BTW, this is also the model used by all CDNs, where the global configuration file containing the certificates, HTTP routing rules etc. for all customers will be updated into into a single-file b-tree structure*, and that "bundle" is distributed among all edge locations frequently.
* I'm yet to see someone use sqlite for this purpose, it's usually DBM style databases like LMDB or Kyoto Cabinet.
Periodically we would run a full sync to replace the database. Between the periodic full syncs, we had a background process keep changes applied on a rolling basis.
All-in-all, it worked pretty well at the time! The full database file sync ensured a bad database was timeboxed and we got a bootstrapping mechanism for free.
I'm not knowledgeable on this, but my understanding was a b-tree is a way of sorting values that could be ordered in a certain way. Like this would be a b-tree of IDs
```
[8]
/ \
[3, 5] [10, 12]
/ | \ / | \
[1] [4] [6,7] [9] [11, 13]
```You traverse by comparing your needle to the root node and going left or right depending on the results.
How is that done with configuration options? That seems like it would just be a regular hashmap which is already efficient to read. What would a b-tree of key/values even look like that wouldn't be less efficient than a hashmap?
Databases including SQLite usually use b+tree for tables (a variant where only the leaves have data, the interior nodes only have keys) and regular btrees for indexes.
So it might seem that simply enumerating the data (sorted or not) would be a better option for a file. (After all, the receiver will read everything anyway.) I guess frequent updates make this inefficient, so a tree helps.
Now, that's a name I've not heard in a long time.
Are people still using Kyoto Cabinet in new projects? Are people still using DBM-style storage generally? I thought that whole branch of the evolutionary tree had sort of died out.
Cloudflare used to use Kyoto Cabinet[1] and moved to LMDB[1] in 2020; other implementations that I'm familiar with (but don't have a link to share) also use LMDB.
> Are people still using DBM-style storage generally?
It's fairly common in these scenarios, as well as the underlying key-value store for popular software like Consul[3].
[1] https://blog.cloudflare.com/kyoto-tycoon-secure-replication/
[2] https://blog.cloudflare.com/introducing-quicksilver-configur...
https://blog.cloudflare.com/moving-quicksilver-into-producti...
In any case, SQLite would serve this solution, but almost certainly with a measurable level of inefficiency built in.
Edit: @akira2501: SQLite comes with different tradeoffs, sometimes superior and other times less so, depending on the use case. Blanket statements without evidence are unproductive in progressing the conversation..
https://www.sqlite.org/optoverview.html
Chapter 2. Chapter 7. Chapter 8.
It should be _zero_ surprise to you that SQLite is not as powerful as other full database query engines.
But then it was quite fast.
But as with all things SQLite, the answer is generally "it's easy to predict so just improve your query" and/or "there's a plugin for that" so it tends to win in the end with a bit of care.
SQLite's planner is anything but simple. It has a beam search over different plans, basically, it approximates full NP-hard solution search process using bounded space and time.
My strategy for resolving this is to fetch the flag value once, but to store it in the request object, so that a) you never have to take the expensive lookup hit more than once per request, and b) there's no risk of an inconsistent value if the flag is updated mid-request.
I can see why you need to check multiple different flags in a session and I understand the parent point about looking in SQLite for them (effectively a function call into a library in process address space rather than a call over the network for each flag).
One example is a multistep transaction processing request. The feature flag could gate several branch points.
A memory-mapped SQLite file is great too, but the strategy I describe above is less code, quicker to implement, and should get you a very similar performance improvement.
In Ruby we called this “hammerspace” https://github.com/airbnb/hammerspace
2. If you were making it today would you just use SQLite?
The write tracking needs to intercept all writes with a custom VFS, but once registered and configured, it's regular in-memory SQLite (no additional daemon).
I'm honestly surprised it isn't more pervasive.
You could look into WAL replication if you wanted an efficient way to update the copies. Something like Litestream.
This doesn't sound right. A feature flag only requires checking if a request comes from a user that is in a specific feature group. This is a single key:value check.
The business logic lies in assigning a user to a specific feature group, which the simplest way means pre assigning the user and in the most complex cases takes place at app start/first request to dynamically control dialups.
Either way, it's a single key: value check where the key is user ID+feature ID, or session ID + feature ID.
I mean, I guess you can send a boat load of data to perform the same complex query over and over again. I suppose. But you need to not have invested any thought onto the issue and insisted in making things very hard for you, QAs, and users too. I mean, read your own description: why are you making the exact same complex query over and over and over again, multiple times in the same request? At most, do it once, cache the result, and from therein just do a key:value check. You can use sqlite for that if you'd like.
The number of feature flags is irrelevant. In fact, the feature flag and A/B testing services I used always returned all default treatment overrides in a single request.
This meant you could roll features out to:
- Specific user IDs
- Every user ID in a specific group
- Every object owned by a specific user ID (feature flags might apply to nested objects in the system)
- Requests from IP addresses in certain countries
- Requests served by specific website TLDs
- Users who are paid members of a specific plan
- etc etc etc
It was an enormously complicated system, that had evolved over 5-10 years.
Not saying that level of complexity is recommended, but that's what we had.
Looks like I gave a talk about this back in 2014: https://speakerdeck.com/simon/feature-flags
That's how feature flag services work.
The whole point is that this is not a querying problem. That's a group assignment problem.
Users/sessions are assigned to a feature group according to business rules. These can be as complex as anyone wants them to be.
Once a user/session is assigned to a feature group, the problem of getting features is a key-value query.
That said, it may not be practical to have hundreds of read-only nodes, but for moderate-size needs, should work fine.
Disclaimer: I'm the creator of rqlite.
That said, I do think it depends on what you consider important, and what your experience has been in the past. I used to value simplicity above all, so reducing the number of moving pieces was important to my designs. For the purpose of this discussion let's count a service as a single moving piece.
But over time I've decided that I also value reliability. Operators don't necessarily want simplicity. What they want is reliability and ease-of-use. Simplicity sometimes helps you get there, but not always.
So, yes, rqlite means another service. But I put a lot of emphasis on reliability when it comes to rqlite, and ease-of-operation. Because often when folks want something "simple" what they really want is "something that just works, works really well, and which I don't have to think about". SQLite certainly meets that requirement, that is true.
As requests come in, gather their features, convert to effectively an int vector, filter for row where match and serve request
* SQLite, but every change you make increments a global version counter.
* Some way to retrieve and/or apply just the changes between version counter numbers.
Then, you could layer all manner of distribution mechanisms on top of that. You could imagine gossip networks, an explicit tree-shaped cache/distribution network, etc. Anyone who has a later version than you would be able to give you the update.
What would this get you?
* You can bundle the current version of a DB into your app artifact, but efficiently update it once your app runs. This would let you fall back to much better defaults than having no data at all. (eg. For feature flags or live config, this would let you fall back to a recent state rather than code-level defaults.)
* Any kind of client can send the global counter as an ETAG and get just the delta
* Reconnections if the network blips are really simple and cheap
* If the system also let you keep a couple of minutes of history, you could annotate the counter onto a microservice call (using W3C Baggage headers, for example), and evaluate your flags/config/data/whatever at a single version number across multiple microservices. Even without evaluate-at-time functionality, logging the generation number would help a lot when debugging what happened after the fact
While SQLite can be used on SSDs, it is not as heavily optimized for SSDs as RocksDB. SQLite is a general-purpose embedded database and was primarily designed for use on devices like mobile phones or lightweight embedded systems, where storage is often slower and smaller (e.g., flash storage or basic SSDs).
SQLite’s traditional B-tree structure involves more random I/O compared to RocksDB's LSM tree, which is less ideal for maximizing the performance of high-speed SSDs like NVMe.
SQLite’s limited concurrency (single-writer, multiple-readers) also means it cannot fully utilize the parallelism and high throughput potential of NVMe SSDs.
Here it’s being used for web application firewall rules.
Another place I’ve thought about using this is feature flag configuration. Feature flags can be checked dozens of times per request and often need the kind of queries (user is a member of group A and has an IP located in country B) which could be well served by a local SQLite - and feature flags have a tolerance for updates taking a few seconds (or longer) to roll out.