I’ve found this article lacking. Like some other articles in this space, it introduces isolation levels through the lens of the phenomena described in the SQL standard, but I find that there’s a different, more intuitive approach.
I think it’s more tractable to define this problem space starting from the concept of (strict) serializability, which is really a generalization of the concept of thread safety. Every software engineer has an intuitive understanding of it. Lack of serializability can lead to execution-dependent behavior, which usually results in hard-to-diagnose bugs. Thus, all systems should strive towards serializability, and the database can be a tool in achieving it.
Various non-serializable levels of database transaction isolation are relaxations of the serializability guarantee, where the database no longer enforces the guarantee and it’s up to the database user to ensure it through other means.
The isolation phenomena are a useful tool for visualizing various corner cases of non-serializability, but they are not inherently tied to it. It's possible to achieve serializability while observing all of the SQL phenomena. For example, a Kubernetes cluster with carefully-written controllers can be serializable.
I think this is a great post to have but I'm going to make a critical usability suggestion:
* the videos should have "pause" and a "step at a time" control *
Even at the "half speed", without a deep knowledge of the context, the videos move way too fast for me to read the syntax that's invoking and line it up with the data on the left side. I (and im definitely not the only one) need to be able to sit on one step and stare at the whole thing without the latent anxiety of the state changing before I've had a chance to grok the whole thing.
this has nothing to do with familiarity with the concepts (read my profile). I literally need time to read all the words and connect them together mentally (ooh, just noticed this is pseudo-SQL syntax also, e.g. "select id=4", that probably added some load for me) without worrying they're going to change before watching things move.
A lot of database tools these days prioritize instant sharing of updates over transactions and ACID properties. Example: Airtable. As soon as you update a field the update shows up on your coworkers screen who also has the same table open. The downside of this is that Airtable doesn't do transactions. And the downside of not doing transactions is potentially dangerous data inconsistencies. More about that here: https://visualdb.com/blog/concurrencycontrol/
This actually used to be one of my favorite interview questions for backend engineers. Everyone has used transactions but depending on your seniority you'd understand it to different degrees.
And no I'd never expect people to know the isolation levels by heart, but if you know there are different ones and they behave differntly that's pretty good and tells me you are curious about how things work under the hood.
> At this stage, it has nothing to do with xmin and xmax, but rather because other transactions cannot see uncommitted data
Am I missing something or this statement is incomplete? Also I find the visualization of commit weird, it “points to” the header of the table, but then xmax gets updated “behind the scenes”? Isnt xmax/xmin “the mechanism behind how the database knows what is committed/not committed”? Also, there could be subtransactions, which make this statement even more contradictory?
I enjoyed the visualizations and explanations otherwise, thanks!
Seems like a frequent surprise is that Postgres and MySQL don't default to serializable (so not fully I in ACID). They do read-committed. I didn't see this article mention that, but maybe I missed it. The article says read-committed provides "slightly" better performance, but it's been way faster in my experience. Forget where, but I think they said they chose this default for that reason.
Using read-committed ofc means having to keep locking details in mind. Like, UNIQUE doesn't just guard against bad data entry, it can also be necessary for avoiding race conditions. But now that I know, I'd rather do that than take the serializable performance hit, and also have to retry xacts and deal with the other caveats at the bottom of https://www.postgresql.org/docs/current/transaction-iso.html
Have you ever seen anyone changing transaction isolation levels in code? I think pessimistic or optimistic locking is preferred way to handle transaction concurrency.
But are you sure it was correct? Typical web apps are riddled with race conditions due to incorrect use of database locking and transactions, in my experience.
I thought this was pretty good, not least because it attempts to explain isolation levels, something I always found pretty tricky when teaching SQL. Mind you, I was only teaching SQL, and so isolation, as part of C and C++ courses so that our clients could do useful stuff, but explaining what levels to use was always tuff.
Is it just me, or are the final results of the deadlock visualisations incorrect? In both animations (mysql/pg), the final `SELECT balance from account...` queries appear to show the result of the two sessions, which have been terminated.
never used planetscale but I’ve always liked their blog, and other content. One of the founders interviewed on the software engineering daily podcast and it was super interesting
> A phantom read is one where a transaction runs the same SELECT multiple times, but sees different results the second time around
> Under the SQL standard, the repeatable read level allows phantom reads, though in Postgres they still aren't possible.
This is bad wording which could lead to an impression that a repeatable read may show different values. Values in rows will be the same but new rows may be added to the second result set. New rows is important as no previously read rows can be either changed or deleted — otherwise there will be no repetition for those rows second time around.
I like to think of transactions, in an MVCC system like Postgres, as being like snapshots in copy-on-write filesystems like btrfs or zfs. When you BEGIN a transaction, the DB takes a snapshot of your data, so now there are two versions of the data, the snapshot (visible to everyone else) and the "private" version visible only to your transaction. Then as you run UPDATEs, the new data is written to the private copy, but everyone else continues to work with the snapshot. (And might be creating their own private copies for other transactions).
If you do a ROLLBACK, then your private copy of the data is discarded, and its changes never make it into the official copy. But if you do a COMMIT, then your private snapshot is made public and is the new, official, copy for everyone else to read from. (Except those who started a transaction before you ran COMMIT: they made their private copies from the older snapshot and don't have a copy of your changes).
This is probably obvious to nearly everyone here, but I figured I'd write it anyway. You never know who might read an analogy like this and have that lightbulb moment where it suddenly makes sense.
P.S. Another analogy would be Git branches, but I'll write that in a different comment.
One way to think about transactions, as I wrote in an earlier comment, would be to think of them as being like snapshots in a copy-on-write filesystem like btrfs or zfs. But another way to think of them is being like Git branches.
When you BEGIN a transaction, you're creating a branch in Git. Everyone else continues to work on the master branch, perhaps making their own branches (transactions) off of it while you're working. Every UPDATE command you run inside the transaction is a commit pushed to your branch. If you do a ROLLBACK, then you're deleting the branch unmerged, and its changes will be discarded without ever ending up in the master branch. But if you instead do a COMMIT, then that's a `git merge` command, and your changes will be merged into the master branch. If they merge cleanly, then all is well. If they do NOT merge cleanly, because someone else merged their own branch (committed their own transaction) that touched the same files that you touched (updated rows in the same table), then the DB will go through the file line by line (go through the table row by row) to try to get a clean merge. If it can successfully merge both changes without conflict, great. If it can't, then what happens depends on the transaction settings you chose. You can, when you start the transaction, tell the DB "If this doesn't merge cleanly, roll it back". Or you can say "If this doesn't merge cleanly, I don't care, just make sure it gets merged and I don't care if the conflict resolution ends up picking the "wrong" value, because for my use case there is no wrong value." This is like using "READ UNCOMMITTED" vs "SERIALIZABLE" transaction settings (isolation levels): you would use "READ UNCOMMITTED" if you don't care about merge conflicts in this particular table, and just want a quick merge. You would use "SERIALIZABLE" for tables with data that must, MUST, be correct, e.g. account balances. And there are two more levels in between for subtle differences in your use case's requirements.
As with my previous comment, this is probably obvious to 98.5% of people here. But maybe it'll help someone get that "ah-ha!" moment and understand transactions better.
21 comments
[ 3.7 ms ] story [ 48.1 ms ] threadI think it’s more tractable to define this problem space starting from the concept of (strict) serializability, which is really a generalization of the concept of thread safety. Every software engineer has an intuitive understanding of it. Lack of serializability can lead to execution-dependent behavior, which usually results in hard-to-diagnose bugs. Thus, all systems should strive towards serializability, and the database can be a tool in achieving it.
Various non-serializable levels of database transaction isolation are relaxations of the serializability guarantee, where the database no longer enforces the guarantee and it’s up to the database user to ensure it through other means.
The isolation phenomena are a useful tool for visualizing various corner cases of non-serializability, but they are not inherently tied to it. It's possible to achieve serializability while observing all of the SQL phenomena. For example, a Kubernetes cluster with carefully-written controllers can be serializable.
* the videos should have "pause" and a "step at a time" control *
Even at the "half speed", without a deep knowledge of the context, the videos move way too fast for me to read the syntax that's invoking and line it up with the data on the left side. I (and im definitely not the only one) need to be able to sit on one step and stare at the whole thing without the latent anxiety of the state changing before I've had a chance to grok the whole thing.
this has nothing to do with familiarity with the concepts (read my profile). I literally need time to read all the words and connect them together mentally (ooh, just noticed this is pseudo-SQL syntax also, e.g. "select id=4", that probably added some load for me) without worrying they're going to change before watching things move.
please add a step-at-a-time button!
And no I'd never expect people to know the isolation levels by heart, but if you know there are different ones and they behave differntly that's pretty good and tells me you are curious about how things work under the hood.
Am I missing something or this statement is incomplete? Also I find the visualization of commit weird, it “points to” the header of the table, but then xmax gets updated “behind the scenes”? Isnt xmax/xmin “the mechanism behind how the database knows what is committed/not committed”? Also, there could be subtransactions, which make this statement even more contradictory?
I enjoyed the visualizations and explanations otherwise, thanks!
Using read-committed ofc means having to keep locking details in mind. Like, UNIQUE doesn't just guard against bad data entry, it can also be necessary for avoiding race conditions. But now that I know, I'd rather do that than take the serializable performance hit, and also have to retry xacts and deal with the other caveats at the bottom of https://www.postgresql.org/docs/current/transaction-iso.html
It goes into not only different isolation levels, but also some ambiguity in the traditional ACID definition.
I believe a 2nd edition is imminent.
> Under the SQL standard, the repeatable read level allows phantom reads, though in Postgres they still aren't possible.
This is bad wording which could lead to an impression that a repeatable read may show different values. Values in rows will be the same but new rows may be added to the second result set. New rows is important as no previously read rows can be either changed or deleted — otherwise there will be no repetition for those rows second time around.
If you do a ROLLBACK, then your private copy of the data is discarded, and its changes never make it into the official copy. But if you do a COMMIT, then your private snapshot is made public and is the new, official, copy for everyone else to read from. (Except those who started a transaction before you ran COMMIT: they made their private copies from the older snapshot and don't have a copy of your changes).
This is probably obvious to nearly everyone here, but I figured I'd write it anyway. You never know who might read an analogy like this and have that lightbulb moment where it suddenly makes sense.
P.S. Another analogy would be Git branches, but I'll write that in a different comment.
When you BEGIN a transaction, you're creating a branch in Git. Everyone else continues to work on the master branch, perhaps making their own branches (transactions) off of it while you're working. Every UPDATE command you run inside the transaction is a commit pushed to your branch. If you do a ROLLBACK, then you're deleting the branch unmerged, and its changes will be discarded without ever ending up in the master branch. But if you instead do a COMMIT, then that's a `git merge` command, and your changes will be merged into the master branch. If they merge cleanly, then all is well. If they do NOT merge cleanly, because someone else merged their own branch (committed their own transaction) that touched the same files that you touched (updated rows in the same table), then the DB will go through the file line by line (go through the table row by row) to try to get a clean merge. If it can successfully merge both changes without conflict, great. If it can't, then what happens depends on the transaction settings you chose. You can, when you start the transaction, tell the DB "If this doesn't merge cleanly, roll it back". Or you can say "If this doesn't merge cleanly, I don't care, just make sure it gets merged and I don't care if the conflict resolution ends up picking the "wrong" value, because for my use case there is no wrong value." This is like using "READ UNCOMMITTED" vs "SERIALIZABLE" transaction settings (isolation levels): you would use "READ UNCOMMITTED" if you don't care about merge conflicts in this particular table, and just want a quick merge. You would use "SERIALIZABLE" for tables with data that must, MUST, be correct, e.g. account balances. And there are two more levels in between for subtle differences in your use case's requirements.
As with my previous comment, this is probably obvious to 98.5% of people here. But maybe it'll help someone get that "ah-ha!" moment and understand transactions better.