After some overly-aggressive cleaning of my GOPATH, vim “hung” with the message “vim-go: initializing gopls” the next time I edited a .go file. I discovered that running :GoInstallBinaries in vim would “fix” the problem and re-install the missing packages. vim-go: fillstruct not found. Installing github.com/davidrjenni/reftools/cmd/fillstruct@master to folder /Users/jlavoie/go/bin/ vim-go: godef not found. Installing github.com/rogpeppe/godef@master to folder /Users/jlavoie/go/bin/ vim-go: motion not found. Installing github.com/fatih/motion@master to folder /Users/jlavoie/go/bin/ vim-go: errcheck not found. Installing github.com/kisielk/errcheck@master to folder /Users/jlavoie/go/bin/ vim-go: dlv not found. Installing github.com/go-delve/delve/cmd/dlv@master to folder /Users/jlavoie/go/bin/ vim-go: gorename not found. Installing golang.org/x/tools/cmd/gorename@master to folder /Users/jlavoie/go/bin/ vim-go: iferr not found. Installing github.com/koron/iferr@master to folder /Users/jlavoie/go/bin/ vim-go: golint not found. Installing golang.org/x/lint/golint@master to folder /Users/jlavoie/go/bin/ vim-go: gotags not found. Installing github.com/jstemmer/gotags@master to folder /Users/jlavoie/go/bin/ vim-go: impl not found. Installing github.com/josharian/impl@master to folder /Users/jlavoie/go/bin/ vim-go: goimports not found. Installing golang.org/x/tools/cmd/goimports@master to folder /Users/jlavoie/go/bin/ vim-go: golangci-lint not found. Installing github.com/golangci/golangci-lint/cmd/golangci-lint@master to folder /Users/jlavoie/go/bin/ vim-go: gomodifytags not found. Installing github.com/fatih/gomodifytags@master to folder /Users/jlavoie/go/bin/ vim-go: keyify not found. Installing honnef.co/go/tools/cmd/keyify@master to folder /Users/jlavoie/go/bin/ vim-go: staticcheck not found. Installing honnef.co/go/tools/cmd/staticcheck@latest to folder /Users/jlavoie/go/bin/ vim-go: asmfmt not found. Installing github.com/klauspost/asmfmt/cmd/asmfmt@master to folder /Users/jlavoie/go/bin/ vim-go: installing finished! Press ENTER or type command to continue Unless I’m missing something, go seems a bit lacking to me in its package management tooling. I don’t see a good way to “clean up” unused packages without removing dependencies from another package. ...

A cow-orker came to me with a puppet issue today. He was trying to remove a package from a fleet of RedHat servers, using ensure => absent in the package resource, but it was failing: Error: Execution of '/bin/rpm -e firefox' returned 1: error: "firefox" specifies multiple packages: firefox-78.9.0-1.el7_9.x86_64 firefox-78.9.0-1.el7_9.i686 Error: /Stage[main]/Profile::Base::Firefox/Package[firefox]/ensure: change from '78.9.0-1.el7_9' to 'absent' failed: Execution of '/bin/rpm -e firefox' returned 1: error: "firefox" specifies multiple packages: firefox-78.9.0-1.el7_9.x86_64 firefox-78.9.0-1.el7_9.i686 A quick search showed that rpm has an --allmatches option. From rpm(8): ...

Single-homed SBC In planning to migrate phone traffic from PRI to SIP, we decided to use an existing pair of session border controllers (SBCs) that were already in production for another (smaller) deployment. Before cutting over the whole organization’s voice traffic, I revisited the (3-year old) network design. While the two SBCs are in separate datacenters in separate buildings, each SBC is only single-homed. This means that there is SBC high-availability in terms of new calls, but existing calls will be dropped if there is a failure or maintenance on the switch. ...

A few times in the past, I’ve had the need to take a subdirectory of an existing repository and move it to a new repository, while preserving history. I always had to look up the syntax for git filter-branch to do this; it worked, but wasn’t very straightforward or easy to remember. At some point, a subtree split command was added to git that makes this process much simpler. My real-world use case was in the migration and modernization of our puppet installation. A local module for managing our Opsview installation was kept in the control repo. Over the years, our locally-maintained nagios plugins have grown to a point where they may be better maintained in a separate repository. ...

The problem A common AWS connectivity design is to have a direct connect (DX) connection with a VPN backup. There are some routing concerns to consider when implementing this design to make sure that traffic prefers the DX circuit and only uses the backup VPN path if the DX is unavailable. Traffic from AWS transit gateway (TGW) will always prefer the direct connect gateway (DXGW) path, but traffic in the other direction (to AWS) is dependent on the customer gateway (CGW) routing policy. ...

I maintain a local mirror site for the Linux distributions we use. This is a simple rsync setup using ftpsync and Apache. I recently added Ubuntu to the list, but ran into an issue when I tested an automated install. The installer complained it was “Unable to locate package puppet.” In the preseed file, I tell the installer to also install this package with a pkgsel directive. (Later, using a late_command directive, the service is configured and started.) ...

What it does By default an ASA does stateful inspection of all traffic. It must see the entire conversation to be able to set up the connection and pass the traffic. If traffic is asymmetric, such that the ASA only sees traffic in one direction, the packets will not be passed. Additionally, even if the traffic is symmetric and a new connection is established, subsequent fast path packets will be inspected for things such as TCP sequence number randomization, TCP normalization, and other checks. ...

Introduction AWS has made much progress over the years with IPv6 support. From S3, EC2, Cloudfront, Route53, and EC2 support back in 2016, to more recent updates to NLB and the EC2 API, I’ve appreciated every advancement and patiently waited for the next. Unfortunately, there are still pieces missing that prevent me from making full use of IPv6 in my employer’s current environment. Existing architecture The architecture is modeled after one of AWS’s recommended connectivity designs. VPCs attach to a per-region transit gateway (TGW) for access to each other, shared services, on-prem network, our Azure VNets, and Internet access. In practice, a set of TGW route tables (common, campus, etc.) allow association and propagation with and to these various routes. ...

Today I was doing some empirical testing of an application’s behavior when one of its authentication servers becomes unreachable. I typically do this with a null route on an upstream device, but noticed that iproute2 has this built in with a nice, memorable syntax. According to ip-route(8), one of the route types is blackhole: blackhole - these destinations are unreachable. Packets are discarded silently. The local senders get an EINVAL error. ...

My experience A couple of months ago, while searching google for some odd python syntax, I received an invitation. I documented my code progress and experience over the following week in a GitHub repository as I solved each one of the problems. This was a ton of fun! What is the Google Foobar Challenge Google Foobar challenge is a set of programming challenges. It is reported that Google has used this as recruiting tool to find candidate developers to hire. (Indeed, there’s a recruitme command available once you reach the upper levels.) It consists of a (somewhat contrived) sci-fi adventure story where you are the hero trying to stop an evil antagonist from destroying a planet solving programming challenges of increasing difficulty to advance the plot. ...

The problem In the process of modernizing our puppet infrastructure, I’ve been able to remove/delete many of the custom modules I had written many years ago and use community developed and supported modules from the puppet forge. Many of these modules accommodate a pattern of a single include in the manifest, and putting all the site-specific configuration data (if any) in hiera. That said, some don’t allow this, and resources must be explicitly configured. In the past, we’ve used create_resources in combination with hiera lookups. ...

I recently had a revisit an old terraform project and update it. I had built a dev environment for our applications team, and they wanted to move it to production. Typically, whenever I go through a process like this, I take the opportunity to update things like pre-commit hooks and bump the terraform version to the most recent stable release. This happened to be a migration from a 0.12.x to a 0.14.x version. After updating the provider definitions, I ran terraform init and received the following error: ...

When the licensing is updated on certain Cisco switches, the reported model number also changes. One of my coworkers ran into this issue recently while trying to coordinate an RMA with TAC for a 3850 switch. He replicated this in the lab and sent me some screenshots of his terminal session to document what he saw. I thought I’d share it here to help others. Out of the box, with the ipbase license, the switch shows up as an “-S” model. ...

We make heavy use of puppet environments in our workflow. Using r10k, git branches are magically mapped to environments. This allows a process where anyone one the team can individually work on a new feature or change, and then we can collaborate and review/revise/test in a controlled manner. We can rebase to the production branch, and use the diff output as part of our change-management documentation. Once the change is merged, however, sometimes the original branch is not deleted. ...

Since version 0.13, terraform has support for custom validation rules for input variables. The example in the documentation shows how to test a single value: variable "image_id" { type = string description = "The id of the machine image (AMI) to use for the server." validation { # regex(...) fails if it cannot find a match condition = can(regex("^ami-", var.image_id)) error_message = "The image_id value must be a valid AMI id, starting with \"ami-\"." } } But, what to do if you want to validate a more complex object, such as list(string) (or other, more complicated types)? Terraform 0.14 introduced the alltrue function that makes this much easier and readable: ...

As of version 0.14, terraform now produces a .terraform.lock.hcl file to record which versions of dependencies – currently, just providers – were chosen when terraform init was run. They recommend adding this file to your version control system so that all future runs will use and verify those same dependencies. These can be manually upgraded by running terraform init -upgrade. I commonly will develop locally and generate the lock file on my Mac. Later, as I push to production, I will migrate the workspace to Terraform Cloud, and get the following error for each provider in the lock file: ...

By default VMware virtual machines only PXE boot on first install. Once an operating system has been installed on the hard drive, it will boot that and never try to network boot again. This is due to the default BIOS boot order. By changing the boot order, they can be configured to try a network boot first and after a short timeout boot from disk. In the vSphere client, find the VM in question, and chose Edit Settings. Under VM Options, expand Boot Options and enable the “Force BIOS setup” option. ...

Motivation I was creating a TextFSM template, and was disappointed with the lack of syntax highlighting support for this filetype in my favorite text editor. Typically, this is just a plugin away, but an exhaustive search turned up nothing. So, like all good geeks, I rolled up my sleeves and made one myself. Process Over the years, I have written little snippets in my .vim/syntax/ directory, or added some missing syntax to someone else’s syntax plugin, but had never written one from scratch. ...

I have a nice netboot setup where we can PXEboot hosts to an automated installer. The last step ensures the puppet agent is running and pointed at the correct puppetmaster. The .preseed files are generated from an erb template that ends in the following: [...] <% if @distcodename == "jessie" -%> d-i preseed/late_command string \ echo -e 'DAEMON_OPTS="--server <%= @puppetmaster %>"' > /target/etc/default/puppet ; \ rm -f /target/var/lib/puppet/state/agent_disabled.lock <% else -%> d-i preseed/late_command string \ in-target sed -i '/\[main\]/a server = <%= @puppetmaster %>' /etc/puppet/puppet.conf ; \ in-target ln -s /lib/systemd/system/puppet.service /etc/systemd/system/multi-user.target.wants/puppet.service <% end -%> Older distribution versions allowed us to populate the --server option in /etc/default/puppet. This is addressed in the first part of the if clause. ...

The Problem A common issue with infrastructure as code, is that it is often possible for someone to go in after deployment and manually change things. I still want to preserve the ability for the infrastructure folks to go in and make emergency changes, but I also want to discourage this practice as much as possible. To this end, I’ve been using a pattern where any “out of band” changes are alerted to the rest of the team. That way, everyone can be aware there was a change made, and can go back afterwards and follow the standard procedures for the change. ...