Tag Archives: biodata

How to build an operational datastore on AWS with S3 & Redshift


You’re building your data warehouse, and getting data into Redshift. You’ve got your ETL pipeline running, and presentation layer talking to the warehouse. Great.

But how to get access to that source data? Wouldn’t it be nice if that was close by too?

Join 35,000 others and follow Sean Hull on twitter @hullsean.

It may be you have 10-zillion rows of source data and don’t want or need to get all of that into Redshift and keep it there. But it would be nice to have access to it when you do.

Enter EXTERNAL tables, aka Spectrum. Now you can keep all your raw data in S3, an in place operational datastore of data before it’s been reworked and transformed. Use SQL to access it right where it sits.

Get all the advantages of lifecycle management in S3, and don’t pay all the redshift costs for data you don’t need all the time. Cool!

Let’s see how it works.

What is an EXTERNAL table?

Spectrum is Amazon’s rebranding of an old database technology called EXTERNAL TABLES. Back in the 90’s Oracle pioneered this work, allowing you to essentially map a CSV file, that sits outside the database proper. This means you can query all that juicy data sitting in flat files. Cool!

Athena allows you to query this stuff as a service, native to AWS. Spectrum allows you to create those external tables inside of Redshift.

Also: Top serverless interview questions for hiring aws lambda experts

Give Redshift permissions

Go into IAM and create a new role called “SeanSpectrumRole”. Assign the policy AmazonS3ReadOnlyPolicy. It looks like this:

"Version": "2012-10-17",
"Statement": [
"Effect": "Allow",
"Action": [
"Resource": "*"

If you’re using the dashboard you just pick the policy from the named list. However if you’re using CloudFormation, you’ll use the code above.

Now navigate your aws console to the Redshift dashboard, click clusters, and click the checkbox for your cluster. Probably there’s only one.

Now click the “Manage IAM Roles” button, and a dialog should popup.. Select the role you created earlier, SeanSpectrumRole. Then click “Apply Changes”.

The beauty of the AWS world is that servers themselves can have API permissions. In this case we gave the redshift cluster or server itself, access to S3 for our use below!

Related: Which engineering roles are in greatest demand?

Create your spectrum schema

First you must create a spectrum schema. Here’s the syntax:

create external schema spectrum
from data catalog
database 'sean'
region 'us-east-1'
iam_role 'arn:aws:iam::9999999999999:role/SeanSpectrumRole';

Read: Can on-demand consulting save startups time & money?

Upload your data to S3 bucket

Here we create an s3 bucket called sean_spectrum, then upload one csv file named sean_numbers.txt.

$ aws s3api create-bucket --bucket sean_spectrum --region us-east-1
"Location": "/sean_spectrum"
$ cd spectrum/
$ cat sean_numbers.txt
$ aws s3 cp sean_numbers.txt s3://sean_spectrum/
upload: ./sean_numbers.txt to s3://sean_spectrum/sean_numbers.txt
$ aws s3 ls s3://sean_spectrum/
2017-05-18 20:28:41 193 sean_numbers.txt

Note the names. The table name won’t turn out to be sean_numbers. It will be called sean_spectrum, and all files inside that directory will be queried. So make sure they have consistent formats!

Also: 30 questions to ask a serverless fanboy

Create & query your external table

Here’s how you create your external table. Note this is just a map to data. The data is still stored in S3. it is not brought into Redshift except to slice, dice & present.

mydb=# create external table spectrum_schema.sean_numbers(
id int,
fname string,
lname string,
phone string)
row format delimited
fields terminated by ','
stored as textfile
location 's3://sean_spectrum/';

Here’s how you query it:

mydb=# select * from spectrum_schema.sean_numbers order by id;
id | fname | lname | phone
12 | Scarlet | Witch | 23-35-47-555
15 | Wonder | Woman | 999-324-7878
21 | Dr. | Who | 44-22-55-77-88
32 | Quick | Silver | 22-33-77-99
35 | Bat | Man | 317-222-4777
75 | Marvel | Girl | 222-333-9595
99 | Storm | Cloud | 367-399-6767

Cool. We reordered data read from an S3 file!!!

Although you can’t create a view over a redshift table *AND* an S3 external table, you can query them together.

So for example if I have a table in redshift with addresses, I can join them together:

mydb=# select a.id, a.fname, a.lname, b.address from spectrum_schema.sean_numbers a, sean_addresses b
where a.id = b.id order by id;

id | fname | lname | phone | address
12 | Scarlet | Witch | 23-35-47-555 | 10 main st
15 | Wonder | Woman | 999-324-7878 | 25 center st
21 | Dr. | Who | 44-22-55-77-88 | 32 broadway
32 | Quick | Silver | 22-33-77-99 | 1 first st
35 | Bat | Man | 317-222-4777 | 99 west st
75 | Marvel | Girl | 222-333-9595 | 66 East Ave
99 | Storm | Cloud | 367-399-6767 | 50 North st

Also: What can startups learn from the DYN DNS outage?

Get more. Grab our exclusive monthly Scalable Startups. We share tips and special content. Our latest Why I don’t work with recruiters

A roughneck walk down database alley


I was just responding to some Disqus comments on a recent blog post. Admittedly it had a provocative title Will SQL databases just die already. What do you think?

Join 34,000 others and follow Sean Hull on twitter @hullsean.

A reader pointed out that some No-SQL databases do support joins. Huh? My face contorts in total confusion. How? Why?

For years SQL was misunderstood & unloved

Relational databases have been around for decades. 43 years according to the StackExchange article. That’s a lot of years. I’ve spent a few years as a dba, aka database administrator. The role can be distilled down as a herder of sorts. Keep all the data bits in the right boxes with the right labels. A digital librarian, that makes sure the books don’t get lost.

Of course patrons don’t always put books back where they should, and strict rules get put in place to avoid losing that one volume of shakespear in miles of shelves.

In the fast moving world of web + mobile, product is king, and agile rules the day. And anything that can make us more agile also wins. SQL, much maligned & misunderstood, was not one of those things.

Also: Top serverless interview questions for hiring aws lambda experts

NoSQL burst on the scene with much fanfare

With all that pressure, it was no wonder engineers thought, there must be a better way. Then along comes the No SQL database. I mean just the name speaks volumes about the design goal.

We’ll sacrifice anything, just please don’t make me write SQL!

The promises…

1. Never have to deal with pesky SQL that we don’t understand!
2. Interact with the database like any other data structure in our code!
3. Be schemaless! Crotchety Database Administrators be damned!
4. Be distributed. Be everywhere consistent! Be indistinguishable from magic!
5. Always be fast.

In that rush into the abyss, we lost track of durability. And down the rabbithole we went!

Related: Which engineering roles are in greatest demand?

Relational databases tried to be key-value

Then I started hearing about crazy things, like MySQL providing a memcache plugin, so you could use it albeit lightening fast, as a key-value store. You could sidestep that pesky SQL engine, and get right down to the bare metal. But why? Memcache & Redis were already doing that & purpose built. Why indeed?

I started to argue maybe we shouldn’t be muddying the waters. I mean stick with what you know!

Read: Can on-demand consulting save startups time & money?

War was won, success declared

Around this I think was when Mongodb was declaring the war won. We had finally left SQL databases in the dustheap of history. It may or may not have inspired this popular youtube skit…

Also: 30 questions to ask a serverless fanboy

Meanwhile hadoop is losing ground. Bigquery & Redshift both speak SQL

But then something funny started to happen. It seemed there was a backlash against Mongodb. A lot of customers were losing data. (Yep that’s what durability means guys…) And the hype started reversing. Even the mighty hadoop has been losing popularity of late. How long does it take to write an EMR job versus an SQL query. Let’s be honest?

I asked myself, Is Hadoop losing ground to SQL warehouses like Redshift & Bigquery?. I wonder.

Also: What can startups learn from the DYN DNS outage?

NoSQL databases are looking for JOINs?

Recently I bumped into some interesting blog comments & discussions about how Orientdb was trying to add joins to their product.

As certain relational databases try to become No SQL databases, other No SQL databases are trying to add more complex SQL, because well somehow their product is missing something.

Also: What can startups learn from the DYN DNS outage?

Engineering truth versus fashion

43 years is a lot of years. And when we drop all the fashion trends in tech, and the new database du jour, what do we find?

There is room for No SQL databases. Yep. And the do certain things, and solve certain types of problems well. But their not general workhorses, nor can they slice and dice your data however you like. And when you get to that point in your project, you’re going to want to ask interesting questions of your data.

And surely that’s where SQL excels. It ain’t going anywhere, folks!

Also: What can startups learn from the DYN DNS outage?

Get more. Grab our exclusive monthly Scalable Startups. We share tips and special content. Our latest Why I don’t work with recruiters