Q: Does MySQL support ACID? A: Yes

I was recently asked this question by an experienced academic at the NY Oracle Users Group event I presented at.

Does MySQL support ACID? (ACID is a set of properties essential for a relational database to perform transactions, i.e. a discrete unit of work.)

Yes, MySQL fully supports ACID, that is Atomicity, Consistency, Isolation and Duration. (*)

This is contrary to the first Google response found searching this question which for reference states “The standard table handler for MySQL is not ACID compliant because it doesn’t support consistency, isolation, or durability”.

The question is however not a simple Yes/No because it depends on timing within the MySQL product’s lifecycle and the version/configuration used in deployment. What is also *painfully* necessary is to understand why this question would even be asked of the most popular open source relational database.

MySQL has a unique characteristic of supporting multiple storage engines. These engines enabling varying ways of storing and retrieving data via the SQL interface in MySQL and have varying features for supporting transactions, locking, index strategies, compression etc. The problem is that the default storage engine from version 3.23 (1999) to 5.1 (2010) was MyISAM, a non-transactional engine, and hence the first point of confusion.

The InnoDB storage engine has been included and supported from MySQL 3.23. This is a transactional engine supporting ACID properties. However, not all of the default settings in the various MySQL versions have fully meet all ACID needs, specifically the durability of data. This is the second point of confusion. Overtime other transactional storage engines in MySQL have come and gone. InnoDB has been there since the start so there is no excuse to not write applications to fully support transactions. The custodianship of Oracle Corporation starting in 2010 quickly corrected this *flaw* by ensuring the default storage engine in MySQL 5.5 is InnoDB. But the damage to the ecosystem that uses MySQL, that is many thousands of open source projects, and the resources that work with MySQL has been done. Recently working on a MySQL 5.5 production system in 2016, the default engine was specifically defined in the configuration defined as MyISAM, and some (but not all tables) were defined using MyISAM. This is a further conversation as to why, is this a upgrade problem? Are there legacy dependencies with applications? Are the decision makers and developers simply not aware of the configuration? Or, are developers simply not comfortable with transactions?

Like other anti-reasonable MySQL defaults the unaware administrator or developer could consider MySQL as supporting ACID properties, however until detailed testing with concurrency and error conditions not realize the impact of poor configuration settings.

The damage of having a non-transactional storage engine as the default for over a decade has created a generation of professionals and applications that abuses one of the primary usages of a relational database, that is a transaction, i.e. to product a unit for work that is all or nothing. Popular open source projects such as WordPress, Drupal and hundreds more have for a long time not supported transactions or used InnoDB. Mediawiki was at least one popular open source project that was proactive towards InnoDB and transaction usage. The millions of plugins, products and startups that build on these technologies have the same flaws.

Further confusion arises when an application uses InnoDB tables but does not use transactions, or the application abuses transactions, for example 3 different transactions that should really be 1.

While newer versions of MySQL 5.6 and 5.7 improve default configurations, until these versions a more commonly implemented non-transactional use in a relational database will remain. A recent Effective MySQL NYC Meetup survey showed that installations of version 5.0 still exist, and that few have a policy for a regular upgrade cadence.

Do you control your database outages?

Working with a client last week I noted in my analysis, “The mysql server was restarted on Thursday and so the [updated] my.cnf settings seems current”. This occurred between starting my analysis on Wednesday and delivering my findings on Friday.

# more /var/lib/mysql/ip-104-238-102-213.secureserver.net.err
160609 17:04:43 [Note] /usr/sbin/mysqld: Normal shutdown

The client however stated they did not restart MySQL and would not do that at 5pm which is still a high usage time of the production system. This is unfortunately not an uncommon finding, that a production system had an outage and that the client did not know about it and did not instigate this.

There are several common causes and the “DevOps” mindset for current production systems has made this worse.

  • You have managed hosting and they perform software updates with/without your knowledge. I have for example worked with several Rackspace customers and there would be an outage because Rackspace engineers decided to apply an upgrade at a time that suited them, not their customers.
  • You have chosen automate updates for your Operating System.
  • Your developers update the software when they like.
  • You are using a 3rd party product that is making an arbitrary decision.

In this case the breadcrumbs lead to the last option, that cPanel is performing this operation as hinted by the cPanel specific installed MySQL binaries.

$ rpm -qa | grep -i mysql

Also note that cPanel still uses MySQL 5.1 shared libraries.

So why did cPanel perform not one shutdown, but two in immediate succession? The first was 17 seconds, the second was 2 seconds. Not being experienced with cPanel I cannot offer an answer for this shutdown occurance. I can for others which I will detail later.

160609 17:04:24 [Note] /usr/sbin/mysqld: Normal shutdown
160609 17:04:28 [Note] /usr/sbin/mysqld: Shutdown complete
160609 17:04:41 [Note] /usr/sbin/mysqld: ready for connections.
160609 17:04:43 [Note] /usr/sbin/mysqld: Normal shutdown
160609 17:04:45 [Note] /usr/sbin/mysqld: ready for connections.

And why did the customer not know about the outage? If you use popular SaaS monitoring solutions such as New Relic and Pingdom you would not have been informed because these products have a sampling time of 60 seconds. I use these products along with Nagios on my personal blog site as they provide adequate instrumentation based on the frequency of usage. I would not recommend these tools as the only tools used in a high volume production system simply because of this one reason. In high volume system you need sampling are much finer granularity.

So just when you were going to justify that 17 seconds while unexpected is tolerable, I want to point out that this subsequently occurred and the outage was over 4 minutes.

160619 11:58:07 [Note] /usr/sbin/mysqld: Normal shutdown
160619 12:02:26 mysqld_safe Starting mysqld daemon with databases from /var/lib/mysql

An analysis of the MySQL error log which is correctly not rolled as I always recommend showed a pattern of regular MySQL updates, from 5.5.37 thru 5.5.50. This is the most likely reason a 3rd party product has performed a database outage, to perform a software update at their choosing, not yours.

150316  3:54:11 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.37-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150316  3:54:22 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.37-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150316 19:07:31 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.37-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150317  2:05:45 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.40-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150317  2:05:54 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.40-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150319  1:17:26 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.42-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150319  1:17:34 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.42-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150616  1:39:44 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.42-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

150616  1:39:52 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.42-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

151006  1:01:45 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.45-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

151006  1:01:54 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.45-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

151027  1:21:12 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.46-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160105  1:31:35 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.47-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160211  1:52:47 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.48-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160211  1:52:55 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.48-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160503  1:14:59 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160503  1:15:03 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160521 18:46:24 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160522 11:51:45 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160529 15:26:41 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160529 15:30:12 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160604 23:29:15 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.49-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160609 17:04:41 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.50-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160609 17:04:45 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.50-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  MySQL Community Server (GPL)

160619 12:21:58 [Note] /usr/sbin/mysqld: ready for connections.
Version: '5.5.50-cll'  socket: '/var/lib/mysql/mysql.sock'  port: 0  MySQL Community Server (GPL)

What is intriguing from this analysis is that several versions were skipped including .38, .39, .41, .43, .44. One may ask the question why?

For Clients

This leads to several questions of the strategy used in your organization for controlling outages of your MySQL infrastructure for upgrades or for other reasons.

  • What is an acceptable outage time?
  • What is the acceptable maintenance window to perform outages?
  • What is your release cadence for MySQL upgrades?
  • Who or what performs updates?
  • Can your monitoring detect small outages?

You should also consider in your business strategy having a highly available (HA) MySQL infrastructure to avoid any outage, or application intelligence to support varying levels of data access as I describe in Successful MySQL Scalability Principles.

Understanding the MySQL Release Cadence

At the recent New York Oracle Users Group summer general meeting I gave a presentation to the Oracle community on the MySQL product release cycle. Details included:

  • Identifying the server product options covering community, enterprise and ecosystem.
  • Describe MySQL enterprise products, features and support options.
  • Describing the DMR, RC, GA, EOL and labs product lifecycle.
  • Discussing the GA release frequency.
  • Talking about the MySQL Upgrade path.

Utilizing OpenStack Trove DBaaS for deployment management

Trove is used for self service provisioning and lifecycle management for relational and non-relational databases in an OpenStack cloud. Trove provides a RESTful API interface that is same regardless of the type of database. CLI tools and a web UI via Horizon are also provided wrapping Trove API requests.

In simple terms. You are a MySQL shop. You run a replication environment with daily backups and failover capabilities which you test and verify regularly. You have defined DBA and user credentials ACL’s across dev, test and prod environments. Now there is a request for using MongoDB or Cassandra, the engineering department has not decided but they want to evaluate the capabilities. How long as a operator does it take to acquire the software, install, configure, setup replication, backups, ACLs and enable the engineering department to evaluate the products?

With Trove DBaaS this complexity is eliminated due to a consistent interface to perform the provisioning, configuration, HA, B&R, ACL across other products the exact same way you perform these tasks for MySQL. This enables operations to be very proactive to changing technology requests supporting digital transformation strategies.

Enabling this capability is not an automatic approval of a new technology stack. It is important that strategic planning, support and management is included in the business strategy to understanding the DBaaS capability for your organization. Examples of operations due diligence would include how to integrate these products into your monitoring, logging and alerting systems. Determine what additional disk storage requirements may be needed. Test, verify and time recovery strategies.

Trove specifically leverages several other OpenStack services for source image and instance management. Each trove guest image includes a base operating system, the applicable database software and a database technology specific trove guest agent. This agent is the intelligence that knows the specific syntax and version needs to perform the tasks. The agent is also the communication mechanism between Trove and the running nova instance.

Trove is a total solution manager for the instance running your chosen database. Instances have no ssh, telnet or other general access. The only access is via the SQL communication via the defined ports, e.g. 3306 for MySQL.

The Trove lifecycle management covers the provisioning, management, security, configuration and tuning of your database. Amrith Kumar in a recent presentation at the NYC Postgres meetup provides a good description of the specifics.

Trove is capable of describing and supporting clustering and replication topologies for the various data stores. It can support backup and restore, failover and resizing of clusters without the operator needing to know the specific syntax of complexities of a database product they are unfamiliar with.

A great example is the subtle difference in MySQL replication management using GTID’s between MySQL and MariaDb. To the developer, the interaction between MySQL and MariaDB via SQL is the same, the management of a replication topology is not identical, but is managed by the Trove guest agent. To the operator, the management is the same.

Also in his presentation, Kumar described Tesora, an enterprise class Trove service provided with a number of important additional features. Tesora supports additional database products including Oracle and DB2Express as well as commercial versions for Oracle MySQL, EnterpriseDB, Couchbase, Datastax, and mongoDB. Using the Horizon UI customizations with pre-defined trove instances greatly reduces the work needed for operators and deployers to build there own.

Are you a responsible developer?

What is a good example of individual developer responsibility? Here is just one example.

A developer downloads a copy of the core production database to their own development laptop. Why? Because it’s easy to work with real data, and it’s hard to consider building applicable test data that all engineers can utilize.

What could be wrong with this approach? Here are a few additional points.

  • Security. Should the developer accidentally leave their laptop on that 90 minute train commute each way daily, could that data end up with a result of negative publicity for the organization. For employees that work at more sensitive organization is theft a possibility? Or, does that employee become disgruntled by lack of management and with poor ethical values take the names, emails, addresses and purchase history of your customers so it can be used for other means.
  • Data Clensing. This includes removing pay rate information of employees of the company that developers should never have access to. It is about obfuscating email address of millions of customers so that test code to improve receipt generate doesn’t accidentally email 1,000 existing customers with a repeat receipt that now contains invalid data. It is about providing a subset of information that is applicable and relevant.
  • Testing philosophy. Testing is all about trying to break your software, not testing that one small feature works in the likely path of use. It is easy to unit test the developer change for editing a customer profile to add a emergency contact field. It is right to consider the lifecycle of customer data. Is it knowing you need to consider the full workflow and the multiple paths to creating and editing a customer profile that causes the responsibility of the organization’s need to be consistent for the entire experience, not just one singular perspective . In simple terms it is about functionality testing at the time of development, not the narrow view of unit testing and that other detailed testing is somebody else’s responsibility.
  • Time. How long did it take to download the 10G dataset and import it? How much of that data is really needed. Does five years of historical products and orders ensure adequate unit and functional testing. Sure it is easy to have the available disk space however what efficiency improvements could exist for a data set 20x smaller. If it took five minutes to reset the test data for development instead of one hour would a developer refresh more often?

Before considering the means to meet an immediate problem such as this one example, stop, think, and act about improving the process for benefit of all technical resources. This is what sets apart an engineer that is just a coder and a software developer.

It is unfortunate that engineering managers are not constantly focussed on process and productivity improvements for sustaining software for the entire lifecycle of a product. The reality is many have worked as developers without applicable mentoring and management and an entire generation of software developers are now influencing the next generation. Historically, the rigidness of the traditional waterfall approach to the software development lifecycle instills a number of key principles that agile only environments have not fostered or understood.

Understanding the DBaaS capability for your organization

As your organization transforms to embrace the wealth of digital information that is becoming available, the capability to store, manage and consume this data in any given format or product becomes an increasing burden for operations.

How does your organization handle the request, “I need to use product Z to store data for my new project?” There are several responses I have experienced first-hand with clients.

  1. Enforce the company policy that Products O and S are all that can be used.
  2. Ignore the request.
  3. Consider the request, but antagonize your own internal organization with long wait times (e.g. months or years) and with repeated delays to evaluate a product you simply do not want to support.
  4. Do whatever the developers say, they know best.

Unfortunately I have seen too many organizations use the first three options as the answer. The last option you make consider as a non valid answer however I have also seen this prevalent when there is no operations team or strategic technical oversight.

Ignorance of the question only leads to a greater pressure point at a later time. This may be when your executive team now enforces the requirement with their timetable. I have seen this happen and with painful ramifications. With the ability to consume public cloud resources with only access to a credit card, development resources can now proceed unchecked more easily if ignored or delayed. When a successful proof of concept is produced this way and now a more urgent need is required to deploy, support and manage, the opportunity to have a positive impact on the design decision of a new data product has passed.

Using DBaaS is one enabling tool within a strategic business model for your organization to satisfy this question with greater control. This however is not the solution but rather one tool combined with applicable processes. In order to scope the requirements for the original question, your model also needs to consider the following:

  • Provisioning capabilities
  • Strategic planning and insight
  • Support and management
  • Release criteria


This is the strength of DBaaS. Operations can enable development to independently provision resources and technology with little additional impeding dependency. There is input from operations to enable varying products to be available by self servicing, however there is also some control. DBaaS can be viewed as a controlled and flexible enabler. A specific example is an organization that uses the MySQL relational database, and now a developer wishes to use the MongoDB NoSQL unstructured store. An operator may cringe at the notion of a lack of data consistency, structure data query access or performance capabilities. These are all valid points, however those are discussions at a strategic level discussion your workflow pipeline and should not be an impediment to iterate quickly. Without oversight, to iterate quickly can lead to unmanageable outcomes.

Strategic Planning

There always needs to be oversight combined with applicable strategy. A single developer stating they want to use the new product Z for a distribution key/value store needs to be vetted first within the engineering organization and its own developer peers. If another project is already using Product Y that has the same core data access and features, this burden of an additional product support should be a self contained discussion validating the need first.

This is one strength of a good engineering manager that balances the requirements of the business needs and objectives with the capabilities of the resources available, including staff, tools and technology. Applicable principles put in place should also ensure that some aspect of planning is instilled into the development culture.

Support and Management

The development and engineering resources rarely consider the administration and support required for the suite of products and services used in an organization. The emphasis is on feature development and customer requirements, not the sustainability, longevity and security of any system. Operational support is a long list of needs, just a few include:

  • Information security.
  • Information availability.
  • Service level agreements (SLAs) between partners, service providers and the internal organization
  • The backup ecosystem, time taken, consistency, point-in-time recovery, testing and verification, cost of H/W, S/W, licenses.
  • Internet connectivity.
  • Capacity planning and cost analysis of storing and archiving ever increasing sources of data.
  • Hardware and software upgrades.

Two way communication which is often overlooked is the start of better understanding. That is, operations being included and involved in strategic development planning, and engineering resources included in operations needs and requirements for ensuring those new product features operate for the benefit of customers. In summary, “bridging the communication chasm”.

DevOps is an abused term, this implies that developers now perform a subset of responsibilities of Operations. As an individual that has worked in both development teams and lead operations teams, your resources skills, personality, rational thinking and decision making needs are vastly different between an engineering task and a production operations task.

Developers need to live a 24 hour day (with the unnecessary 3am emergency call) in the shoes of an operator. The reverse is also true, however the ramifications to business continuity are not the same. Just one factor, the cost, or more specifically the loss to the business due to a production failure alters the decision making process. Failure can be anything from a hardware or connectivity problem, bad code that was released to a data breach.

Release Criteria

If an organization has a strong (and flexible) policy on release criteria, all parties from the stack-holder, executive, engineering, operations and marketing should be able to contribute to the discussion and decision for a new product, and applicable in-house or third-party support. This discussion is not a pre-requisite for any department or developer to iterate quickly, however it is pre-requisite to migrate from a proof-of-concept prototype to a supported feature. Another often overlooked criteria in the pursuit for rapid deployment of new features which are significantly more difficult to remove after publication.

Expired MySQL passwords

I was surprised to find on one of my websites the message “Connect failed: Your password has expired. To log in you must change it using a client that supports expired passwords.

Not knowing that I was using a MySQL password expiry policy I reviewed the 5.7 documentation quickly which *clearly* states “The default default_password_lifetime value is 0, which disables automatic password expiration.”.

I then proceeded to investigate further, my steps are below the following comment.

However, it is always important with MySQL documentation and a new feature (in this case a 5.7 feature) to review release notes when installing versions or to least read ALL the documentation, because you may miss important information, such as.

From MySQL 5.7.4 to 5.7.10, the default default_password_lifetime value is 360 (passwords must be changed approximately once per year). For those versions, be aware that, if you make no changes to the default_password_lifetime variable or to individual user accounts, all user passwords will expire after 360 days, and all user accounts will start running in restricted mode when this happens. Clients…

I would encourage you to view the MySQL password expiry policy to see the full note. I have only include the intro here are a teaser, because you need to read the entire note.


Back to impatient analysis steps.

$ mysql -u admin -p 

| 5.7.9-log |

| Variable_name             | Value |
| default_password_lifetime | 360   |

SELECT host,user,password_last_changed 
FROM mysql.user
WHERE password_last_changed + INTERVAL @@default_password_lifetime DAY < CURDATE();
| host      | user         | password_last_changed |
| localhost | XXX          | 2014-12-01 12:53:36   |
| localhost | XXXXX        | 2014-12-01 12:54:04   |
| localhost | XX_XXXX      | 2015-06-04 11:01:11   |

Indeed there are some passwords that have expired.

After finding the applicable application credentials I looked at verifying the problem.

$ mysql -uXX_XXXX -p
Enter password: 
Welcome to the MySQL monitor.  Commands end with ; or \g.
Server version: 5.7.9-log


Interesting, there was no error to make a client connection, however.

mysql> use XXXX;
ERROR 1820 (HY000): You must reset your password using ALTER USER statement before executing this statement.

I then proceeded to change the password with the applicable hint shown.

ALTER USER XX_XXXX@localhost IDENTIFIED BY '*************************';

I chose to reuse the same password because changing the password would require a subsequent code change. MySQL accepted the same password. (A topic for a separate discussion on this point).

A manual verification showed the application and users operating as it should be, so immediate loss of data was averted. Monitoring of the sites home page however did not detect this problem of a partial page error, so should a password expiry policy be used, an applicable check in a regularly scheduled operational task is a good feature request.

All of this could have been avoided if my analysis started with reading the documentation and the note (partly shown above) which has an alternative and potentially more practical immediate solution.

In a firefighting operational mode it can be a priority to correct the problem, however more detailed analysis is prudent to maintain a "Being proactive rather than reactive" mindset. Being a Friday, I feel the old saying "There is more than one way to skin a cat" is applicable.

I am also more familiar with the SET PASSWORD syntax, so reviewing this 5.7 manual page is also a good read to determine what specific syntax is now deprecated and what "ALTER USER is now the preferred statement for assigning passwords" also.