optimize #crisp kvrocks rocksdb configuration file on server I migrated yesterday because it's burning through the SSD pretty fast, something must be sub-optimal here

migrate all legacy redis commands i used back in 2019 to talk to SSDB for #crisp so that we can now talk to kvrocks

create 2x new DigitalOcean servers to setup another kvrocks cluster to replace old legacy SSDB servers for #crisp

Add workspace-based token for #crisp API

monitor #crisp kvrocks cluster setup after 48h run and system metrics are looking insanely good this thing is so lightweight on resources

fine tune #crisp rocksdb-powered kvrocks server configuration file so that we get more cache hits in the in-memory block cache vs hitting the disk, which can save us when the hypervisor throttles disk I/O for some reason (happens)

monitor kvrocks setup and check if it hammers the SSD with write I/O and no it’s good now #crisp

finish ingesting all of #crisp 300M+ keys from old kv store to kvrocks, and kvrocks is really chill on server resources especially SSD write I/O, migration was totally worth it. won't have to emergency migrate server every year or so because I burnt the SSD and it's reporting bad sectors 🤣

monitor #crisp 300M keys redis ingestion to kvrocks and it's progressing slowly but surely

deploy kvrocks in production at #crisp for session storage system replacement that's more modern & safe at scale (300M total keys to migrate)

Push a bunch of improvements to #crisp

add all my public oss to #crisp homebrew cask to ease with local #crisp infrastructure setup

replicate #crisp 200GB redis database at 2Gbps speed using kvrocks on test server — how can this be that fast compared to the previous system we use

add disclaimer in #crisp documentation that we do not accept security vulnerability reports from LLMs and other automated systems, because I'm getting spammed by LLMs over email and it's humanly impossible to filter what's legit (maybe 0.1% of reports) over what's automated — maybe LLMs will comply when seeing that but I bet they'll just ignore, so if that is not effective then I'll nuke our reponsible vulnerability report bounty program which is sad

renew important domain names like #crisp for the maximum possible time

start testing migration of cold storage redis data for #crisp to kvrocks

improve openvpn bastion vpn security to access #crisp infrastructure and other critical servers, adding a CRL to be able to revoke certificates on-demand if leak

type all mongoose ORM objects in #crisp API + storage backend

start background process of mirroring 150M files to a new S3 bucket, ETA 10 days #crisp

run some optimizations on #crisp image resize service, following the migration of all stored files to a new S3 backend, so that if cloudflare / cloudflare tunnel is down in AMS DC we can still serve images wordwide