Tag Archives: open source

All posts relating to Open Source software, mostly but not exclusively UNIX focused.

MongoDB document depth headache

We ran into a weird problem recently where we were unable to sync a replica set running MongoDB 3.4 when adding new members to the replica set.

The sync would begin, but at some point during the sync it would always fail with:

[replication-0] collection clone for 'database.collection' failed due to Overflow:
While cloning collection 'database.collection' there was an error
'While querying collection 'database.collection' there was an error 
'BSONObj exceeded maximum nested object depth: 200''

(For extra annoyance the sync would continue with syncing all the other databases and collections on the replica set, before then only realising it had actually failed earlier at the very end of the sync and then restarting the sync from the beginning again).

The error means that one or more documents has a max depth over 200. This could be a chain of objects, or a chain of arrays in a document – a mistake that isn’t too tricky to cause with a buggy loop or ORM.

But how is it possible that this document could be in the database in the first case? Surely it should have been refused at time of insert? Well the nested document limit size and enforcement has changed at various times in past versions and a long-lived database such as ours from early MongoDB 2.x days may have had these bad documents inserted before the max depth limit was enforced and only now when we try to use the document do the limits become a problem.

In our case the document was old, but didn’t have any issues syncing back on Mongo 3.0 but now failed with Mongo 3.4.

Finding the document is tricky – the replication process helpfully does not log the document ID, so you can’t go and purge it from the collection to resolve the issue.

With input from my skilled colleagues with better Mongo skills than I, we figured out three queries that allowed us to identify the bad documents.

1. This query finds any documents that have a long chain of nested objects inside them.

db.collection.find({ $where: function() { return tojsononeline(this).indexOf("} } } } } } } } }") != -1 } })

2. This query finds any documents that have a long chain of nested arrays. This was the specific issue in our case and this query successfully identified all the bad documents.

db.collection.find({ $where: function() { return tojsononeline(this).indexOf("] ] ] ] ] ] ]") != -1 } })

3. And if you get really stuck, you can find any bad document (for whatever reason) by reading the document and then re-writing it back out to another collection. This ensures the document gets all the limits applied at write time and will identify their ID, regardless of the specific reason for them being refused.

db.collection.find({}).forEach(function(d) { print(d["_id"]); db.new_collection.insert(d) });

Note that all of these queries tend to be performance impacting since you’re asking your database to read every single document. And the last one, copying collections, could take considerable time and space to complete.

I recommend restoring the replica set to a test system and performing the operation there where you know it’s not going to impact production if you have any data of notable size.

Once you find your bad document, you can display it with:

db.collection.find({ _id: ObjectId("54492129902178d6f600004f") });

And delete it entirely (assuming nothing important in it!) with:

db.collection.deleteOne({ _id: ObjectId("54492129902178d6f600004f") });

DevOpsDays NZ 2016

Leave a reply

I recently spoke at the inaugural DevOpsDays NZ in Wellington. The team whom put together the conference did an amazing job and it’s one of the few conferences that I’ve really enjoyed recently. If they put together a subsequent conference next year, I recommend attending if possible.

I presented about our DevOps practises and tooling at Fairfax Media / stuff.co.nz which you can find at the recording below:

Whilst the vast majority of the content of the conference was really good, the following were clear standouts to me that I recommend watching:

Bridget Cowie from Datacom talking about the issues they faced when trying to debug production outages and the cultural and technology issues encountered.
Damian Brady from OctopusDeploy presenting on the issue of meetings and disruptions in your work day and how they tackled the problem at a previous organisation.
Troy Cornwell & Alex Corkin from HealthLink walking us through how they tried to change the culture and technology of a very traditional organisation, including an honest look at what failed and went wrong, as well as what succeeded.

You can find these (and other) presentations from the conference on this Youtube page.

Puppet Training

Leave a reply

I recently ran a training session for our development team at Fairfax introducing them to the fundamentals of Puppet.

To assist with this training, I’ve developed a bunch of scripts and learning modules which I’ve now open sourced at https://github.com/jethrocarr/puppet-training

Using these modules you can:

Provision a pre-configured Puppet master + Puppet client to use for exercises.
Learn the basics of an r10k/git workflow for Puppet modules.
Create a module and get used to Puppet resources like package, file and service.
Learn the basics of ordering and dependencies.
Use Hiera to set params.

It’s not as complete a course as I’d like. I did about half a day using these modules and another half the day adhoc. Ideally I’d like to finish off writing modules at some point, but it takes a reasonably long period to write anything like this and there’s only so many hours in a week :-)

Putting up here as they might be of interest to people. PRs are always welcome as well.

Building a mail server with Puppet

2 Replies

A few months back I rebuilt my personal server infrastructure and fully Puppetised everything, even the mail server. Because I keep having people ask me how to setup a mail server, I’ve gone and adjusted my Puppet modules to make them suitable for a wider audience and open sourced them.

Hence announcing – https://github.com/jethrocarr/puppet-mail

This module has been designed for hobbyists or small organisation mail server operators whom want an easy solution to build and manage a mail server that doesn’t try to be too complex. If you’re running an ISP with 30,000 mailboxes, this probably isn’t the module for you. But 5 users? Yourself only? Keep on reading!

Mail servers can be difficult to configure, particularly when figuring out the linking between MTA (eg Postfix) and LDA (eg Dovecot) and authentication (SASL? Cyrus? Wut?), plus there’s the added headaches of dealing with spam and making sure your configuration is properly locked down to prevent open relays.

By using this Puppet module, you’ll end up with a mail server that:

Uses Postfix as the MTA.
Uses Dovecot for providing IMAP.
Enforces SSL/TLS and generates a legitimate cert automatically with LetsEncrypt.
Filters spam using SpamAssassin.
Provides Sieve for server-side email filtering rules.
Simple authentication against PAM for easy management of users.
Supports virtual email aliases and multiple domains.
Supports CentOS (7) and Ubuntu (16.04).

To get started with this module, you’ll need a functional Puppet setup. If you’re new to Puppet, I recommend reading Setting up and using Pupistry for a master-less Puppet setup.

Then it’s just a case of adding the following to r10k to include all the modules and dependencies:

mod 'puppetlabs/stdlib'

# EPEL & Jethro Repo modules only required for CentOS/RHEL systems
mod 'stahnma/epel'
mod 'jethrocarr/repo_jethro'

# Note that the letsencrypt module needs to be the upstream Github version,
# the version on PuppetForge is too old.
mod 'letsencrypt',
  :git    => 'https://github.com/danzilio/puppet-letsencrypt.git',
  :branch => 'master'

# This postfix module is a fork of thias/puppet-postfix with some fixes
# to make it more suitable for the needs of this module. Longer-term,
# expect to merge it into this one and drop unnecessary functionality.
mod 'postfix',
  :git    => 'https://github.com/jethrocarr/puppet-postfix.git',
  :branch => 'master'

And the following your Puppet manifests (eg site.pp):

class { '::mail': }

And in Hiera, define the specific configuration for your server:

mail::server_hostname: 'setme.example.com'
mail::server_label: 'My awesome mail server'
mail::enable_antispam: true
mail::enable_graylisting: false
mail::virtual_domains:
 - example.com
mail::virtual_addresses:
  'nickname@example.com': 'user'
  'user@example.com': 'user'

That’s all the Puppet config done! Before you apply it to the server, you also need to make sure both your forward and reverse DNS is correct in order to be able to get the SSL/TLS cert and also to ensure major email providers will accept your messages.

$ host mail.example.com
mail.example.com has address 10.0.0.1

$ host 10.0.0.1
1.0.0.10.in-addr.arpa domain name pointer mail.example.com.

For each domain being served, you need to setup MX records and also a TXT record for SPF:

$ host -t MX example.com
example.com mail is handled by 10 mail.example.com.

$ host -t TXT example.com
example.com descriptive text "v=spf1 mx -all"

Note that SPF used to have it’s own DNS type, but that was replaced in favour of just using TXT.

The example above tells other mail servers that whatever system is mentioned in the MX record is a legitmate mail server for that domain. For details about what SPF records and what their values mean, please refer to the OpenSPF website.

Finally, you should read the section on firewalling in the README, there are a number of ports that you’ll probably want to restrict to trusted IP ranges to prevent attackers trying to force their way into your system with password guess attempts.

Hopefully this ends up being useful to some people. I’ve replaced my own internal-only module for my mail server with this one, so I’ll continue to dogfood it to make sure it’s solid.

That being said, this module is new and deals with a complex configuration so I’m sure there will be some issues people run into – please raise any problems you have on the Github issues page and I’ll do my best to assist where possible.

Faking a Time Capsule with a GNU/Linux server

Leave a reply

Apple MacOS’s Time Machine feature is a great backup solution for general desktop use, but has some annoying limitations such as only working with either locally attached storage devices or with Apple’s Time Capsule devices.

Whilst the Time Capsules aren’t bad devices, they offer a whole bunch of stuff I already have and don’t need – WiFi access point, ethernet router, and network attached storage and they’re not exactly cheap either. They also don’t help anyone wanting to backup to an off-site cloud server/VPS via a VPN.

So instead of a Time Capsule, I’m using a project called netatalk to allow a GNU/Linux server to provide an AFP file share to MacOS which acts as a Time Machine suitable target.

There’s an annoyance with Time Machine where it only officially works with AFP shares specially flagged as “Time Machine” shares. So whilst Apple has embraced SMB2 as the file sharing protocol of future use, you can’t use SMB2 for Time Machine backups (Well technically you can by enabling unsupported volumes in MacOS, but then you lack the ability to restore from backup via the MacOS recovery tools).

To make life easy, I’ve written a Puppet module that install netatalk and configures a Debian GNI/Linux server to act as a Time Capsule for all local users.

After installing the Puppet module (r10k or puppet module tools), you can simply define the directory and how much space to report to each client:

class { 'timemachine':
  location     => '/mnt/backup/timemachine',
  volsizelimit => '1000000', # 1TB per user backing up
}

To setup each MacOS machine, you will need to first connect to the share using Finder. You can do this with Finder -> Go -> Connect to Server and then entering afp://SERVERNAME and authenticating with your PAM credentials for the server.

After connecting, the share should now appear under Time Machine preferences. If you experience any issues connecting, check the /var/log/afpd.log file for debug information on the server – common issues include not having created the directories for the shares or having incorrect permissions on them.

Easy IKEv2 VPN for mobile devices (inc iOS)

15 Replies

I recently obtained an iPhone and needed to connect it to my VPN. However my existing VPN server was an OpenVPN installation which works lovely on traditional desktop operating systems and Android, but the iOS client is a bit more questionable having last been updated in September 2014 (pre iOS 9).

I decided to look into what the “proper” VPN option would be for iOS in order to get something that should be supported by the OS as smoothly as possible. Last time I looked this was full of wonderful horrors like PPTP (not actually encrypted!!) and L2TP/IPSec (configuration hell), so I had always avoided like the plague.

However as of iOS 9+, Apple has implemented support for IKEv2 VPNs which offers an interesting new option. What particularly made this option attractive for me, is that I can support every device I have with the one VPN standard:

IKEv2 is built into iOS 9 and MacOS El Capitan.
IKEv2 is built into Windows 10.
Works on Android with a third party client (hopeful for native integration soonish?).
Naturally works on GNU/Linux.

Whilst I love OpenVPN, being able to use the stock OS features instead of a third party client is always nice, particularly on mobile where power management and background tasks behaviour can be interesting.

IKEv2 on mobile also has some other nice features, such as MOBIKE, which makes it very seamless when switching between different networks (like the cellular to WiFi dance we do constantly with phones/tablets). This is something that OpenVPN can’t do – whilst it’s generally fast and reliable at establishing a connection, a change in the network means issuing a reconnect, it doesn’t just move the current connection across.

Given that I run GNU/Linux servers, I went for one of the popular IPSec solutions available on most distributions – StrongSwan.

Unfortunately whilst it’s technical capabilities are excellent, it’s documentation isn’t great. Best way to describe it is that every option is documented, but what options and why you’d want to use them? Not so much. The “left” vs “right” style documentation is also a right pain to work with, it’s not a configuration format that reads nicely and clearly.

Trying to find clear instructions and working examples of configurations for doing IKEv2 with iOS devices was also difficult and there’s some real traps for young players such as generating SHA1 certs instead of SHA2 when using the tools with defaults.

The other fun is that I also wanted my iOS device setup properly to:

Use certificate based authentication, rather than PSK.
Only connect to the VPN when outside of my house.
Remain connected to the VPN even when moving between networks, etc.

I found the best way to make it work, was to use Apple Configurator to generate a .mobileconfig file for my iOS devices that includes all my VPN settings and certificates in an easy-to-import package, but also (critically) allows me to define options that are not selectable to end users, such as on-demand VPN establishment.

After a few nights of messing around and cursing the fact that all the major OS vendors haven’t just implemented OpenVPN, I managed to get a working connection. To avoid others the same pain, I considered writing a guide – but it’s actually a really complex setup, so instead I decided to write a Puppet module (clone from github / or install from puppetforge) which does the following heavy lifting for you:

Installs StrongSwan (on a Debian/derived GNU/Linux system).
Configures StrongSwan for IKEv2 roadwarrior style VPNs.
Generates all the CA, cert and key files for the VPN server.
Generates each client’s certs for you.
Generates a .mobileconfig file for iOS devices so you can have a single import of all the configuration, certs and ondemand rules and don’t have to have a Mac to use Apple Configurator.

This means you can save yourself all the heavy lifting and setup a VPN with as little as the following Puppet code:

class { 'roadwarrior':
   manage_firewall_v4 => true,
   manage_firewall_v6 => true,
   vpn_name           => 'vpn.example.com',
   vpn_range_v4       => '10.10.10.0/24',
   vpn_route_v4       => '192.168.0.0/16',
 }

roadwarrior::client { 'myiphone':
  ondemand_connect       => true,
  ondemand_ssid_excludes => ['wifihouse'],
}

roadwarrior::client { 'android': }

The above example sets up a routed VPN using 10.10.10.0/24 as the VPN client range and routes the 192.168.0.0/16 network behind the VPN server back through. (Note that I haven’t added masquerading options yet, so your gateway has to know to route the vpn_range back to the VPN server).

It then defines two clients – “myiphone” and “android”. And in the .mobileconfig file generated for the “myiphone” client, it will specifically generate rules that cause the VPN to maintain a constant connection, except when connected to a WiFi network called “wifihouse”.

The certs and .mobileconfig files are helpfully placed in /etc/ipsec.d/dist/ for your rsync’ing pleasure including a few different formats to help load onto fussy devices.

Hopefully this module is useful to some of you. If you’re new to Puppet but want to take advantage of it, you could always check out my introduction to Puppet with Pupistry guide.

If you’re not sure of my Puppet modules or prefer other config management systems (or *gasp* none at all!) the Puppet module should be fairly readable and easy enough to translate into your own commands to run.

There a few things I still want to do – I haven’t yet done IPv6 configuration (which I’ll fix since I run a dual-stack network everywhere) and I intend to add a masquerade firewall feature for those struggling with routing properly between their VPN and LAN.

I’ve been using this configuration for a few weeks on a couple iOS 9.3.1 devices and it’s been working beautifully, especially with the ondemand configuration which I haven’t been able to do on any other devices (like Android or MacOS) yet unfortunately. The power consumption overhead seems minimal, but of course your mileage may vary.

It would be good to test with Windows 10 and as many other devices as possible. I don’t intend for this module to support non-roadwarrior type configs (eg site-to-site linking) to keep things simple, but happy to merge any PRs that make it easier to connect more mobile devices or branch routers back to a main VPN host. Also happy to merge PRs for more GNU/Linux distribution support- currently only support Debian/Ubuntu, but it shouldn’t be hard to add others.

If you’re on Android, this VPN will work for you, but you may find the OpenVPN client better and more flexible since the Android client doesn’t have the same level of on demand functionality that iOS has built in. You may also find OpenVPN a better option if you’re regularly using restrictive networks that only allow “HTTPS” out, since it can be run on TCP/443, whereas StrongSwan IKEv2 runs on UDP port 500 or 4500.

Upcycling 32-bit Mac Minis

13 Replies

The first generation Intel Apple Mac Mini (Macmini1,1) has a special place as the best bang-for-buck system that I’ve ever purchased.

Purchased for around $1k NZD in 2006, it did a stint as a much more sleep-friendly server back after I started my first job and was living at my parents house. It then went on to become my primary desktop for a couple of years in conjunction with my laptop. And finally it transitioned into a media centre and spent a number of years driving the TV and handling long running downloads. It even survived getting sent over to Sydney and running non-stop in the hot blazing hell inside my apartment there.

My long term relationship on the left and a more recent stray obtained second hand.

My long term relationship on the left and a more recent stray I obtained. Clearly mine takes after it’s owner and hasn’t seen the sun much.

Having now reached it’s 10th birthday, it’s started to show it’s age. Whilst it handles 720p content without an issue, it’s now hit and miss whether 1080p H264 content will work without unacceptable jitter.

It’s previously undergone a few upgrades. I bumped it from the original 512MB RAM to 2GB (the max) years ago and it’s had it’s 60GB hard drive replaced with a more modern 500GB model. But neither of these will help much with the video decoding performance.

Given we had recently obtained something that the people at Samsung consider a “Smart” TV, I decided to replace the Mac Mini with the Plex client running natively on the TV and recycle the Mac Mini into a new role as a small server to potentially replace a much more power hungry AMD Phenom II system that performs somewhat basic storage and network tasks.

Unfortunately this isn’t as simple as it sounds. The first gen Intel Mac Minis arrived on the scene just a bit too soon for 64-bit CPUs and so are packing the original Intel Core Solo or Intel Core Duo (1 or 2 cores respectively) which aren’t clocked particularly high and are only 32bit capable.

Whilst GNU/Linux *can* run on this, supported versions of MacOS X certainly can’t. The last MacOS version supported on these devices is Mac OS X 10.6.8 “Snow Leopard” 32-bit and the majority of app developers for MacOS have decided to set their minimum supported platform at 64-bit MacOS X 10.7.5 “Lion” so they can drop the old 32-bit stuff – this includes the popular Chrome browser which now only provides 64-built builds. Basically OS X Snow Leopard is the Win XP of the MacOS world.

Even running 32-bit GNU/Linux can be an exercise in frustration. Some distributions now only ship 64-bit builds and proprietary software vendors don’t always bother releasing 32-bit builds of their apps limiting what you can run on them.

On the plus side, this earlier generation of Apple machines was before Apple decided to start soldering everything together which means not only can you replace the RAM, storage, drives, WiFi card, you can also replace the CPU itself since it’s socketed!

I found a great writeup of the process at iFixit which covers the process of replacing the CPU with a newer model.

Essentially you can replace the CPUs in the Macmini1,1 (2006) or Macmini2,1 (2007) models with any chip compatible with Intel Socket M, the highest spec model available being the Intel Core 2 Duo 2.33 Ghz T7600.

At ~$60NZD for the T7600, it was a bit more than I wanted to spend for a decade old CPU. However moving down slightly to the T7400, the second hand price drops to around ~$20NZD per CPU with international shipping included. And at 2.177Ghz it’s no slouch, especially when compared to the original 1.5Ghz single core CPU.

It took a while to get here, I used this seller after the first seller never delivered the item and refunded me when asked about it. One of my CPUs also arrived with a bent pin, so there was some rather cold sweat moments straightening the tiny pin with a screw driver. But I guess this is what you get for buying decade old CPUs from a mysterious internet trader.

I was surprised at the lack of dust inside the unit given it’s long life, even the fan duct was remarkably dust-free.

The replacement is a bit of a pain, you have to strip the Mac Mini right down and take the motherboard out, but it’s not the hardest upgrade I’ve ever had to do – dealing with cheap $100 cut-your-hand-open PC cases were much nastier than the well designed internals of the Mac. The only real tricky bit is the addition and removal of the heatsink which worked best with a second person helping remove the plastic pegs.

I did it using a regular putty knife, needle-nose pliers, phillips & flat head screw drivers and one Torx screw driver to deal with a single T10 screw that differs from the rest of the ones in the unit.

Recommend testing this things *before* putting the main case back together, they’re a pain to open back up if it doesn’t work first run.

The end result is an upgrade from a 1.5 Ghz single core 32bit CPU to 2.17 Ghz dual core 64bit CPU – whilst it won’t hold much to a modern i7, it will certainly be able to crunch video and server tasks quite happily.

The next problem was getting an OS on there.

This CPU upgrade opens up new options for MacOS fans, if you hack the installer a bit you can get MacOS X 10.7.5 “Lion” on there which gives you a 64-bit OS that can still run much of the current software that’s available. You can’t go past Lion however, since the support for the Intel GMA 950 GPU was dropped in later versions of MacOS.

Given I want them to run as servers, GNU/Linux is the only logical choice. The only issue was booting it… it seems they don’t support booting from USB flash drives.

These Mac Minis really did fall into a generational gap. Modern enough to have EFI and no legacy ports, yet old enough to be 32-bit and lack support for booting from USB. I wasn’t even sure if I would even be able to boot 64-bit Linux with a 32-bit EFI…

Given it doesn’t boot from USB and I didn’t have any firewire devices lying around to try booting from, I fell back to the joys of optical media. This was harder than it sounds given I don’t have any media and barely any working drives, but my colleague thankfully dug up a couple old CD-R for me.

“Daddy are those shiny things floppy disks?”

I also quickly remembered why we all moved on from optical media. My first burn appeared to succeed but crashed trying to load the bootloader. And then refused to eject. Actually, it’s still refusing to eject, so there’s a Debian 8 installer that might just be stuck in there until it’s dying days… The other unit’s optical drive didn’t even work at all, so I couldn’t even do the pain of swapping around hardware to get a working combination.

Having exhausted the optional of a old-school CD-based GNU/Linux install, I started digging into ways to boot from another partition on the machine’s hard drive and found a project called rEFInd.

This awesome software is an alternative boot manager for EFI. It differs from a boot loader slightly, in a traditional BIOS -> Boot Loader -> OS world, rEFInd is equivalent to a custom BIOS offering better boot functionality than the OEM vendor.

It works by installing itself into a small FAT partition that lives on the hard disk – it’s probably the easiest low-level tool I’ve ever installed – download, unzip, and run the installer from either MacOS or Linux.

Disturbingly easy from the existing OS X installation

Once installed, rEFInd kicks in at boot and offers the ability to boot from USB flash drives, in addition to the hard drive itself!

The USB flash installer has been detected as “Legacy OS from whole disk volume”.

Yusss, Debian installer booted from USB via rEFInd!

A typical Debian installation followed, only thing I was careful about was not to delete the 209.7MB FAT filesystem used by EFI – I figured I didn’t want to find out what deleting that would mean on a box that was hard enough to boot as it is…

The small < 1MB free space between the partitions here irks me so much, I blame MacOS for aligning the partitions weirdly.

Once installed. rEFInd detected Linux as the OS installed on the hard drive and booted into GRUB and from there the usual Linux boot process works fine.

Launch the penguins!

This spec sheet violates the manafacturer's warranty!

Final result -2GB RAM, 64bit CPU, delicious delicious GNU/Linux x86_64

I can confirm that both 32bit and 64bit Debian works nicely on this box (I installed 32-bit first by mistake) – so even without doing the CPU upgrade, if you want to get a bit more life out of these early unsupported Mac Minis, they’d happily run a 32-bit Debian desktop so you can enjoy wonders like a properly patched browser and operating system.

Not all other distributions will work – Ubuntu for example don’t include EFI support on their 32-bit installer which will probably cause some headaches. You should be OK with any of the major 64-bit distributions since they tend to always support EFI.

The final joy I ran into is that when I set up the Mac Mini as a headless box, it didn’t boot… it just turned on and never appeared on the network.

Seems that the Mac Minis (even the later unibody generation) have some genius firmware that disables the GPU hardware if no screen is attached, which then messes up most operating systems on it.

The easy fix, is to hack together a fake VGA load by connecting a 100Ω resister between pins 2 and 7 of a DVI-to-VGA adaptor (such as the one that ships with the Mac Mini).

I need to make a tidier/better version of this, but it works!

No idea what engineer thought this was a good feature, but thankfully it’s an easy and cheap fix, especially since I have a box littered with these now-useless adaptors.

The end result is that I now have 2x 64-bit first gen Mac Minis running Debian GNU/Linux for a cost of around $20NZD and some time dismantling/reassembling them.

I’d recommend these small Mac Minis for server purposes, but the NZ second hand prices are still a bit too expensive for their age to buy specifically for this… Once they start going below $100 they’d make reasonable alternatives to something like the Intel NUC or Raspberry Pi for small serving tasks.

The older units aren’t necessarily problem free either. Whilst the build quality is excellent, after 10 years things don’t always work right. Both of my optical drives no longer function properly and one of the Mac Minis has a faulty RAM slot, limiting it to 1GB instead of the usual 2GB.

And of course at 10 years whom knows how much longer they’ll run for – but it’s been a good run so far, so here’s to another 10 years (hopefully)! The real limiting factor is going to be the 1GB/2GB RAM long term.

How much swap should I use on my VM?

1 Reply

Lately a couple people have asked me about how much swap space is “right” for their servers – especially in the context of running low spec machines like AWS t2.nano/t2.micro or Digital Ocean boxes with low allocations like 1GB or 512MB RAM.

The old fashioned advice was always “your swap space should be double your RAM” but this doesn’t actually make a lot of sense any more. Really swap should be considered a tool of last resort – a hack even – to squeeze a bit more performance out of systems and should be used sparingly where it makes sense.

I tend to look after two different types of systems:

Small systems running a specific dedicated service (eg microservices). These systems might do nothing more than run Nginx/Apache or something like PHP-FPM or Unicorn with a few workers. They typically have 512MB-1GB of RAM.
Big heavy servers running heavy weight applications, typically Java. These systems will be configured with large memory allocations (eg 16GB) and be configured to allocate a specific amount of memory to the application (eg 10GB Java Heap) and to keep the rest free for disk cache and background apps.

The latter doesn’t need swap. There’s no time I would ever want my massive apps getting pushed into swap for a couple reasons:

Performance of these systems is critical. We’ve paid good money to allocate them specific amounts of memory which is essentially guaranteed – we know how much the heap needs, how much disk cache we need and how much to allocate to the background apps.
If something does go wrong and starts consuming too much RAM, rather than having performance degrade as the server tries to swap, I want it to die – and die fast. If Puppet has decided it wants 7 GB of RAM, I want the OOM to step in and slaughter it. If I have swap, I risk everything on the server being slowed down as it moves tasks into the horribly slow (even on SSD) swap space.
If you’re paying for 16GB of RAM, why do you want to try and get an extra 512MB out of some swap space? It’s false economy.

For this reason, our big boxes are all swapless. But what about the former example, the small microservice type boxes, or your small personal VPS type systems?

Like many things in IT, “it depends”.

If you’re running stateless clusters, provided that the peak usage fits within the memory allocation, you don’t need swap. In this scenario, your workload is sized appropriately and if anything goes wrong due to an unexpected issue, the machine will either kill the errant process or die and get removed from the pool entirely.

I run a lot of web app workers this way – for example a 1GB t2.micro can happily run 4 Ruby Unicorn workers averaging around 128MB each, plus have space for Puppet, monitoring and delayed jobs. If something goes astray, the process gets killed and the usual automated recovery processes handle things.

However you may need some swap if you’re running stateful systems (pets) where it’s better for them to go slow than to die entirely, or if you’re running a system where the peak usage won’t fit within the memory allocation due to tight budget constraints.

For an example of tight budget constraints – I run this blog on a small machine with only 512MB RAM. With an allocation this small, there’s just not enough memory to run applications like Apache and also be able to handle the needs of background daemons and Puppet runs which can use several hundred MB just by themselves.

The approach I took was to create a small swap volume and size the worker counts in Apache so that the max workers at average size would just fit within the real memory allocation. However any background or system tasks, would have to fight over the swap space.

What you can see from the above is that I’m consuming quite a bit of swap – but my disk I/O is basically nothing. That’s because most of what’s in swap on this machine isn’t needed regularly and the active workload, i.e. the apps actually using/freeing RAM constantly, fit within the available amount of real memory.

In this case, using swap allows me to get better value for money, than using the next size up machine – I’m paying just enough to run Apache and squeezing in the management tools and background jobs onto the otherwise underutilized SSD storage. This means I can spend $5 to run this blog, vs $10. Excellent!

In respects to sizing, I’m running with a 1GB swap on a 512MB RAM server which is compliant with the traditional “twice your RAM” approach to sizing. That being said, I wouldn’t extend past this, even if the system had more RAM (eg 2GB) you should only ever use swap as a hack to squeeze a bit more out of a system. Basically don’t assume swap will scale linearly as memory scales.

Given I’m running on various cloud/VPS environments, I don’t have a traditional swap partition – instead I create an image file on the root filesystem and format it as swap space – I use a third party Puppet module (https://forge.puppetlabs.com/petems/swap_file) to do this:

swap_file::files { 'default':
  ensure            => present,
  swapfile         => '/tmp/swapfile',
  swapfilesize  => '1000 MB’
}

The performance impact of using a swap file ontop of a filesystem is almost nothing and this dramatically simplifies management and allocation of swap space. Just make sure you’re not using tmpfs for that /tmp path or you’ll find that memory benefit isn’t as good as it seems.

/tmp mounted as tmpfs on CentOS

1 Reply

After a recent reboot of my CentOS servers, I’ve inherited an issue where the server comes up running with /tmp mounted using tmpfs. tmpfs is a memory-based volatile filesystem and has some uses for people, but others like myself may have servers with very little free RAM and plenty of disk and prefer the traditional mounted FS volume.

As a service it should be possible to disable this as per the above comment… except that it already is – the following shows the service disabled on both my server and also by default by the OS vendor:

The fact I can’t disable it, appears to be a bug. The RPM changelog references 1298109 and implies it’s fixed, but the ticket seems to still be open, so more work may be required… it looks like any service defining “PrivateTmp=true” triggers it (such as ntp, httpd and others).

Whilst the developers figure out how to fix this properly, the only sure way I found to resolve the issue is to mask the tmp.mount unit with:

systemctl mask tmp.mount

Here’s something to chuck into your Puppet manifests that does the trick for you:

exec { 'fix_tmpfs_systemd':
 path => ['/bin', '/usr/bin'],
 command => 'systemctl mask tmp.mount',
 unless => 'ls -l /etc/systemd/system/tmp.mount 2>&1 | grep -q "/dev/null"'
}

This properly survives reboots and is supposed to survive systemd upgrades.

Secure Hiera data with Masterless Puppet

Leave a reply

One of the biggest limitations with masterless Puppet is keeping Hiera data secure. Hiera is a great way for separating site-specific information (like credentials) from your Puppet modules without making a huge mess of your sites.pp. On a traditional Puppet master environment, this works well since the Puppet master controls access to the Hiera data and ensures that client servers only have access to credentials that apply to them.

With masterless puppet this becomes difficult since all clients have access to the full set of Hiera data, which means your webserver might have the ability to query the database server’s admin password – certainly not ideal.

Some solutions like Hiera-eyaml can still be used, but they require setting up different keys for each server (or group of servers) which is a pain with masterless, especially when you have one value you wish to encrypted for several different servers.

To solve this limitation for Pupistry users, I’ve added a feature “HieraCrypt” in Pupistry version 1.3.0 that allows the hieradata directory to be encrypted and filtered to specific hosts.

HieraCrypt works, by generating a cert on each node (server) you use with the pupistry hieracrypt --generate parameter and saving the output into your puppetcode repository at hieracrypt/nodes/HOSTNAME. This output includes a x509 cert made against the host’s SSH RSA host key and a JSON array of all the facter facts on that host that correlate to values inside the hiera.yaml file.

When you run Pupistry on your build workstation, it parses the hiera.yaml file for each environment and generates a match of files per-node. It then encrypts these files and creates an encrypted package for each node that only they can decrypt.

For example, if your hiera.yaml file looks like:

:hierarchy:
  - "environments/%{::environment}"
  - "nodes/%{::hostname}"
  - common

And your hieradata directory looks like:

hieradata/
hieradata/common.yaml
hieradata/environments
hieradata/nodes
hieradata/nodes/testhost.yaml
hieradata/nodes/foobox.yaml

When Pupistry builds the artifact, it will include the common.yaml file for all nodes, however the testhost.yaml file will only be included for node “testhost” and of course foobox.yaml will only be available on node “foobox”.

The selection of matching files is then encrypted against each host’s certificate and bundled into the artifact. The end result is that whilst all nodes have access to the same artifact, nodes can only decrypt the Hiera files relating to them. Provided you setup your Hiera structure properly, you can make sure your webserver can’t access your database server credentials and vice-versa.