Let’s start with the obvious: what does “ACID compliant” mean? The short answer is that ACID, an acronym for “Atomicity, Consistency, Isolation, and Durability,” is a set of principles that ensure database transactions are processed reliably. When any data storage system upholds those principles, it is said to be ACID compliant.
ACID compliance is valuable because those four strong guarantees provide reliability, correctness, and other essential features. We’ll examine what this means in more detail later. Still, the high-level overview is that ACID compliant database operations must always succeed or fail in reliable and expected ways. In contrast, the database as a whole must always remain in a consistent, well-defined state at all times—regardless of any software errors, hardware errors, or even complete power failures that occur along the way.
Consider the quick and common example of a bank customer using an app on a mobile device to transfer money from one account to another. A complete and correct transaction requires both debiting the origin account and crediting the destination account.
But what if the power fails to the database server after the origin account is debited yet before the destination account is credited? ACID compliance ensures that in such a case the first part of the overall transaction is canceled (or “rolled back”) because the second part didn’t go through, keeping the database in a consistent state in the case of such a failure.
In this article, we’ll explain the four elements of ACID compliance, discuss the common difficulties involved with maintaining those elements when data and transactions are distributed in modern systems and offer some specific suggestions for choosing an ACID compliant storage system to meet your needs.
What Are the Elements of ACID Compliance?
Before we dig into the weeds, it’s worth noting that the elements or properties of ACID compliance generally fall on a sliding scale, from permissive to strict. Weaker guarantees typically produce a faster performance at the cost of correctness, whereas strong guarantees sacrifice speed for correctness and data reliability.
Let’s start with the property of atomicity, which refers to how a given set of database operations must all either complete successfully or fail altogether. This is what transaction means: a complete and indivisible unit of work that may involve many different data operations. Atomicity requires that all those data operations either succeed or fail as a whole.
Returning to our earlier banking example illustrates this nicely: both the debit and credit operations must be completed for the entire transaction to be a success. If either individual operation fails, then the entire transaction must fail and leave the state of the database exactly as it was before the transaction began.
In contrast, consistency is about ensuring the information in the database is always meaningful. Many ACID compliant database systems offer features to enforce various types of constraints, which can require a given piece of data to meet certain requirements or guarantee that a given piece of “parent” information may be linked relationally to many different pieces of “child” information yet prevent child data from being entered without a valid parent. Consistency ensures that if any violation of such data integrity checks (or others) occurs, the entire containing transaction will be aborted, and any changes already made will be rolled back. There’s much more that can be said about consistency
, but we can return to our previous banking example for simplicity for this article.
If the customer’s account transfer would cause an “overdraft” to occur (e.g., an attempt to move more money than is actually in the account), it’s not hard to see why the bank might want to prevent negative balances from happening entirely or perhaps perform additional processing after the fact to assign fees or other penalties. In this sense, consistency differs from the other ACID elements in that it’s largely the responsibility of the application developer(s) to ensure the code they place in transactions does not violate such rules of business and enforces all desired constraints.
The third element of ACID compliance, isolation, refers to the need to separate the details of multiple transactions in process at the same time. To see why, let’s refer back to our example and ask a question: what happens if some automated process that affects the user’s balance is triggered while the user is in the middle of the account transfer? It’s surprising how often this kind of event-driven scenario can occur in the real world.
If that process were to read the balance in the user’s origin account after the debit occurred, yet the account transfer was to fail, then the automated process would move forward based on bad information. In effect, the details of a transaction doomed to fail could “leak” into some other transaction that refers to the same information if the database doesn’t maintain strict isolation between them.
Again, much more can be said about both isolation and consistency levels
. Still, the key is that transactions remain fully “serializable,” each transaction occurring in its proper order without any dependent transactions occurring in tandem to avoid such information leaks and other anomalies. Database systems accomplish this using various tools, but ordering transactions in a global queue for processing is a common approach. Isolation is what allows multiple transactions to execute both simultaneously and safely for greater overall throughput.
Finally, we come to durability, which is perhaps the easiest ACID compliance guarantees to understand. Put simply, durability requires that every successful transaction’s data operations are indelible and cannot be lost—even in the event of a complete system shutdown.
This can be a tricky goal, but it’s commonly achieved by writing the changes to be made to a series of log files before the data operations are committed to the database files. That way, even if the whole database server crashes, those log files will be available to read on startup and either continue the transaction or roll back any changes that had already been made.
It’s easy to see why this too is important, insofar as banking customers who move money from one account to another, perhaps to pay off a loan or some other bill, except that the money will stay where they put it. Nobody wins if the bill marked “paid” doesn’t stay paid because of some hiccup in the database software, network connection, etc.
ACID and Distributed Transactions
Having explained the elements of ACID compliance, let’s look at how maintaining these important guarantees is complicated by modern systems. It is increasingly the case that applications need to store data reliably across multiple servers, perhaps even distributed geographically around the world, which poses significant challenges. It’s easy to understand how a simple, single database server might enforce ACID compliance. But how do you do that when the data operations in a transaction might conceivably be spread over dozens (or more) servers across multiple time zones?
Any failure of any element of any data operation on any of those servers effectively requires that the entire transaction must be canceled and rolled back safely. That’s hard enough when you’re talking about a single piece of software running on a single server. It’s far worse when multiple pieces of software running on multiple servers are processing their individual bits of the transaction and have to be notified, stopped, rolled back, etc. And similarly, even if the transaction succeeds, the overall system has to deal with making sure all those different operations are both correct and durable regardless of which server(s) executed them.
As a matter of practice, ACID compliance with distributed transactions is very difficult to achieve but is doable, at least with some vendors. For example, Fauna
is able to provide strictly serializable, externally consistent transactions because of its architecture and data storage algorithms.
And unlike other systems, Fauna does not require strict, physical clock synchronization across all servers to provide consistency, which avoids the usual limitations on distance between replica servers and is thus practical for deployment around the world at typical, global Internet latencies. Approaches that do require synchronization can result in failures when systems’ clocks or network traffic differs by a matter of milliseconds, whereas Fauna’s more relaxed requirements don’t suffer from such problems. This is possible because Fauna offers a transaction engine inspired by Calvin, an approach to achieving fast distributed transactions across partitioned database systems
. The Fauna transaction engine makes it possible to achieve ”consistency without clocks” courtesy of its distributed transaction protocol
. In effect, Fauna decides in advance in what order transactions should be executed prior to any database writes. The Fauna execution engine then processes them in such a way that the final result is the same as if they’d been processed one at a time in that order.
In effect, you get all the speed and power of distributed transactions executing in parallel on multiple servers yet enjoy all the data-goodness of ACID compliance as if they’d been executed serially on a single server.
ACID compliance in Fauna
To be clear, Fauna is not the usual hosted “database as a service” (DBaaS)
or even some clustered cloud offering, both of which require management. Rather, Fauna is a true “Data API”, meaning developers can simply make calls as needed without any time spent worrying about provisioning or scale, with all the benefits of ACID compliance along the way.
Because of the way Fauna handles distributed transactions, users can avoid the sort of data anomalies that can occur with other systems
. Immortal writes, stale reads, causal reverse, and other such problems can be prevented via strictly serializable multi-region transactions that don't limit the number of keys, documents, or partitions.
And because Fauna is free of all the usual provisioning and configuration headaches, it’s available instantly as a serverless utility. Developers need only an account to get started, which costs nothing and offers a generous starting allowance.
Fauna is also Jepsen tested and approved
, being proven “…architecturally sound, correctly implemented, and ready for enterprise workloads in the cloud.” That means you can trust Fauna to be reliable, correct in its operations, and self-maintaining without the need for provisioning, regular maintenance, etc.
In this article, we’ve explained what ACID compliance means, illustrated the value it brings for common use cases, discussed how the modern need for distributed transactions complicates the picture and offered some specific advice for an easy path to ACID compliance. Fauna is a modern, Data API that provides relational capabilities and ACID properties without the usual limitations or headaches.