There are a few interesting scenarios to keep in mind when writing applications (not just databases!) that interact with read and writes files, particularly in transactional contexts where you actually care about the integrity of the data and when you are editing data in place (versus copy-on-write for example). If I don't say otherwise I'm talking about behavior on Linux. The research version of this blog post is Parity Lost and Parity Regained and Characteristics, Impact, and Tolerance of Partial Disk Failures. These two papers also go into the frequency of some of the issues discussed here. These behaviors actually happen in real life! Thank you to Alex Miller and George Xanthakis for reviewing a draft of this post. Terminology Some of these terms are reused in different contexts, and sometimes they are reused because they effectively mean the same thing in a certain configuration. But I'll try to be explicit to avoid confusion. Sector The smallest amount of data that can be read and written atomically by hardware. It used to be 512 bytes, but on modern disks it is often 4KiB. There doesn't seem to be any safe assumption you can make about sector size, despite file system defaults (see below). You must check your disks to know. Block (filesystem/kernel view) Typically set to the sector size since only this block size is atomic. The default in ext4 is 4KiB. Page (kernel view) A disk block that is in memory. Any reads/writes less than the size of a block will read the entire block into kernel memory even if less than that amount is sent back to userland. Page (database/application view) The smallest amount of data the system (database, application, etc.) chooses to act on, when it's read or written or held in memory. The page size is some multiple of the filesystem/kernel block size (including the multiple being 1). SQLite's default page size is 4KiB. MySQL's default page size is 16KiB. Postgres's default page size is 8KiB. Things that go wrong The data didn't reach disk By default, file writes succeed when the data is copied into kernel memory (buffered IO). The man page for write(2) says: A successful return from write() does not make any guarantee that data has been committed to disk. On some filesystems, including NFS, it does not even guarantee that space has successfully been reserved for the data. In this case, some errors might be delayed until a future write(), fsync(2), or even close(2). The only way to be sure is to call fsync(2) after you are done writing all your data. If you don't call fsync on Linux the data isn't necessarily durably on disk, and if the system crashes or restarts before the disk writes the data to non-volatile storage, you may lose data. With O_DIRECT, file writes succeed when the data is copied to at least the disk cache. Alternatively you could open the file with O_DIRECT|O_SYNC (or O_DIRECT|O_DSYNC) and forgo fsync calls. fsync on macOS is a no-op. If you're confused, read Userland Disk I/O. Postgres, SQLite, MongoDB, MySQL fsync data before considering a transaction successful by default. RocksDB does not. The data was fsynced but fsync failed fsync isn't guaranteed to succeed. And when it fails you can't tell which write failed. It may not even be a failure of a write to a file that your process opened: Ideally, the kernel would report errors only on file descriptions on which writes were done that subsequently failed to be written back. The generic pagecache infrastructure does not track the file descriptions that have dirtied each individual page however, so determining which file descriptors should get back an error is not possible. Instead, the generic writeback error tracking infrastructure in the kernel settles for reporting errors to fsync on all file descriptions that were open at the time that the error occurred. In a situation with multiple writers, all of them will get back an error on a subsequent fsync, even if all of the writes done through that particular file descriptor succeeded (or even if there were no writes on that file descriptor at all). Don't be 2018-era Postgres. The only way to have known which exact write failed would be to open the file with O_DIRECT|O_SYNC (or O_DIRECT|O_DSYNC), though this is not the only way to handle fsync failures. The data was corrupted If you don't checksum your data on write and check the checksum on read (as well as periodic scrubbing a la ZFS) you will never be aware if and when the data gets corrupted and you will have to restore (who knows how far back in time) from backups if and when you notice. ZFS, MongoDB (WiredTiger), MySQL (InnoDB), and RocksDB checksum data by default. Postgres and SQLite do not (though databases created from Postgres 18+ will). You should probably turn on checksums on any system that supports it, regardless of the default. The data was partially written Only when the page size you write = block size of your filesystem = sector size of your disk is a write guaranteed to be atomic. If you need to write multiple sectors of data atomically there is the risk that some sectors are written and then the system crashes or restarts. This is called torn writes or torn pages. Postgres, SQLite, and MySQL (InnoDB) handle torn writes. Torn writes are by definition not relevant to immutable storage systems like RocksDB (and other LSM Tree or Copy-on-Write systems like MongoDB (WiredTiger)) unless writes (that update metadata) span sectors. If your file system duplicates all writes like MySQL (InnoDB) does (like you can with data=journal in ext4) you may also not have to worry about torn writes. On the other hand, this amplifies writes 2x. The data didn't reach disk, part 2 Sometimes fsync succeeds but the data isn't actually on disk because the disk is lying. These are called lost writes or phantom writes. You can be resilient to phantom writes by always reading back what you wrote (expensive) or versioning what you wrote. Databases and file systems generally do not seem to handle this situation. The data was written to the wrong place, read from the wrong place If you aren't including where data is supposed to be on disk as part of the checksum or page itself, you risk being unaware that you wrote data to the wrong place or that you read from the wrong place. This is called misdirected writes/reads. Databases and file systems generally do not seem to handle this situation. Further reading In increasing levels of paranoia (laudatory) follow ZFS, Andrea and Remzi Arpaci-Dusseau, and TigerBeetle.

This is an external post of mine. Click here if you are not redirected.

Last month I completed my first year at EnterpriseDB. I'm on the team that built and maintains pglogical and who, over the years, contributed a good chunk of the logical replication functionality that exists in community Postgres. Most of my work, our work, is in C and Rust with tests in Perl and Python. Our focus these days is a descendant of pglogical called Postgres Distributed which supports replicating DDL, tunable consistency across the cluster, etc. This post is about how I got here. Black boxes I was a web developer from 2014-2021†. I wrote JavaScript and HTML and CSS and whatever server-side language: Python or Go or PHP. I was a hands-on engineering manager from 2017-2021. I was pretty clueless about databases and indeed database knowledge was not a serious part of any interview I did. Throughout that time (2014-2021) I wanted to move my career forward as quickly as possible so I spent much of my free time doing educational projects and writing about them on this blog (or previous incarnations of it). I learned how to write primitive HTTP servers, how to write little parsers and interpreters and compilers. It was a virtuous cycle because the internet (Hacker News anyway) liked reading these posts and I wanted to learn how the black boxes worked. But I shied away from data structures and algorithms (DSA) because they seemed complicated and useless to the work that I did. That is, until 2020 when an inbox page I built started loading more and more slowly as the inbox grew. My coworker pointed me at Use The Index, Luke and the DSA scales fell from my eyes. I wanted to understand this new black box so I built a little in-memory SQL database with support for indexes. I'm a college dropout so even while I was interested in compilers and interpreters earlier in my career I never dreamed I could get a job working on them. Only geniuses and PhDs did that work and I was neither. The idea of working on a database felt the same. However, I could work on little database side projects like I had done before on other topics, so I did. Or a series of explorations of Raft implementations, others' and my own. Startups From 2021-2023 I tried to start a company and when that didn't pan out I joined TigerBeetle as a cofounder to work on marketing and community. It was during this time I started the Software Internals Discord and /r/databasedevelopment which have since kind of exploded in popularity among professionals and academics in database and distributed systems. TigerBeetle was my first job at a database company, and while I contributed bits of code I was not a developer there. It was a way into the space. And indeed it was an incredible learning experience both on the cofounder side and on the database side. I wrote articles with King and Joran that helped teach and affirm for myself the basics of databases and consensus-based distributed systems. Holding out When I left TigerBeetle in 2023 I was still not sure if I could get a job as an actual database developer. My network had exploded since 2021 (when I started my own company that didn't pan out) so I had no trouble getting referrals at database companies. But my background kept leading hiring managers to suggest putting me on cloud teams doing orchestration in Go around a database rather than working on the database itself. I was unhappy with this type-casting so I held out while unemployed and continued to write posts and host virtual hackweeks messing with Postgres and MySQL. I started the first incarnation of the Software Internals Book Club during this time, reading Designing Data Intensive Applications with 5-10 other developers in Bryant Park. During this time I also started the NYC Systems Coffee Club. Postgres After about four months of searching I ended up with three good offers, all to do C and Rust development on Postgres (extensions) as an individual contributor. Working on extensions might sound like the definition of not-sexy, but Postgres APIs are so loosely abstracted it's really as if you're working on Postgres itself. You can mess with almost anything in Postgres so you have to be very aware of what you're doing. And when you can't mess with something in Postgres because an API doesn't yet exist, companies have the tendency to just fork Postgres so they can. (This tendency isn't specific to Postgres, almost every open-source database company seems to have a long-running internal fork or two of the database.) EnterpriseDB Two of the three offers were from early-stage startups and after more than 3 years being part of the earliest stages of startups I was happy for a break. But the third offer was from one of the biggest contributors to Postgres, a 20-year old company called EnterpriseDB. (You can probably come up with different rankings of companies using different metrics so I'm only saying EnterpriseDB is one of the biggest contributors.) It seemed like the best place to be to learn a lot and contribute something meaningful. My coworkers are a mix of Postgres veterans (people who contributed the WAL to Postgres, who contributed MVCC to Postgres, who contributed logical decoding and logical replication, who contributed parallel queries; the list goes on and on) but also my developer-coworkers are people who started at EnterpriseDB on technical support, or who were previously Postgres administrators. It's quite a mix. Relatively few geniuses or PhDs, despite what I used to think, but they certainly work hard and have hard-earned experience. Anyway, I've now been working at EnterpriseDB for over a year so I wanted to share this retrospective. I also wanted to cover what it's like coming from engineering management and founding companies to going back to being an individual contributor. (Spoiler: incredibly enjoyable.) But it has been hard enough to make myself write this much so I'm calling it a day. :) I wrote a post about the winding path I took from web developer to database developer over 10 years. pic.twitter.com/tf8bUDRzjV — Phil Eaton (@eatonphil) February 15, 2025 † From 2011-2014 I also did contract web development but this was part-time while I was in school.

I have the fortune to review a few important blog posts every year and the biggest value I add is to call out sentences or sections that make no sense. It is quite simple and you can do it too. Without clarity only those at your company in marketing and sales (whose job it is to work with what they get) will give you the courtesy of a cursory read and a like on LinkedIn. This is all that most corporate writing achieves. It is the norm and it is understandable. But if you want to reach an audience beyond those folks, you have to make sure you're not writing nonsense. And you, as reviewer and editor, have the chance to call out nonsense if you can get yourself to recognize it. Immune to nonsense But especially when editing blog posts at work, it is easy to gloss over things that make no sense because we are so constantly bombarded by things that make no sense. Maybe it's buzzwords or cliches, or simply lack of rapport. We become immune to nonsense. And even worse, without care, as we become more experienced, we become more fearful to say "I have no idea what you are talking about". We're afraid to look incompetent by admitting our confusion. This fear is understandable, but is itself stupid. And I will trust you to deal with this on your own. Read it out loud So as you review a post, read it out loud to yourself. And if you find yourself saying "what on earth are you talking about", add that as a comment as gently as you feel you should. It is not offensive to say this (depending on how you say it). It is surely the case that the author did not know they were making no sense. It is worse to not mention your confusion and allow the author to look like an idiot or a bore. Once you can call out what does not make sense to you, then read the post again and consider what would not make sense to someone without the context you have. Someone outside your company. Of course you need to make assumptions about the audience to a degree. It is likely your customers or prospects you have in mind. Not your friends or family. With the audience you have in mind, would what you're reading make any sense? Has the author given sufficient background or introduced relevant concepts before bringing up something new? Again this is a second step though. The first step is to make sure that the post makes sense to you. In almost every draft I read, at my company or not, there is something that does not make sense to me. Do two paragraphs need to be reordered because the first one accidentally depended on information mentioned in the second? Are you making ambiguous use of pronouns? And so on. In closing Clarity on its own will put you in the 99th percentile of writing. Beyond that it definitely still matters if you are compelling and original and whatnot. But too often it seems we focus on being exciting rather than being clear. But it doesn't matter if you've got something exciting if it makes no sense to your reader. This sounds like mundane guidance, but I have reviewed many posts that were reviewed by other people and no one else called out nonsense. I feel compelled to mention how important it is. Wrote a new post on the most important, and perhaps least done, thing you can do while reviewing a blog post: edit for clarity. pic.twitter.com/ODblOUzB3g — Phil Eaton (@eatonphil) January 29, 2025