I compress and password protect my entire home folder under Xubuntu? This is on my single-user desktop. I want to backup my whole home folder onto an external drive but I want it to be password protected as well. So far using the GUI and right-clicking -> compress has permissions issues, since I don't know how to do that as root.

If I try using the command line the only stuff I can find online requires me to enter my password in plaintext and I'm also worried about mistyping and messing everything up.

reading posted:

It would be really ideal if I could break this massive resulting file (my whole home directory is 190GB) into lots of 5GB files too.

Cidrick posted:

I'm not trying to be a smartass - do real production environments ever use ACLs? I've never actually seen one in the wild before, only ever in lab environments. Most people are content with standard unix permissions coupled with audited sudo access. I'm curious what places decided to use ACLs and why they chose to go that route.

evol262 posted:

[citation needed]

ACLs are on the RHCE exam and everything, but ACLs and caps aren't even supported in RPM to my knowledge. You can do it in %post, but I don't think we ship a single piece of software with an ACL set by default, and we may not even have one that can manage ACLs from a frontend of any kind.

We have a hard-on for SElinux, in the sense that we aim to have every product usable with SElinux in enforcing, but...