SQLite doesn't run a server like Postgres, so the trick is moving the polling source from interval queries on a SQLite connection to a lightweight stat(2) on the WAL file. Many small queries are efficient in SQLite (https://www.sqlite.org/np1queryprob.html) so this isn't really a huge upgrade, but the cross-language result is pretty interesting to me - this is language agnostic as all you do is listen to the WAL file and call SQLite functions.
On top of the store/notify primitives, honker ships ephemeral pub/sub (like pg_notify), durable work queues with retries and dead-letter (like pg-boss/Oban), and event streams with per-consumer offsets. All three are rows in your app's existing .db file and can commit atomically with your business write. This is cool because a rollback drops both.
This used to be called litenotify/joblite but I bought honker.dev as a joke for my gf and I realized that every mq/task/worker have silly names: Oban, pg-boss, Huey, RabbitMQ, Celery, Sidekiq, etc. Thus a silly goose got its name.
Honker waddles the same path as these giants and honks into the same void.
Hopefully it's either useful to you or is amusing. Standard alpha software warnings apply.
Struggling to see why you would otherwise need this in java/go/clojure/C# your sqlite has a single writer, so you can notify all threads that care about inserts/updates/changes as your application manages the single writer (with a language level concurrent queue) so you know when it's writing and what it has just written. So it always felt simpler/cleaner to get notification semantics that way.
Still fun to see people abuse WAL in creative ways. Cool to see a notify mechanism that works for languages that only have process based concurrency python/JS/TS/ruby. Nice work!
Cron jobs might need to coordinate with webservers. Even heavily threaded webservers might have some subprocesses/forking to manage connection pools and hot reloads and whatnot. Suid programs are process-separated from non-suid programs. Plenty of places are in the "permanent middle" of a migration from e.g. Java 7 to Java 11 and migrate by splitting traffic to multiple copies of the same app running on different versions of the runtime.
If you're heavily using SQLite for your DB already, you probably are reluctant to replace those situations with multiple servers coordinating around a central DB.
Nit:
> languages that only have process based concurrency python/JS/TS/ruby
Not true. There are tons and tons of threaded Python web frameworks/server harnesses, and there were even before GIL-removal efforts started. Just because gunicorn/multiprocessing are popular doesn't mean there aren't loads of huge deployments running threads (and not suffering for it much, because most web stacks are IO bound). Ruby's similar, though threads are less heavily-used than in Python. JS/TS as well: https://nodejs.org/api/worker_threads.html
But this is actually a great main benefit as well.
Unless you have a single "reader", you don't mind the delay, and don't worry about redoing a bunch of notifications after a crash (and so, can delay claims significantly), concurrency will kill this.
The experiment and back-of-the-envelope calculations show that it can only support ~ 5 jobs/sec. The only major factor to increase throughput is to increase the size of group commits.
I dont think shipping CDC instead of whole sqlite files will change the calculations as the number of writes mattered in this experiment.
So yes, the number of writes (min. of 3) can support very low throughputs.
Another maybe stupid question, would something like inotify(7) help to get rid of any active polling?
In other words, there’s a lot of unmeasured performance degradation that’s a side effect of doing many syscalls above and beyond the CPU time to enter/leave the kernel which itself has shrunk to be negligible. But there’s a reason high performance code is switching to io_uring to avoid that.
But I agree with the conclusion, system calls are still pretty fast compared to a lot of other things.
But I was clarifying because the wording could be taken as data/instruction cache and there generally isn’t a full flush of that just to enter/leave kernel.
https://sqlite.org/pragma.html#pragma_data_version
Or for a C API that's even better, `SQLITE_FCNTL_DATA_VERSION`:
https://sqlite.org/c3ref/c_fcntl_begin_atomic_write.html#sql...
> [SQLITE_FCNTL_DATA_VERSION] is the only mechanism to detect changes that happen either internally or externally and that are associated with a particular attached database.
Another user itt says the stat(2) approach takes less than 1 μs per call on their hardware.
I wonder how these approaches compare across compatibility & performance metrics.
Aside from this - SQLite has tons of cool features, like the session extension.
I may be wrong, but I think you wrote somewhere that you're looking at the WAL size increasing to know if something was committed. Well, the WAL can be truncated, what then? Or even, however unlikely, it could be truncated, then a transaction comes and appends just enough to it to make it the same size.
If SQLite has an API it guarantees can notify you of changes, that seems better, in the sense that you're passing responsibility along to the experts. It should also work with rollback mode, another advantage. And I don't think wakes you up if a large transaction rolls back (a transaction can hit the WAL and never commit).
That said, I'm not sure what's lighter on average. For a WAL mode database, I will say that something that has knowledge of the WAL index could potentially be cheaper? That file is mmapped. The syscalls involved are file locks, if any.
Can you use it also as a lightweight Kafka - persistent message stream? With semantics like, replay all messages (historical+real time) from some timestamp for some topics?
As with pub/sub, you can reproduce this with some polling etc but as you say, that's not optimal.
Would it help if subscriber states were also stored? (read position, queue name, filters, etc) Then instead of waking all subscription threads to do their own N=1 SELECT when stat(2) changes, the polling thread could do Events INNER JOIN Subscribers and only wake the subscribers that match.
Thanks all for your feedback, responses, and discussion. I've done a PR here taking your suggestions into account:
https://github.com/russellromney/honker/pulls/1
The PR implements a three-layer polling architecture: - PRAGMA data_version every 1ms - stat every 100ms - retry connection to handle blips
1. PRAGMA data_version every 1ms replaces stat-based (size, mtime) change detection. This is SQLite's own commit counter: monotonic, immune to clock skew, correctly handles WAL truncation and rolled-back transactions. ~3µs nonblocking query. Credit to ncruces for pointing to this. This is not done for performance but for correctness as it is slightly slower. tuo-lei also pointed out truncation risk, which turned out to be more real than i thought.
Interesting note: I found in testing that the C API's SQLITE_FCNTL_DATA_VERSION does not work cross-connection. So for now honker continues paying the cost of going through the VFS layer which vlovich123 pointed out and now we tradeoff explicitly.
2. Reconnect-on-error: if the data_version query fails (disk blip, NFS hiccup, corrupted connection), honker tries to reconnect and wakes subscribers as a precaution. zbentley pointed me in this direction.
3. stat identity check every 100ms: compares (dev, ino) against startup values to detect file replacement (atomic rename, litestream restore, volume remount). data_version can't catch this because it polls through the open fd, which follows the original inode even after replacement. Credit to zbentley for the file-replacement scenarios.
Again - thanks for the discussion, honker got better because of it and I learned some stuff. See you round
You want EVFILT_VNODE with NOTE_WRITE. That's hooked up to VNOP_WRITE in the kernel, the call made to the relevant filesystem to actually perform the write.
The specific thing I'm talking about is this: write events don't fire until the file handle is closed. [1] I didn't validate this myself btw, but my original design was certainly trying to use notify events rather than stat polling. My research (heavily AI assisted of course) led me away from that path as platforms differ in behavior and I wanted to avoid that.
one thing i'm curious about: WAL checkpoint. when SQLite truncates WAL back to zero, does the stat() polling handle that correctly? feels like there's a window where events could get lost.
notifs are extremely cheap, either in the old stat(2) mode or the new PRAGMA page_version (see my update on feeback comment). Some other comments mentioned that stat(2) is about 1µs.
I have a proliferation of small apps backed by SQLite. And most of these need a queue and scheduler.
I home rolled some stuff for it but was always pining for the elegance of the Postgres solutions.
Will give this a spin very soon
What happens on WAL checkpoint? When the file shrinks back, does that trigger a wakeup, or does the poller filter size drops?
It is possible to achieve with external IPC, but require a lot of very careful programming.
Question: any thoughts on what breaks first when a single process has 10k+ concurrent listeners? I'm curious whether the SQLite side can sustain what Postgres does cheaply.
After peeking the source, a few possible areas of improvement:
- You can use `fstat` and keep a file handle around, likely further improving performance (well, reducing the performance hit to other users of the filesystem by not resolving vfs nodes). If you do this, you'll have to check for file deletions.
- If you do stick with stat(2), it might be a good idea to track the inode number from the stat result in addition to the time,size tuple. That handles the "t,s = 1,2; honker gets SIGSTOPped/CRIU'd; database file replaced; honker started again", as well as renameat/symlink-swap fiddling. Changing inode probably should just trigger a crash.
- Also check the device number from the stat call. It sounds fringe, but the number of weird hellbugs I've dealt with in my career caused by code continually interacting with a file at the same time as something else mounted an equivalent path "over" the directory the file was originally in is nonzero.
- It's been a few years since I fought with this, but aren't there edge cases here if the system clock goes backwards? IIRC the inode timestamp isn't monotonic--right? There are various strategies for detecting clock adjustment, of various reliability, that you could use here, if so. Just checking if the mtime-vs-system-clock diff is negative is a start.
That covers the more common of the "vanishingly uncommon but I've still seen 'em" cases related to file modification detection. Whether you choose to cope with people messing with the file via utime(2) is up to you (past a point, it feels like coping with malicious misuse rather than edge cases). But since your code runs in a loop, you're well-positioned to do that (and detect drift/manipulations of the system clock): track a monotonic clock and use it to approximate the elapsed wall time between honker poller ticks (say it fast with an accent, and you get https://www.bbc.com/news/world-latin-america-11465127); if the timestamp reported by (f)stat(2) ever doesn't advance at the same rate, fall back to checksumming the file, or crashing or something. But this is well into the realm of abject paranoia by now.
It's been a decade or so since I worked in this area, so some of that knowledge is likely stale; you probably know a lot more than I do after developing this library even before considering how out-of-date my knowledge might be. When I worked on this stuff, I remember that statx(2) was going to solve all the problems any day now, and then didn't. More relevant, I also remember that the lsyncd (https://github.com/lsyncd/lsyncd) and watchman (https://github.com/facebook/watchman) codebases were really good sources of "what didn't I think of" information in this area.
But seriously, again, nice work! Those are nitpicks; this is awesome as-is!
I actually looked at fstat, but the "check for deletions" piece, given I'm polling at 1kHZ, was the reason I decided not to use it. Older hardware actually made this a big issue but it's fast enough now I decided it wasn't a problem.
I'll ignore the malicious ones bc [out of scope declaration]. Object paranoia is an artifact of build trama and I respect that lmao.
I've just looked into the device number and system clock issues. I think what i'll end up doing is actually a combo of ncruces's above comment and your feedback: a 1kHZ data_version and a 10HZ stat() with version check. This gets around syscall load, avoid clock issues, avoids the WAL truncation issues that others have mentioned, and is both lighter weight and less bugabooable than my previous design.
Thanks again.
Any conflicts or issues when running Litestream as well?