John McCormack DBA

SQL Server Databases and AWS Cloud Solutions

How to create a table in AWS Athena

By

How to create a table in AWS Athena

Before you learn how to create a table in AWS Athena, make sure you read this post first for more background info on AWS Athena.

Background

When you create a table in Athena, you are really creating a table schema. The underlying data which consists of S3 files does not change. You are simply telling Athena where the data is and how to interpret it. Therefore, tables are just a logical description of the data. Just like a traditional relational database, tables also belong to databases. Therefore, databases are also logical objects, which exist to group a collection of tables together. Databases and tables do not need to be created before the data is placed in to AWS S3. Similarly, if a table or database is dropped, the data will remain in S3.

All DDL statements in Athena use HiveQL DDL. Thankfully, you don’t need to be an expert in HiveQL DDL to create tables, you can learn as you go along. You can even use a wizard in the AWS console to create tables. You can script out the DDL from existing tables using the Athena console and this will give you guide for future tables.

The data used in the demo is a free download from data.gov.uk. They also have loads of data in various formats which you can use for testing.

About HiveQL DDL

Some syntax in HiveQL DDL is similar to ANSI SQL however there are are few key differences.

  • CREATE TABLE should included the keyword EXTERNAL. CREATE EXTERNAL TABLE
  • ROW FORMAT SERDE – This describes which SerDe you should use. (More about that in the about SerDe section)
  • SERDEPROPERTIES – e.g a set of rules which is applied to each row that is read, in order to split the file up into different columns. If you are not sure about this, read up more on SerDe
  • LOCATION – the S3 bucket and folder where the data resides. No filename is required, just the location. e.g. s3://testathenabucket/traffic

About SerDe

SerDes are libraries which tell Hive how to interpret your data. SerDe is short for Serializer/Deserializer. There are a few to choose from that Athena supports and you cannot currently add you own.

  • Apache Web Logs
    • org.apache.hadoop.hive.serde2.RegexSerDe
  • CSV
    • org.apache.hadoop.hive.serde2.OpenCSVSerde
  • TSV
    • org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
  • Custom Delimiters
    • org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
  • Parquet
    • org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe
  • Orc
    • org.apache.hadoop.hive.ql.io.orc.OrcSerde
  • JSON
    • org.apache.hive.hcatalog.data.JsonSerDe
    • org.openx.data.jsonserde.JsonSerDe

Create a table in AWS Athena using Create Table wizard

You can use the create table wizard within the Athena console to create your tables. Just populate the options as you click through and point it at a location within S3. You must have access to the underlying data in S3 to be able to read from it. This method is slightly laborious as a result of all the screens and dropdowns, however it is reasonable enough when you only need to create a small number of tables.

Add Table

First of all, select from an existing database or create a new one. Give your table a name and point to the S3 location.

add table in aws athena

Data format

Various data formats are acceptable. Parquet and ORC are compressed columnar formats which certainly makes for cheaper storage and query costs and quicker query results. Other formats such as JSON and CSV can also be used, these can be compressed to save on storage and query costs however you would still select the data format as the original data type. e.g. For .csv.gz – you would choose CSV.

How to create a table in AWS Athena

Columns

Column names and data types are selected by you. As a result, you need to know the structure of your data for this (or open the file to check)

add column to table in aws athena

Partitions

Above all, data should be partitioned where appropriate, such as by day or by customer ID. Wherever it makes sense as this will reduce the amount of data scanned by Athena which reduces cost and improves query performance even more than compression alone.

add partitions to tables in aws athena

Create a table in AWS Athena automatically (via a GLUE crawler)

An AWS Glue crawler will automatically scan your data and create the table based on its contents. Due to this, you just need to point the crawler at your data source. Once created, you can run the crawler on demand or you can schedule it. Hence, scheduling is highly effective for loading in new data and updating data where underlying files have changed.

Give your crawler a name and description

Glue crawler info

Point the crawler to your data store.

AWS Glue Crawler Add Data Store

Select or create an IAM role. The crawler runs under an IAM role which must have the correct permission to create tables and read the data from S3.

AWS Glue Choose IAM Role

Choose a schedule for your Glue Crawler.

AWS Glue Crawler Schedule

Declare the output location for your data.

AWS Glue Crawler Output

Finally, query your data in Athena. You can type SQL into the new query window, or if you just want a sample of data you can click the ellipses next to the table name and click on preview table.

Create a table in AWS Athena using HiveQL (Athena Console or JDBC connection)

This method is useful when you need to script out table creation. As well as the AWS Athena console, you can also use programs such SQL Workbench/J which rely on a JDBC connection.


CREATE EXTERNAL TABLE `demo_traffic`(
`region name (go)` string,
`ons lacode` string,
`ons la name` string,
`cp` bigint,
`s ref e` bigint,
`s ref n` bigint,
`s ref latitude` double,
`s ref longitude` double,
`road` string,
`a-junction` string,
`a ref e` bigint,
`a ref n` bigint,
`b-junction` string,
`b ref e` bigint,
`b ref n` bigint,
`rcat` string,
`idir` string,
`year` bigint,
`dcount` string,
`hour` bigint,
`pc` bigint,
`2wmv` bigint,
`car` bigint,
`bus` bigint,
`lgv` bigint,
`hgvr2` bigint,
`hgvr3` bigint,
`hgvr4` bigint,
`hgva3` bigint,
`hgva5` bigint,
`hgva6` bigint,
`hgv` bigint,
`amv` bigint)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
STORED AS INPUTFORMAT
'org.apache.hadoop.mapred.TextInputFormat'
OUTPUTFORMAT
'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION
's3://johnbox-athena/Traffic/'
TBLPROPERTIES (
'compressionType'='none',
'delimiter'=',',
'objectCount'='1',
'skip.header.line.count'='1',
'typeOfData'='file')

Further resources

Introduction to AWS Athena

By

Introduction to AWS Athena

AWS AthenaThis introduction to AWS Athena gives a brief overview of what what AWS Athena is and some potential use cases. I discuss in simple terms how to optimize your AWS Athena configuration for cost effectiveness and performance efficiency, both of which are pillars of the AWS Well Architected Framework. My Slides | AWS White Paper

AWS’s own documentation is the best place for full details on the Athena offering, my page hopes to serve as further explanation and also act as an anchor to some more detailed information. As it is a managed service, Athena requires no administration, maintenance or patching. Performance may not be on par with a Relational Database Management System (RDBMS) but costs can be much cheaper as you are running no servers and there are no licencing implications.

What is AWS Athena

Athena is a service provided by AWS. It allows users to query static files, such as CSVs (which are stored in AWS S3) using SQL Syntax. The queries are made using ANSI SQL so many existing users of database technologies such as MySQL or SQL Server can adapt quickly to using ANSI. New users can learn the commands easily.

How does it save me money?

“Object based storage” like Amazon S3 is a lot cheaper than “block based storage” such as EBS. This means you can store large data sets as CSV files on Amazon S3 at a fraction of the price it would cost to store the data using EBS. You are then charged for each query (currently $5 per 5TB scanned). Clever use of compression and partitioning can reduce the amount of data scanned, meaning queries will be cheaper. AWS Athena is described as serverless which means the end user doesn’t need to manage or administer any servers, this is all done by AWS.

Save more using compression, partitioning and columnar data formats

If you notice from the previous paragraph that the query cost is $5 per 5TB scanned so the pricing is quite straightforward. Athena uses per megabyte charging, with a 10MB minimum. You can save by compressing, partitioning and/or converting data to a columnar format. The less data that needs to be scanned, the cheaper the query.

  • Compression
    • As Athena natively reads compressed files, the same query that works against a CSV file will also work against data compressed into one of the following formats:
      • Snappy (.snappy)
      • Zlib (.bz2)
      • GZIP (.gz)
    • As less data is scanned, the overall cost is lower
  • Partitioning
    • Tables can be partitioned on any key. e.g. OrderDate
    • If the query can use the key, there is no need to scan all the other partitions,  only the relevant partition needs to be scanned.
    • Compression and partitioning can be used together to further reduce the amount of scanned data.
  • Converting to columnar
    • Columnar formats such as ORC and Parquet are supported
    • Converting may add complexity to your workload
    • However it will save money on querying due to the columnar format, data scanned is reduced and speed is improved
    • Here’s a tutorial, it will require an intermediate knowledge of EMR

Use Cases

  • Huge, infrequently accessed data sets in RDBMS (Extracted to a flat file format in S3)
  • Apache Web Logs
  • AWS Cloudwatch logs
  • Ad hoc querying where you don’t want to introduce the complexity of EMR or Redshift

Further Reading

Deleting data from a data warehouse

By

t-sql tuesday logoThis post about deleting data from a data warehouse is my first post in the #tsql2sday series.  This month’s host is Arun Sirpal. (blog)

T-SQL Tuesday #98 – Your Technical Challenges Conquered

Please write about and share with the world a time when you faced a technical challenge that you overcame.

Deleting data from a data warehouse

I was tasked with deleting a large amount of data from our data warehouse. This was because we had sold off a small part of our company (based in a different country) and as such, all associated data had to be removed from our data warehouse. The data warehouse pooled the data from our various OLTP data sources and stored it in one data warehouse.

Challenges

  • Find out the scope of the data
  • Identify the data to be purged
  • Identify the order for the purge to take place
    • Why was this important?
  • Create a process for deleting data that will not interfere with our daily loads
  • Test, and then deploy with confidence
Find out the scope of the data & Identify the data to be purged
  1. I had to identify which tables held the data. I created a tracking table and inserted the names of all of the tables in our two databases which held customer data.
  2. Looping through all of the tables, I identified which tables had 0 rows where sourceid = 5 and marked these as completed in my tracking table.
  3. The remaining tables containing rows where sourceid = 5 would be the tables where the purge was needed.
Identify the order for the purge to take place

It was not possible or practical to just delete the data based on a list of table names where data was held. I had to consider referential integrity and identify a precedence order in which to carry out the deletes, table by table. Many of our tables had Foreign Key constraints so SQL Server simply prevents you from deleting out of order. Some of our tables had multiple Foreign Key relationships and some went down through many levels. This made determining the order of deletions a difficult task.

Explanation:
If you try to delete a row in a primary key table, the delete will fail when the primary key value corresponds to a value in the foreign key constraint of another table. To make this change, you must do your delete of the foreign key data in the foreign key table first, and then delete from the primary key table.

I wrote a stored procedure which reviewed all of the foreign keys in the database and identified the correct order in which tables could be deleted from. Some tables had no foreign key relationships so these could be done in any order. They were the lowest precedence and were done first. The next set of tables were those foreign key tables referenced by a table with a primary key, but did not reference any other tables. These tables were processed next. This process continued on creating a hierarchy which eventually allowed me to identify the correct order in which the tables could be processed.

Create a process for deleting data that will not interfere with our daily loads

Over the years I have used SSIS for many large administration projects. It is useful where a complex workflow has to be identified and where there are many moving parts. SSIS can do so much more than just ETL. With this in mind, I created an SSIS package which  would contain all of the code.

A main outline of the SSIS package

Steps to create objects only do so if the object doesn’t already exist.

  • Create and pre-populate (with table names and expected row counts) a metadata tracking table. As the package was running through, this would allow me to tell at a glance how many rows had been deleted, how many tables had been completed and how far the task had progressed overall. This information was important and also allowed me to provide regular updates to management.
  • Create a stored procedure which would determine the order in which tables should be purged.
  • Run the stored procedure and update the metadata tracking table with the precedence values.
  • Delete the rows. The deletes would be done one table at a time, in optimised batches following the precedence order set out in the metadata tracking table to prevent error. The number of rows deleted for each table was updated in the tracking table.

With the package created, I set up a SQL Agent Job and set a schedule for times outside of the data loads. I also added in a step to ensure the loads were not still running when the job went to run.

Test, and then deploy with confidence

With a process created that worked, I had to set about testing it on a large scale to ensure a smooth process by the time my package was going to target our production database. Fortunately, we had 4 full scale dev and QA environments, each of them had a full set of data. There were slight differences in data volumes due to refresh dates, and also some newer tables existed that weren’t in production.

Having so many environments allowed me to focus on batch size to get the most efficient number of rows to delete per run. I set this as a user variable in SSIS which allowed me to pass in the value via the agent job step. If I felt it needed adjusted, I could just adjust the job step without amending the package.

Overall Result

The production run completed over a couple of weeks. In that time, several billion rows of data over hundreds of tables were deleted.

Alternative options

As I was confident all related data would be deleted, I could have disabled the Foreign Key constraints and deleted in any order. I decided against this for the following reasons.

  • The full purge would take two weeks. As each of the tables also held data that should remain, and also because we would still be running daily loads, I felt this was unacceptable as it could lead to periods of time where the referential integrity was in doubt.
  • I could forget to re-enable some constraints.
  • Once re-enabled, the foreign key constraints would be untrusted. They would need to be checked to reset the is_not_trusted value to 0 in sys.foreign_keys.

EC2 SQL Server Backups to Amazon S3

By

How to write your EC2 SQL Server Backups to Amazon S3

How to write your EC2 SQL Server Backups to Amazon S3This post specifically discusses how to write your EC2 SQL Server Backups to Amazon S3. It should not be confused with running SQL Server on RDS which is Amazon’s managed database service. To back up to S3, you will need to have an AWS account and a bucket in S3 that you want to write to. You’ll also need a way of authenticating into S3 such as an access key or IAM roles.

Background

Recently, I have been investigating running SQL Server in Amazon EC2. One issue to resolve is where to store database backups. Amazon S3 is far cheaper and has much higher redundancy than Amazon EBS (Elastic Block Store) so storing the backups here is an attractive idea. Furthermore, we can use rules in S3 to auto delete old backups or move them to Glacier which is cheaper than S3 Standard. However,  there is no obvious or inbuilt way to write your EC2 SQL Server Backups to Amazon S3 directly.

Example Costs as of December 2017 (US-EAST-1 Region) for storing each Terabyte of backup.

EBS – General purpose SSD Amazon S3 Amazon Glacier
$1228.80 $282.62 $49.15

Why is this a problem? (Block storage vs object storage)

SQL Server expects to be able to write backups to a file location (block storage) but S3 is Object Storage. This makes S3 unsuitable for direct backups. Newer versions of SQL Server (2016 onwards) also allow a BACKUP DATABASE TO URL option but this is designed for Microsoft Azure Blob Storage and as such, an S3  bucket cannot be used as the URL.

As such, we need a method to get the file into S3, either directly or indirectly because the storage cost is so much cheaper than EBS. Amazon recommend using File Gateway but this involves running another EC2 machine. This isn’t cost effective or easy to implement so I’ve discarded this method for now. The two options which I have tried and tested are below, they both have their pros and cons. I’ve also listed 3 other methods which may help you write your EC2 SQL Server Backups to Amazon S3 directly but these are untested by me as I rejected the approach.

A note on Ola

The Ola Hallengren backup solution (link) works well in EC2 when writing to EBS storage as this is just treated like a local drive. It also works well if writing to Azure Blob Storage. Whilst it also works with 3rd party backup tools such as LiteSpeed, it doesn’t cater for the latest LiteSpeed functionality that does allow writes directly to S3. I have written to Ola to ask if this is something that could be considered for a future release but until then, I am unable to use Ola’s solution.

Option 1 – Write to EBS and Sync with S3

Take a database backup using t-sql. Better still, schedule a job via SQL Agent to run at a set interval.

BACKUP DATABASE [Test]
TO DISK = N'D:\MSSQL\BACKUP\Test.bak' WITH NOFORMAT,
NOINIT,
NAME = N'Test-Full Database Backup',
SKIP,
NOREWIND,
NOUNLOAD,
STATS = 10

Using the AWS command line, copy the backup from your EBS volume to your S3 bucket. This should be step 2 of your agent job.

aws s3 sync D:\MSSQL\BACKUP s3://sqlpoc-backups/prod-backups

Delete the backup on EBS so the space will be available for the next backup. I have used PowerShell to do this but you can choose any method you prefer. This should be step 3 of your agent job.

remove-item D:\MSSQL\BACKUP\* -Recurse

Note: When sizing your EBS volume, it needs to be at least the size of your biggest database backup (with some room for growth). If this is the size you choose, bear in mind you can only back up one database at a time. If you size the drive bigger, you can back up more databases in the same job before syncing with S3 however this will be more expensive. With EBS, you pay for the size of the volume allocated, not just what is used. This is different to S3 where you only pay for data stored.

Option 2 – Write directly using LiteSpeed

Quest LiteSpeed is a 3rd party product for backing up SQL Databases. One very useful feature of enterprise edition is that it allows you to backup directly to Amazon S3. This is huge because it takes away the need for any intermediate storage, VMs or other services. It pushes your backup directly to the desired bucket location in S3. To do this, you’ll need to add a few new parameters to your backup calls but overall, it’s saves a load of hassle.

Note: This isn’t an ad for LiteSpeed, my decision to use it is down to my company already having a license agreement with Quest. If I didn’t have this in place, I would more than likely go with option 1, ultimately cost would be the deciding factor unless the difference was small.

Before writing your t-sql for the backup job, you will need to add your cloud account details to the LiteSpeed desktop app. Here you will need to pass in your Cloud Vendor, Display name, AWS Access Key, AWS Secret Key, region, storage class and bucket name.


exec master.dbo.xp_backup_database
@database = N'Test',
@backupname = N'Test - Full Database Backup',
@desc = N'Full Backup of Test on %Y-%m-%d %I:%M:%S %p',
@compressionlevel = 7,
@filename = 'test-backups/Test-Full.bkp', -- folder name within bucket defined here
@CloudVendor = N'AmazonS3',
@CloudBucketName = N' sqlpoc-backups',
@CloudAccessKeyEnc = N'xxxxxxxxxxxx’, -- an encrypted version of your access key (generated by LiteSpeed app)
@CloudSecretKeyEnc = N' xxxxxxxxxxxx, -- an encrypted version of your secret key (generated by LiteSpeed app)
@CloudRegionName = N'us-east-1',
@UseSSL = 1,
@OLRMAP = 1 ,
@init = 1,
@with = N'STATS = 10',
@verify = 1

As Amazon S3 is object storage, it does not support appending to an existing file so if using the same name each time for your backups, the previous backup is deleted and replaced. If you turn on versioning in your S3 bucket, you can keep old versions of the same filename. Or if each backup has a unique name (due to time stamp being appended), it will save them side by side. Do remember to have Object Lifecycle Rules turned on for all or specific items in your bucket to either delete old backups or send them to the much cheaper Glacier storage tier. If you don’t, your bucket size will grow very quickly and costs will escalate.

Option 3 – Map s3 bucket as a drive

If only this was easy. Doing this natively is not possible as far as I can tell. This comes down to a difference in storage types. S3 is object storage and for a drive, we need block storage. That said, there a number of 3rd party solution which I have not tested that seem to offer this service. A quick google suggests CloudBerry and Tntdrive as the most popular providers.

Option 4 – File Gateway/Storage Gateway

When investigating methods for saving my backups to S3, AWS support recommended using File Gateway. This would certainly overcome the technical challenge of getting the files into object storage (S3) as it allows you to mount an NFS Volume, but it adds complexity and cost to the process. File gateway is very useful as a method of connecting an on-premises appliance to cloud based storage however as the instances were already in the cloud (EC2), I didn’t feel like this was a very satisfactory solution.

Option 5 – Other 3rd Party tools

Other 3rd party tools exist such as CloudBerry which allow but I have not looked into them in detail. Prices start around $150 per server but they do offer volume discounts.

Useful references

https://docs.microsoft.com/en-us/sql/relational-databases/backup-restore/sql-server-backup-to-url

https://support.quest.com/technical-documents/litespeed-for-sql-server/8.6/user-guide/8

How do you write your EC2 SQL Server Backups to Amazon S3?

If you use another method listed not here, I’d be delighted to hear about it in the comments.

SQL Server Glasgow Meetup

By

sql server replication basicsI had the pleasure of speaking at last night’s SQL Server Meetup in Glasgow. It was a fairly relaxed affair and I hope that the session was enjoyed by those who attended.

I presented “A guide to SQL Server Replication, how to fix it when it breaks and alternatives to replication“. The slides are pitched at a basic level and don’t dive too deep into replication but give an overview of the different types of replication plus the key replication components.

As promised, I have made the slides available for download.

I also enjoyed doing the replication demo which worked but the content of that is not set out in a step by step fashion so I’ll work on that and aim to share it as soon as I get the chance. I focused the demo on setting up transactional replication and the importance of scripting it out so it is repeatable. I also showed how to use Replication monitor, how to use tracer tokens & how to query the distribution database for errors. There’s more info on that in this blog post.

Thanks to all who attended and thanks to Robert French for hosting.