Rana Dolui

Full stack AI developer

[Rana Dolui @jack718r]

@anuragdotdev College one’s portfolio actually looks very nice 💀

[Rana Dolui @jack718r]

@mirajobs Currently single-node with RDB snapshots and atomic fsync for durability not async replication. It auto-saves every 5 minutes by default and performs a final save on SIGTERM/SIGINT. Multi-node/replication support is currently in development :)

[Rana Dolui @jack718r]

FlashDB is now production-grade with major stability, performance, and memory improvements.

22.04M GET/s
20.91M SET/s
27.02M Pub/Sub/s
Peak memory: 600MB → 247MB (59% drop)
Single-node FlashDB vs 6-node Redis Cluster.
https://t.co/Olz7VhcCdF

#Rust #Redis #Database #OpenSource

[Rana Dolui @jack718r]

@Praha37v https://t.co/wvRdoiEJbO

[Rana Dolui @jack718r]

🚀 CPM 0.2.0 is out!

Fast, isolated & easy-to-use C/C++ package manager for Linux 🐧
📦 GitHub/Nix deps
🔧 Isolated toolchains
⚡ Reproducible builds
🛠️ Auto compile_commands.json
🌐 https://t.co/FnW4lz4x2M
⭐ https://t.co/ul8yI62uEB
#C #Cpp #Linux

[Rana Dolui @jack718r]

@KartikLande15 @arpit_bhayani nice :)

[Rana Dolui @jack718r]

After completing @arpit_bhayani’s Redis video series, I built FlashDB — a lock-free, Redis-compatible database in Rust.

21.3M GET/s
15.7M SET/s
22.5M Pub/Sub/s

Single node Flash vs 6-node Redis Cluster.
🔗 https://t.co/OcxYeFu9LQ

#Rust #Redis #Database #OpenSource

[Rana Dolui @jack718r]

@ditsuke @arpit_bhayani Both FlashDB and Redis were running inside Docker containers on the same Linux machine, so the comparison uses the same containerized environment

[Rana Dolui @jack718r]

@ditsuke @kakashii24 No, not yet. Fingerprinting could reduce pointer chasing, but FlashDB’s 64-bit hash already filters most false matches. It’d also require major changes to the current atomic-pointer design. Worth benchmarking later :)

[Rana Dolui @jack718r]

@sarmag77 @arpit_bhayani Thanks 😊

[Rana Dolui @jack718r]

@CinePill :))

[Rana Dolui @jack718r]

@KakaluoteW45042 Persistence favors Redis here: RDB/AOF are disabled, while FlashDB keeps periodic RDB snapshots (none during the test). Redis also uses 2,500-command pipelines vs 100 for FlashDB. So this is a throughput comparison, not durability.

Benchmark code: https://t.co/UVlxvpi5J5

[Rana Dolui @jack718r]

@helyoussfii @arpit_bhayani https://t.co/jxQkjpPkZb

[Rana Dolui @jack718r]

@kakashii24 It’s primarily based on the load factor. I chose 70% as the resize threshold for the open addressing table to keep probe lengths low. I also benchmarked around that range and found it to be a good trade-off between memory usage and lookup performance

[Rana Dolui @jack718r]

@nhuxhr @arpit_bhayani :)

[Rana Dolui @jack718r]

@WarBornePhoenix @arpit_bhayani thanks

[Rana Dolui @jack718r]

@shankstwin @arpit_bhayani Yeah bro, definitely complete it! It’ll really help you understand how Redis works under the hood

[Rana Dolui @jack718r]

@sarmag77 @arpit_bhayani Mainly because FlashDB uses a multi-threaded, lock-free HashMap help of atomicptr, so readers and writers don't block each other on the hot path. I wrote a deep dive about the custom HashMap if you'd like to check it out - https://t.co/ReZMo9nUgy

[Rana Dolui @jack718r]

@SambitChakrabo8 @arpit_bhayani Sure :)

[Rana Dolui @jack718r]

@sabari_246 Thanks! You can give it a try, or check out this blog post about the custom concurrent hash map I built and use in FlashDB:
https://t.co/ReZMo9nUgy

[Rana Dolui @jack718r]

@devsup_ thanks