Who's using, or planning to use, Terrastore?

[Sergio Bossa]

Hi guys,

if you're using Terrastore or just planning/evaluating it, whether it
is for your company business or your personal pet project, we want to
hear from you.

Your feedback will be very valuable to make Terrastore a better
product, so don't hesitate to write about your experiences, whether
they're good or bad: and if you don't want to share them publicly,
feel free to reach us by private mail at: sergio.bossa (at) gmail.com.

Thanks in advance!
Cheers,

Sergio B.

--
Sergio Bossa
http://www.linkedin.com/in/sergiob

[Giles Velarde]

Hi,

I probably should have said in my last mail: I am evaluating it's use
as a bioinformatics database, and as such it's integration with
elastic search is an appeal. Any more questions please do get in touch.

Best, Giles.

[Sergio Bossa]

Hi Giles,

thanks much for your feedback.
Feel free to get back with questions and suggestions (regarding
Terrastore or its ElasticSearch integration) related to your use case.

Cheers,

Sergio B.

[bhauer]

I am a KV-store noob. So bear with me if anything I say here is
misinformed or outright wrong. I'd be happy to learn that I'm wrong
on any points.

I'm using Terrastore for a pet project right now. This hands-on
evaluation will inform my decision making on future real projects. I
spent several days looking over the sea of options in RDBMS
alternatives. My goodness, there are so many options out there. For
this pet project, I had been seriously considering Project Voldemort.
At first, I liked its simplicity--I liked the pure-Java client API (as
opposed to the low signal-to-noise ratio of Apache Cassandra client
code). But I ultimately decided its architecture steered the design
of applications in a direction I wasn't happy with.

Being a simpler KV store than Terrastore, to my understanding Project
Voldemort offers no ability to leverage the server to evaluate the
Values. In order to, for example, produce a list of documents whose
"publish date" is in the past, it is necessary to either fetch all
documents and evaluate the publish date each time this operation is
needed -or- manage a lookup list of document IDs that were "published"
when the lookup list was created. Both options seem disappointing.

Perhaps I was missing something, but with Voldemort it seemed overly
complicated to do even seemingly trivial operations like manage a
counter to use as an identity for entities. To approximate atomic
updates, you create an "Updater" that continuously Gets the latest
version of a versioned value, updates the value, and attempts to Put
the value. If the version became out of date during that sequence,
the Put operation fails and the Updater repeats until it succeeds.
That's not the end of the world, but it rubs me the wrong way.

Finally, the out-of-the-box serialization of Voldemort is
disappointing. It consumes and produces data structures composed of
Lists, Maps, and so on. It doesn't actually serialize Java objects.
You can plug in Apache Avro, but (a) that's not the standard Voldemort
usage pattern and with something so new, I'd like to stick with the
practices the developers seem to use; and (b) Apache Avro is, in my
opinion, more complex than it needs to be.

Don't get me wrong; I have no metrics upon which to render a negative
judgment of Voledemort. I just decided to pass and look for options
that suited me better.

I am impressed by the number of KV options. In reviewing them, I
dismissed many for the following reasons:

1. The Java APIs appeared to be an after-thought in many cases.
2. The serialization of plain Java objects was either lacking or over-
complicated (as in Voldemort).
3. Many provided some mechanisms for atomic updates (e.g., Scalien's
testAndSet method) and/or provided some ability to leverage the
computational power of the server, but ultimately, not to the degree I
desired.
4. Some seemed to be implemented in a Windows-hostile mindset. I've
come to expect a certain Linux bias in the Java community, but
reviewing some of these KV stores could make you believe Java is a
Linux-only platform.

Arriving at the Terrastore project site, I was delighted by the
description. Document-focus, per-document consistency, scalability at
the server, server-side extensibility, a lightweight HTTP protocol,
and a very clean Java API on top of that HTTP protocol. No XML files
to edit, thank goodness!

I've only started experimenting with Terrastore on my pet project, but
I am having fun so far.

Next to the relatively polished experience of installing and
evaluating an RDMBS, the experience evaluating KVs has been
challenging. The Terrastore team deserves a lot of credit (I can't
stress this enough) for taking the time to build an installation and
deployment mechanism that is mostly based on Ant rather than Unix
shell scripts. I do Java on Windows and I don't like Cygwin.
Terrastore's installation and deployment is 95% Ant. With relatively
minimal effort, I was able to rework the remaining .sh files that were
included with the Master server and Terrastore servers into plain
batch files.

I was off and running. Since then it's been a lot of experimentation
with the various capabilities, some head-scratching, some a-ha
moments, and ultimately a bunch of questions. I hope you won't mind
the long-winded set of questions I'm going to post in a separate
thread.

Overall, I'd say Terrastore is extremely promising. I like the clean,
concise style of the documentation and the API. I haven't yet had an
opportunity to exercise its performance, but I am optimistic.

A sincere thank you to the architects, designers, developers, and
project sponsors (if there are any)!

[Sergio Bossa]

Hi bhauer,

I wanted to thank you for your feedback and your kind words and
interest in Terrastore: mails like your one really make up for all the
hard work, and it surely made my day :)

Now going on answering your next mail, stay tuned ;)
Cheers!

Sergio B.

--
Sergio Bossa
http://www.linkedin.com/in/sergiob