"Idle cost is that one lightweight SELECT per millisecond per database — no page-cache pressure, no writer-lock contention, no kernel file watcher in the mix."
I think (respectfully) the LLM that probably wrote this overshot the mark here because busy-polling a select does not actually sound better to me than a "kernel file watcher".
> Once real work flows through a SQLite-backed app, you need a queue. The usual answer is “add Redis + Celery.”
Are they joking? SQLite is usually used for single-process (mutliple threads) applications. The proper way to communicate between threads/processes is a ring buffer, where you allocate structs (allocation typically is incrementing a pointer), and futex/eventfd for notifications (+ some spinlocking to avoid going to kernel when the tasks arrive quickly). Why do you need redis for that? If you need persistent tasks, then you can store them in the table, and still use futex for notifications. This polling is inefficient and they should not make it a library which will cause other lazy developers add it to their app.
> honker polls SQLite’s PRAGMA data_version every millisecond. That’s a monotonic counter SQLite increments on every commit from any connection, journal mode, or process — a ~3 µs read for a precise wake signal
That's 3 ms per second = 0.3% CPU time wasted for every waiting thread.
Like Electron, this feels like written by a web developer and not a real programmer.
>That's 3 ms per second = 0.3% CPU time wasted for every waiting thread.
I suspect that's actually "per process, per database (usually 1)", and not based on number of threads or tables. `data_version` semantics mean there's no need for more than one connection polling it, and it's being used as a relatively lightweight "DB has changed, check queues" check.
Also I believe this is mostly intended for multi-process use, e.g. out-of-process workers, so an in-process dirty tracker (e.g. just check after insert/update/delete) isn't sufficient.
So I do think it's somewhat crazy, but it is at least very simple. fsnotify-like monitoring seems like a fairly obvious improvement tho, not sure why that isn't part of it.
Nevertheless, expect articles like "We replaced our redis cluster with this simple extension and got it N times faster".
It’s an interesting approach and can be quite fun to use for new projects.
> How it works: honker polls SQLite’s PRAGMA data_version every millisecond. That’s a monotonic counter SQLite increments on every commit from any connection, journal mode, or process — a ~3 µs read for a precise wake signal.
Prior discussion a few days ago: https://news.ycombinator.com/item?id=47874647
Reminds me of Litestack for Rails. Eventually, it was abandoned because Rails itself started going all out on SQLite.
All in*
I've implemented something similar in the past, but using inotify. You need to watch the -wal file for IN_MODIFY. To make it work reliably I found I had to run:
BEGIN IMMEDIATE TRANSACTION; ROLLBACK;
At the end it says: "pg-boss and Oban are the Postgres-side gold standards" -- but Oban supports SQLite now too https://github.com/oban-bg/oban
There's also Graphile Worker. https://github.com/graphile/worker
Almost feels like someone is trying to joke about similar postgres application .
To make it look even more absurd . SQLite is not concurrent and you’ll have tons of problems using it practically .
This seems especially appealing in the awkward middle: too serious for in-memory queues, not big enough to justify Kafka-shaped machinery.
Suggestion for the author wind back the polling to once a second when nothing is happening.
I can’t see any benchmarks or performance stats.
I’d like to see messages per second.
Could this work with Turso, the SQLite rust rewrite?
"one lightweight SELECT per millisecond"
This reminds me of the teenager who told her dad that she was just a tiny little bit pregnant.
Thing of the battery!
(read that in the way of "think of the children!")
[flagged]
Hold on -- if it really is "one lightweight SELECT per millisecond", and you're saying a select is "a couple hundred microseconds", say generously 200us?, then you're spending 200us out of every 1000us just selecting. That's a lot of polling!
I mean only in the same sense that you spend 1 second per second doing something. Time is probably not the best way to evaluate the resources this consumes and I doubt it takes much of anything else either.
It does seem weird though even for sqlite. I wonder how oban does it. I also wonder if OP knows oban can run on sqlite.
Yeah, again, to be clear: I get how SQLite works and I'm not dunking on the design, I'm just saying the comparison set up on this page snags. It's a classic LLM negated triptych, but "one of these things is not like the other": cache pressure: bad, writer contention: bad, kernel file watcher: ... good, actually? Intuitively seems better than this design?
to me it sounds like they asked it to not make a kernel file watcher, and now it writes that into every comment everywhere, despite not even being in the implementation
If you're not making any changes to the database, does the SELECT "kill" you?
And if you are making changes, don't you have to poll regardless after the file watcher wakes you?
For WAL mode, SQLite can probably satisfy this query just by inspecting some shared memory. But it is busy waiting, sure.
SQLite has a wal hook which calls you back every time a transaction is committed to the WAL. https://www.sqlite.org/c3ref/wal_hook.html
That only catches changes made by the database connection being "hooked."
This has a thread running in the background trying to catch changes made by other connections, potentially (I'm not sure here, but I suspect as much) in different processes that are modifying the same database.
good point. but ime and as seems to be widely understood writing from multiple connections is a bit of a minefield in SQLite. and afaik it still would be possible to have a hook on all connections you expect to be writing?
[dead]
Yeah, I had the same instinct - this feels very much like a "nice idea" but the execution falls short. I mean - busily banging on sqlite like this? Shit at that point just use Redis.
For what it's worth, Kine (software that k3s uses to replace etcd with SQL databases) implements etcd watches on SQLite through polling[1]. The reason being that SQLite does not offer NOTIFY/LISTEN like MySQL and Postgres do. Ironically, Honkey attempts implementing NOTIFY/LISTEN through polling.
k3s has been running on my home server for about three years now (using the default SQLite backend), and there doesn't seem to be excessive CPU usage despite dozens of watches existing in the simulated etcd. Of course, this doesn't say much about Honker, but it's nonetheless worth pointing out that sometimes the choice of database forces one towards a certain design.
[1] https://github.com/k3s-io/kine/blob/648a2daa/pkg/logstructur...
Are you trying to avoid sleep?
With SQLite, you're basically funneled towards a single-writer / single-process design anyway ... in which case why not use a more traditional condvar + mutex rather than polling?
I'm not even saying it's unworkable, just, my intuition is not that the "lightweight per-millisecond select" is an optimal design.
Really might be in sqlite. I've learned to never trust my intuition about performance with that thing. So many times I've gone to "optimize" something and discovered that the naive hack way I had been doing it was faster anyway. It's built for this sort of bullshit.
Oh, yes, I see what you mean now.
What's the CPU usage? Like 2%?
I had a manual fs polling thing a while back. It was ugly (low time budget, didn't wanna mess with the native watchers), just scanned the whole thing once per second. It averaged out to like 0.3% CPU.
Not elegant, but acceptable for my purposes! (Small-ish directory, and "ping me within a second or two" was realtime enough for this use case.)
i mean, technically this is once per millisecond, so this would happen 1000x more. In your case due to the kernel overhead you would likely not even be able to do it (300% CPU?).
Either way this does seem like a very large overhead due to the fact that there's just no other way to do it without a deeper kernel integration which might be outside the scope of what sqlite is trying to do.
> one lightweight SELECT per millisecond
For the low, low cost of $1 per minute, you can also lease a supercar.
[flagged]