Go, Bleve and Library oriented software
By R. S. Doiel, 2018-02-19 (updated: 2018-02-22)
In 2016, Stephen Davison, asked me, "Why use Go and Blevesearch for our library projects?" After our conversation I wrote up some notes so I would remember. It is now 2018 and I am revising these notes. I think our choice paid off. What follows is the current state of my reflection on the background, rational, concerns, and risk mitigation strategies so far for using Go and Blevesearch for Caltech Library projects.
I first came across Go a few years back when it was announced as an Open Source project by Google at an Google I/O event (2012). The original Go authors were Robert Griesemer, Rob Pike, and Ken Thompson. What I remember from that presentation was Go was a rather consistent language with the features you need but little else. Go developed at Google as a response to high development costs for C/C++ and Java in addition to challenges with performance and slow compilation times. As a language I would put Go between C/C++ and Java. It comes the ease of writing and reading you find in languages like Python. Syntax is firmly in the C/C++ family but heavily simplified. Like Java it provides many modern features including rich basic data structures and garbage collection. It has a very complete standard library and provides very good tooling. This makes it easy to generate code level documentation, format code, test, efficiently profile, and debug.
Often programming languages develop around a specific set of needs. This is true for Go. Given the Google origin it should not be surprising to find that Go's primary strengths are working with structured data, I/O and concurrency. The rich standard library is organized around a package concept. These include packages supporting network protocols, file and socket I/O as well as various encoding and compression scheme. It has particularly strong support for XML, JSON, CSV formatted data out of the box. It has a template library for working with plain text formats as well as generating safe HTML. You can browse Go's standard library https://golang.org/pkg/.
An additional feature is Go's consistency. Go code that compiles under version 1.0 still compiles under 1.10. Even before 1.0 code changes that were breaking came with tooling to automatically updates existing code. Running code is a strong element of Go's evolution.
Go is unsurprising and has even been called boring. This turns out to be a strength when building sustainable projects in a small team.
Why do I write Go?
For me Go is a good way to write web services, assemble websites, create search appliances and write command line (cli) utilities. When a shell script becomes unwieldy Go is often what I turn to. Go is well suited to building tools as well as systems. Go based command line tools are very easy to orchestrate with shell and Python.
Go runs on all the platforms I actively use - Windows, Mac OS X, Linux on both Intel and ARM (e.g. Raspberry Pi, Pine64). It has experimental support for Android and iOS. I've used a tool called GopherJS to write web browser applications that transform my command line tools into web tools with a friendlier user interface (see our BibTeX Tools).
Go supports cross compiling out of the box. This means a production system running on AWS, Google's compute engine or Microsoft's Azure can be compiled from Windows, Mac OS or even a Raspberry Pi! Deployment is a matter of copying the (self contained) compiled binary onto the production system. This contrasts with other platforms like Perl, PHP, Python, NodeJS and Ruby where you need to install not only your application code but all dependencies. While interpretive languages retain an advantage of having a REPL, Go based programs have advantages of fast compile times and easy deployment.
In many of the projects I've written in Go I've only required a few (if any) 3rd party libraries (packages in Go's nomenclature). This is quite a bit different from my experience with Perl, PHP, Python, NodeJS and Ruby. This is in large part a legacy of having grown up at Google before become an open source project. While the Go standard packages are very good there is a rich ecosystem for 3rd party packages for specialized needs. I've found I tend to rely only on a few of them. The one I've used the most is Bleve.
Bleve is a Go package for building search engines. When I originally came across Bleve (around 2014), it was described as "Lucene lite". "Lucene lite" was an apt description, but I find it easier to use than Lucene. When I first used Bleve I embedded its functionality into the tools I used to process data and present web services. It did not have much in the way of stand alone command line tooling. Today I increasingly think of Bleve as "Elastic Search lite". It ships with a set of command line tools that include support for building Bleve's indexes. My current practice is to only embed the search portion of the packages. I can use the Bleve command line for the rest. In 2018, Bleve is being actively developed, has a small vibrant community and is used by Couchbase, a well established NoSQL player.
Who is using Go?
Many companies use Go. The short list includes Google, Amazon, Netflix, Dropbox, Box, eBay, Pearsons and even Walmart and Microsoft. This came to my attention at developer conferences back in 2014. People from many of these companies started presenting at conferences on pilot projects that had been successful and moved to production. Part of what drove adoption was the ease of development in Go along with good system performance. I also think there was a growing disenchantment with alternatives like C++, C sharp and Java as well as the weight of the LAMP, Tomcat, and OpenStack.
Highly visible Go based projects include
Hugo - the fast/popular static website generator, an alternative to Jekyll, for those who want speed
Caddy - a Go based web server trying to unseat Apache/NGinX focusing on easy of use plus speed
IPFS - a cutting edge distributed storage system based on block chains
Who is using Blevesearch?
Here's some larger projects using Bleve.
Couchbase, a NoSQL database platform are replacing Lucene with Bleve. Currently the creator of Bleve works for them.
Hugo can integrate with Bleve for search and index generation
Caddy integrates with Bleve to provide an embedded search capability
In 2014 Go was moving from bleeding to leading edge. Serious capital was behind its adoption and it stopped being an exotic conference item. In 2014 Bleve was definitely bleeding edge. By late 2015 and early 2016 the program level API stabilized. People were piloting projects with it. This included our small group at Caltech Library. In 2015 non-English language support appeared followed by a growing list of non-European languages in 2016. By mid 2016 we started to see missing features like alternative sorting added. While Bleve isn't yet 1.0 (Feb. 2018) it is reliable. The primary challenge for the Bleve project is documentation targeting the novice and non-Programmer users. Bleve has proven effective as an indexing and search platform for archival, library, and data repository content.
Adopting new software comes with risk. We have mitigated this in two ways.
- Identify alternative technology (a plan B)
- Architect our systems for easy decomposition and re-composition
In the case of Go, packages can be compiled to a C-Shared library. This allows us to share working Go packages with languages like Python, R, and PHP. We have included shared Go/Python modules on our current road map for projects.
For Blevesearch the two alternatives are Solr and Elastic Search. Both are well known, documented, and solid. The costs would be recommitting to a Java stack and its resource requirements. We have already identified what we want to index and that could be converted to either platform if needed. If we stick with Go but dropped Blevesearch we would swap out the Bleve specific code for Go packages supporting Solr and Elastic Search.
The greatest risk in adopting Go for library and archive projects was knowledge transfer. We addressed this by knowledge sharing and insuring the Go codebase can be used via command line programs. Additionally we are adding support for Go based Python modules. Training also is available in the form of books, websites and online courses (lynda.com offers a "Up Running Go" course).
What are the benefits?