SQLite pushed hard

[gedw99]

https://www.terlici.com/2023/11/06/pushing-sqlite-limits.html Shows sone good benchmarks

sqlite and golang

24K requests per second, 72Mbit/s traffic, 271MB RAM used. That’s with writes concurrently

am think it might be good to change the benchmarks to writes / second styje results because that is how users would want to do the metrics on their servers. Then we can start to see comparative results .

[ganigeorgiev]

Thanks for sharing. Benchmarking PocketBase applications is a little more complicated vs raw SQLite as there multiple layers before between the request and the database (routing middlewares, collection API rules, validations, json response filtering, etc.).

For the most recent PocketBase benchmarks you can explore the Results files.

[gedw99]

Yeah I looked already art the result files.

I was hoping to get an idea of currency aspects, which those results don't show.

Main thing is the response time based on number of concurrent users. for example

Creating posts (10k, 25k, 50k, 100k)

Creating 5000 posts10k [reqs:5000, conc:500, rule:""]

┌─ Best:      7.481876ms
├─ Worst:     6.382093667s
├─ Completed: 6.540550371s
└─ Errors:    0

I assume conc: 500 means that its 500 concurrent users ?

[ganigeorgiev]

I assume conc: 500 means that its 500 concurrent users ?

Yes.

But again the linked article cannot be compared directly to PocketBase because we do a lot more work. You are not taking in consideration:

• route handling and middlewares (even a simple middleware function call can have drastic impact on micro benchamarks)
• activity logger (aka. db request log writes)
• API security and filter rules
• data validations
• data serialization
• hook calls
• the benchmark test runs as part of the same application process that started the PocketBase server
• pure Go (PocketBase default) vs CGO SQLite drivers (which the article uses; the linked article also uses prepared statements)
• difference in the running hardware
• connection limits to avoid starving the file descriptors and memory - we have default limits for the max open and idle connections
• PocketBase maintain separate read and write pools + have auto fail-retry handling to take care for the SQLITE_BUSY errors
• and probably others...