Golang parallelism issues causing “too many open files” error

I’ve been hacking in golang for a while, but I’ll admit that I didn’t get too deep into some of the language nuances until more recently. Since some of them have started to bite me, here’s a little post-mortem of one of the problems I was having.

After hacking and testing code all day, I made a seemingly innocuous change, and when running my program, I saw the following error:

2015/07/10 14:34:12 too many open files

I didn’t know what I broke, but it was obviously my fault. I reverted my recent changes, but still the error persisted. Internet searches and many painful hours of debugging ensued.

I had definitely hit some sort of Heisenbug.

What had gone wrong? Digging around my system, I noticed something weird in my ps output:

$ ps auxww | grep go
james     3446  0.0  0.1 197392  9240 pts/4    Sl   11:48   0:00 go run ./main.go
james     3457  0.0  0.0   6268  1980 pts/4    Sl   11:48   0:00 /tmp/go-build030949627/command-line-arguments/_obj/exe/event
james     3487  0.0  0.1 197392  9184 pts/4    Sl   11:49   0:00 go run ./main.go
james     3501  0.0  0.0   6268  2040 pts/4    Sl   11:49   0:00 /tmp/go-build037131602/command-line-arguments/_obj/exe/event
james     3556  0.0  0.1 197392  9168 pts/4    Sl   11:49   0:00 go run ./main.go
james     3567  0.0  0.0   6268  1976 pts/4    Sl   11:49   0:00 /tmp/go-build957487534/command-line-arguments/_obj/exe/event
james     3788  0.0  0.0 197392  1636 pts/4    Sl   Jul04   0:07 go run ./main.go
james     3800  0.0  0.0   5180  1944 pts/4    Sl   Jul04   0:01 /tmp/go-build552106841/command-line-arguments/_obj/exe/event
[...]

Hoards and hoards of lingering go build artefacts, were still running. At one time I noticed over 42 of these! I quickly killed them all off:

# processes are named `event`, and I don't have any unrelated event processes running.
$ killall -9 event
kernel: ahh, much better! :)

Which brought my program back to life! Heisenbug gone… or was it? I soon noticed, that each time I ran my program, the left over process count would increment by one. What was causing this? After another session of debugging, I found that these leftovers were caused by a lack of clean up due to some buggy code. That buggy code is the interesting part. Let’s look at it:

for v := range obj.GetSomeChannel() {
    fmt.Printf("Processing: %v\n", v.Name)
    wg.Add(1)
    // BAD
    go func() {
        defer wg.Done()
        v.Start() // some method
        fmt.Printf("Finished: %v\n", v.Name)
    }()
}

I’m not sure how common this issue is, so if you’re not yet familiar with it, take a moment to try and figure it out.

…

Okay. The issue is that when you iterate through the loop, the v value which is passed in to the function (the closure) is actually referencing the memory space of v. As a result, whenever the v value changes (as it does in the loop) the v variable instantly contains the new value, and the go routine will see the value of whatever it happens to be when it uses it.

To get around this race (and it is a race) you need to copy in the value to the goroutine:

for v := range obj.GetSomeChannel() {
    fmt.Printf("Processing: %v\n", v.Name)
    wg.Add(1)
    // GOOD
    go func(v *Objtype) {
        defer wg.Done()
            v.Start() // some method
        fmt.Printf("Finished: %v\n", v.Name)

    }(v)
}

It so happens that v is a pointer, but that’s irrelevant. The value of the pointer still needs to be copied in to the goroutine that is being called to use it. In my case, v needs to be a pointer, because we want the same copy of our data to be used throughout the code.

Many thanks to bleidl for helping me with some of the analysis!

As a quick aside, I’m using this WaitGroup pattern, which replaced the much uglier version of this loop which I had previously written. For a language that claims to not be pattern and idiom heavy, there sure are a bunch that I’ve found so far, many of which come with gotchas.

Happy hacking!

James

Git archive with submodules and tar magic

Git submodules are actually a very beautiful thing. You might prefer the word powerful or elegant, but that’s not the point. The downside is that they are sometimes misused, so as always, use with care. I’ve used them in projects like puppet-gluster, oh-my-vagrant, and others. If you’re not familiar with them, do a bit of reading and come back later, I’ll wait.

I recently did some work packaging Oh-My-Vagrant as RPM’s. My primary goal was to make sure the entire process was automatic, as I have no patience for manually building RPM’s. Any good packager knows that the pre-requisite for building a SRPM is a source tarball, and I wanted to build those automatically too.

Simply running a tar -cf on my source directory wouldn’t work, because I only want to include files that are stored in git. Thankfully, git comes with a tool called git archive, which does exactly that! No scary tar commands required:

Nobody likes tar

Here’s how you might run it:

$ git archive --prefix=some-project/ -o output.tar.bz2 HEAD

Let’s decompose:

The --prefix argument prepends a string prefix onto every file in the archive. Therefore, if you’d like the root directory to be named some-project, then you prepend that string with a trailing slash, and you’ll have everything nested inside a directory!

The -o flag predictably picks the output file and format. Using .tar.bz2 is quite common.

Lastly, the HEAD portion at the end specifies which git tree to pull the files from. I usually specify a git tag here, but you can specify a commit id if you prefer.

Obligatory, “make this article more interesting” meme image.

This is all well and good, but unfortunately, when I open my newly created archive, it is notably missing my git submodules! It would probably make sense for there to be an upstream option so that a --recursive flag would do this magic for you, but unfortunately it doesn’t exist yet.

There are a few scripts floating around that can do this, but I wanted something small, and without any real dependencies, that I can embed in my project Makefile, so that it’s all self-contained.

Here’s what that looks like:

sometarget:
    @echo Running git archive...
    # use HEAD if tag doesn't exist yet, so that development is easier...
    git archive --prefix=oh-my-vagrant-$(VERSION)/ -o $(SOURCE) $(VERSION) 2> /dev/null || (echo 'Warning: $(VERSION) does not exist.' && git archive --prefix=oh-my-vagrant-$(VERSION)/ -o $(SOURCE) HEAD)
    # TODO: if git archive had a --submodules flag this would easier!
    @echo Running git archive submodules...
    # i thought i would need --ignore-zeros, but it doesn't seem necessary!
    p=`pwd` && (echo .; git submodule foreach) | while read entering path; do \
        temp="$${path%\'}"; \
        temp="$${temp#\'}"; \
        path=$$temp; \
        [ "$$path" = "" ] && continue; \
        (cd $$path && git archive --prefix=oh-my-vagrant-$(VERSION)/$$path/ HEAD > $$p/rpmbuild/tmp.tar && tar --concatenate --file=$$p/$(SOURCE) $$p/rpmbuild/tmp.tar && rm $$p/rpmbuild/tmp.tar); \
    done

This is a bit tricky to read, so I’ll try to break it down. Remember, double dollar signs are used in Make syntax for embedded bash code since a single dollar sign is a special Make identifier. The $(VERSION) variable corresponds to the version of the project I’m building, which matches a git tag that I’ve previously created. $(SOURCE) corresponds to an output file name, ending in the .tar.bz2 suffix.

    p=`pwd` && (echo .; git submodule foreach) | while read entering path; do \

In this first line, we store the current working directory for use later, and then loop through the output of the git submodule foreach command. That output normally looks something like this:

james@computer:~/code/oh-my-vagrant$ git submodule foreach 
Entering 'vagrant/gems/xdg'
Entering 'vagrant/kubernetes/templates/default'
Entering 'vagrant/p4h'
Entering 'vagrant/puppet/modules/module-data'
Entering 'vagrant/puppet/modules/puppet'
Entering 'vagrant/puppet/modules/stdlib'
Entering 'vagrant/puppet/modules/yum'

As you can see, this shows that the above read command, eats up the Entering string, and pulls the quoted path into the second path variable. The next part of the code:

        temp="$${path%\'}"; \
        temp="$${temp#\'}"; \
        path=$$temp; \
        [ "$$path" = "" ] && continue; \

uses bash idioms to remove the two single quotes that wrap our string, and then skip over any empty versions of the path variable in our loop. Lastly, for each submodule found, we first switch into that directory:

        (cd $$path &&

Run a normal git archive command and create a plain uncompressed tar archive in a temporary directory:

git archive --prefix=oh-my-vagrant-$(VERSION)/$$path/ HEAD > $$p/rpmbuild/tmp.tar &&

Then use the magic of tar to overlay this new tar file, on top of the source file that we’re now building up with each iteration of this loop, and then remove the temporary file.

tar --concatenate --file=$$p/$(SOURCE) $$p/rpmbuild/tmp.tar && rm $$p/rpmbuild/tmp.tar); \

Finally, we end the loop:

    done

Boom, magic! Short, concise, and without any dependencies but bash and git.

Nobody should have to figure that out by themselves, and I wish it was built in to git, but until then, here’s how it’s done! Many thanks to #git on IRC for pointing me in the right direction.

This is the commit where I landed this patch for oh-my-vagrant, if you’re curious to see this in the wild. Now that this is done, I can definitely say that it was worth the time:

Is it worth the time? In this case, it was.

With this feature merged, along with my automatic COPR builds, a simple ‘make rpm‘, causes all of this automation to happen, and delivers a fresh build from git in a few minutes.

I hope you enjoyed this technique, and I hope you have some coding skills to get this feature upstream in git.

Happy Hacking,

James

Oh-My-Vagrant “Mainstream” mode and COPR RPM’s

Making Oh-My-Vagrant (OMV) more developer accessible and easy to install (from a distribution package like RPM) has always been a goal, but was previously never a priority. This is all sorted out now. In this article, I’ll explain how “mainstream” mode works, and how the RPM work was done. (I promise this will be somewhat interesting!)

Prerequisites:

If you haven’t read any of the previous articles about Oh-My-Vagrant, I’d recommend you start there. Many of the articles include screencasts, and combined with the examples/ folder, this is probably the best way to learn OMV, because the documentation could use some love.

Installation:

OMV is now easily installable on Fedora 22 via COPR. It probably works on other distros and versions, but I haven’t tested all of those combinations. This is a colossal improvement from when I first posted about this publicly in 2013. There is still one annoying bug that I occasionally hit. Let me know if you can reproduce.

Install from COPR:

james@computer:~$ sudo dnf copr enable purpleidea/oh-my-vagrant

You are about to enable a Copr repository. Please note that this
repository is not part of the main Fedora distribution, and quality
may vary.

The Fedora Project does not exercise any power over the contents of
this repository beyond the rules outlined in the Copr FAQ at
, and
packages are not held to any quality or security level.

Please do not file bug reports about these packages in Fedora
Bugzilla. In case of problems, contact the owner of this repository.

Do you want to continue? [y/N]: y
Repository successfully enabled.
james@computer:~$ sudo dnf install oh-my-vagrant
Last metadata expiration check performed 0:05:08 ago on Tue Jul  7 22:58:45 2015.
Dependencies resolved.
================================================================================
 Package           Arch     Version            Repository                  Size
================================================================================
Installing:
 oh-my-vagrant     noarch   0.0.7-1            purpleidea-oh-my-vagrant   270 k
 vagrant           noarch   1.7.2-7.fc22       updates                    428 k
 vagrant-libvirt   noarch   0.0.26-2.fc22      fedora                      57 k

Transaction Summary
================================================================================
Install  3 Packages

Total download size: 755 k
Installed size: 2.5 M
Is this ok [y/N]: n
Operation aborted.
james@computer:~$ sudo dnf install -y oh-my-vagrant
Last metadata expiration check performed 0:05:19 ago on Tue Jul  7 22:58:45 2015.
Dependencies resolved.
================================================================================
 Package           Arch     Version            Repository                  Size
================================================================================
Installing:
 oh-my-vagrant     noarch   0.0.7-1            purpleidea-oh-my-vagrant   270 k
 vagrant           noarch   1.7.2-7.fc22       updates                    428 k
 vagrant-libvirt   noarch   0.0.26-2.fc22      fedora                      57 k

Transaction Summary
================================================================================
Install  3 Packages

Total download size: 755 k
Installed size: 2.5 M
Downloading Packages:
(1/3): vagrant-1.7.2-7.fc22.noarch.rpm          626 kB/s | 428 kB     00:00    
(2/3): vagrant-libvirt-0.0.26-2.fc22.noarch.rpm  70 kB/s |  57 kB     00:00    
(3/3): oh-my-vagrant-0.0.7-1.noarch.rpm         243 kB/s | 270 kB     00:01    
--------------------------------------------------------------------------------
Total                                           246 kB/s | 755 kB     00:03     
Running transaction check
Transaction check succeeded.
Running transaction test
Transaction test succeeded.
Running transaction
  Installing  : vagrant-1.7.2-7.fc22.noarch                                 1/3 
  Installing  : vagrant-libvirt-0.0.26-2.fc22.noarch                        2/3 
  Installing  : oh-my-vagrant-0.0.7-1.noarch                                3/3 
  Verifying   : oh-my-vagrant-0.0.7-1.noarch                                1/3 
  Verifying   : vagrant-libvirt-0.0.26-2.fc22.noarch                        2/3 
  Verifying   : vagrant-1.7.2-7.fc22.noarch                                 3/3 

Installed:
  oh-my-vagrant.noarch 0.0.7-1                vagrant.noarch 1.7.2-7.fc22       
  vagrant-libvirt.noarch 0.0.26-2.fc22       

Complete!
james@computer:~$

If you’d like to avoid typing passwords over and over again when using vagrant, you can add yourself into the vagrant group. 99% of people do this. The downside is that it could allow your user account to get root privileges. Since most developers have a single user environment, it’s not a big issue. This is necessary because vagrant uses the qemu:///system connection instead of qemu:///session. If you can help fix this, please hack on it.

james@computer:~$ groups
james wheel docker
james@computer:~$ sudo usermod -aG vagrant james
# you'll need to logout/login for this change to take effect...

Lastly, there is a user session plugin addition that is required. Installation is automatic the first time you create a new OMV project. Let’s do that and see how it works!

james@computer:~$ mkdir /tmp/omvtest
james@computer:~$ cd !$
cd /tmp/omvtest
james@computer:/tmp/omvtest$ which omv
/usr/bin/omv
james@computer:/tmp/omvtest$ omv init
Oh-My-Vagrant needs to install a modified vagrant-hostmanager plugin.
Is this ok [y/N]: y
Cloning into 'vagrant-hostmanager'...
remote: Counting objects: 801, done.
remote: Total 801 (delta 0), reused 0 (delta 0), pack-reused 801
Receiving objects: 100% (801/801), 132.22 KiB | 0 bytes/s, done.
Resolving deltas: 100% (467/467), done.
Checking connectivity... done.
Branch feat/oh-my-vagrant set up to track remote branch feat/oh-my-vagrant from origin.
Switched to a new branch 'feat/oh-my-vagrant'
sending incremental file list
./
vagrant-hostmanager.rb
vagrant-hostmanager/
vagrant-hostmanager/action.rb
vagrant-hostmanager/command.rb
vagrant-hostmanager/config.rb
vagrant-hostmanager/errors.rb
vagrant-hostmanager/plugin.rb
vagrant-hostmanager/provisioner.rb
vagrant-hostmanager/util.rb
vagrant-hostmanager/version.rb
vagrant-hostmanager/action/
vagrant-hostmanager/action/update_all.rb
vagrant-hostmanager/action/update_guest.rb
vagrant-hostmanager/action/update_host.rb
vagrant-hostmanager/hosts_file/
vagrant-hostmanager/hosts_file/updater.rb

sent 20,560 bytes  received 286 bytes  41,692.00 bytes/sec
total size is 19,533  speedup is 0.94
Patched successfully!
Current machine states:

omv1                      not created (libvirt)

The Libvirt domain is not created. Run `vagrant up` to create it.
james@computer:/tmp/omvtest$ ls
ansible/  docker/  kubernetes/  omv.yaml  puppet/  shell/
james@computer:/tmp/omvtest$

You can see that the plugin installation worked perfectly, and that OMV created a few files and folders.

More usage:

You can hide that generated mess in a subfolder if you prefer:

james@computer:/tmp/omvtest$ mkdir /tmp/omvtest2
james@computer:/tmp/omvtest$ cd !$
cd /tmp/omvtest2
james@computer:/tmp/omvtest2$ omv init mess
Current machine states:

omv1                      not created (libvirt)

The Libvirt domain is not created. Run `vagrant up` to create it.
james@computer:/tmp/omvtest2$ ls
mess/  omv.yaml@
james@computer:/tmp/omvtest2$ ls -lAh
total 0
drwxrwxr-x. 7 james 160 Jul  7 23:26 mess/
lrwxrwxrwx. 1 james  13 Jul  7 23:26 omv.yaml -> mess/omv.yaml
drwxrwxr-x. 3 james  60 Jul  7 23:26 .vagrant/
james@computer:/tmp/omvtest2$ tree
.
├── mess
│   ├── ansible
│   │   └── modules
│   ├── docker
│   ├── kubernetes
│   │   ├── applications
│   │   └── templates
│   ├── omv.yaml
│   ├── puppet
│   │   └── modules
│   └── shell
└── omv.yaml -> mess/omv.yaml

10 directories, 2 files
james@computer:/tmp/omvtest2$

As you can see all the mess is wrapped up in a single folder. This could even be named .omv if you prefer, and should all be committed inside of your project. Now that we’re installed, let’s get hacking!

Mainstream mode:

Mainstream mode further hides the ruby/Vagrantfile aspect of a Vagrant project and extends OMV so that you can define your entire project via the omv.yaml file, without the rest of the OMV project cluttering up your development tree. This makes it possible to have your project use OMV by only committing that one yaml file into the project repo.

The main difference is that you now control everything with the new omv command line tool. It’s essentially a smart wrapper around the vagrant command, so any command you used to use vagrant for, you can now substitute in omv. It also saves typing four extra characters!

As it turns out (and by no accident) the omv tool works exactly like the vagrant tool. For example:

james@computer:/tmp/omvtest2$ omv status
Current machine states:

omv1                      not created (libvirt)

The Libvirt domain is not created. Run `vagrant up` to create it.
james@computer:/tmp/omvtest2$ omv up
Bringing machine 'omv1' up with 'libvirt' provider...
==> omv1: Box 'centos-7.1' could not be found. Attempting to find and install...
    omv1: Box Provider: libvirt
    omv1: Box Version: >= 0
==> omv1: Adding box 'centos-7.1' (v0) for provider: libvirt
    omv1: Downloading: https://dl.fedoraproject.org/pub/alt/purpleidea/vagrant/centos-7.1/centos-7.1.box
[snip]
james@computer:/tmp/omvtest2$ omv destroy
Unlocking shell provisioning for: omv1...
==> omv1: Domain is not created. Please run `vagrant up` first.
james@computer:/tmp/omvtest2$

BUT THAT’S NOT ALL…

The existing tools you know and love, like vlog, vsftp, vscreen, vcssh, vfwd, vansible, have all been modified to work with OMV mainstream mode as well. The same goes for common aliases such as vs, vp, vup, vdestroy, vrsync, and the useful (but occasionally dangerous) vrm-rf. Have a look at the above links on my blog and the source to see what these do. If it’s not clear enough, let me know!

All of these are now packaged up in the oh-my-vagrant COPR and are installed automatically into /etc/profile.d/oh-my-vagrant.sh for your convenience. Since they’re part of the OMV project, you’ll get updates when new functions or bug fixes are made.

The plumbing:

Mainstream mode is possible because of an idea rbarlow had. He gets full credit for the idea, in particular for teaching me about VAGRANT_CWD which is what makes it all work. I rejected his 6 line prototype, but loved the idea, and since he was busy making juice, I got bored one day and hacked on a full implementation.

james@computer:~/code/oh-my-vagrant$ git diff --stat 853073431d227cbb0ba56aaf4fedd721904de9a8 aa764ae79d69475b87f293c43af4f20fd7d1d000
 DOCUMENTATION.md    | 18 +++++++++++++++
 bin/omv.sh          | 50 +++++++++++++++++++++++++++++++++++++++++
 vagrant/Vagrantfile | 65 ++++++++++++++++++++++++++++++++++-------------------
 3 files changed, 110 insertions(+), 23 deletions(-)
james@computer:~/code/oh-my-vagrant$

It turned out it was a little longer, but I artificially inflated this by including some quick doc patches. What does it actually do differently? It sets VAGRANT_CWD and VAGRANT_DOTFILE_PATH so that the vagrant command looks in a different directory for the Vagrantfile and .vagrant/ directories. That way, all the plumbing is hidden and part of the RPM.

Making the RPM:

The RPM’s happened because stefw made me feel bad about not having them. He was right to do so. In an case, RPM packaging still scares me. I think repetitive work scares me even more. That’s why I automate as much as I can. So after a lot of brain loss, I finally made you an RPM so that you could easily install it. Here’s how it went:

I started by adding the magic so that my Makefile could build an RPM.

This made it so I can easily run make srpm to get a new RPM or SRPM.

Then I added COPR integration, so a make copr automatically kicks off a new COPR build. This was the interesting part. You’ll need a Fedora account for this to work. Once you’re logged in, if you go to https://copr.fedoraproject.org/api you’ll be able to download a snippet to put in your ~/.config/copr file. Lastly, the work happens in copr-build.py where the python copr library does the heavy lifting.

#!/usr/bin/python

# README:
# for initial setup, browse to: https://copr.fedoraproject.org/api/
# and it will have a ~/.config/copr config that you can download.
# happy hacking!

import os
import sys
import copr

COPR = 'oh-my-vagrant'
if len(sys.argv) != 2:
    print("Usage: %s <srpm url>" % sys.argv[0])
    sys.exit(1)

url = sys.argv[1]

client = copr.CoprClient.create_from_file_config(os.path.expanduser("~/.config/copr"))

result = client.create_new_build(COPR, [url])
if result.output != "ok":
    print(result.error)
    sys.exit(1)
print(result.message)

A build looks like this:

james@computer:~/code/oh-my-vagrant$ git tag 0.0.8 # set a new tag
james@computer:~/code/oh-my-vagrant$ make copr 
Running templater...
Running git archive...
Running git archive submodules...
Running rpmbuild -bs...
Wrote: /home/james/code/oh-my-vagrant/rpmbuild/SRPMS/oh-my-vagrant-0.0.8-1.src.rpm
Running SRPMS sha256sum...
/home/james/code/oh-my-vagrant
Running SRPMS gpg...

You need a passphrase to unlock the secret key for
user: "James Shubin (Third PGP key.) <james@shubin.ca>"
4096-bit RSA key, ID 24090D66, created 2012-05-09

gpg: WARNING: The GNOME keyring manager hijacked the GnuPG agent.
gpg: WARNING: GnuPG will not work properly - please configure that tool to not interfere with the GnuPG system!
Running SRPMS upload...
sending incremental file list
SHA256SUMS
SHA256SUMS.asc
oh-my-vagrant-0.0.8-1.src.rpm

sent 8,583 bytes  received 2,184 bytes  4,306.80 bytes/sec
total size is 1,456,741  speedup is 135.30
Build was added to oh-my-vagrant.
james@computer:~/code/oh-my-vagrant$

A few minutes later, the COPR build page should look like this:

A screenshot of the Oh-My-Vagrant COPR build page for people who like to look at pretty pictures instead of just terminal output.

There was a bunch of additional fixing and polishing required to get this as seamless as possible for you. Have a look at the git commits and you’ll get an idea of all the work that was done, and you’ll probably even learn about some new, features I haven’t blogged about yet. It was exhausting!

As a result of all this, you can download fresh builds easily. Visit the COPR page to see how things are cooking:

https://copr.fedoraproject.org/coprs/purpleidea/oh-my-vagrant/

I’ll try to keep this pumping out releases regularly. If I lag behind, please holler at me. In any case, please let me know if you appreciate this work. Comment, tweeter, or contact me!

Happy Hacking,

James