Neo4j has an interesting approach to handling complexity beyond the core query model - a plugin system which adds procedures. I had a query that was running out of memory. Turns out they have a batching system in one of their plugins. https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ragtjsm2j2vknwkz3zp4oxrd&cid=bafkreiam2mzrkt72utyuniysvwutyimkqlprv6kq5kfsacxofr64v2xf3e

Mark Hamill and George Takei RE:

Bluesky Social

Blake Stovall (@mstrofbass.bsky.social)

Which two of the ten accounts with the largest follower accounts share the most followers.

I've loaded up neo4j with a pretty big amount of the follow graph. Anybody got a query they want me to run?

I've developed an algorithm to find out who are my real ones and who are just faking it. Which my therapist says is good behavior https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ragtjsm2j2vknwkz3zp4oxrd&cid=bafkreihbtrtct2k3by2tikhbjyylqqtom5r63sojm3vvssqzhf4wrkoxfy

Does it count as inventing it if you Google and find out it already exists

watching htop while I run neo4j commands with the same energy that other people watch sports

We're in business. 4.9M nodes, 348M follows, imported into neo4j in 11.5min Listing my follows takes 775ms. Listing all my followers takes 1236ms.

neo4j import tool has a --bad-tolerance flag, which more software should have

I feel differently about it now but back in ‘04 I did download a car

[code talk] #atproto I now have a snapshot dataset of ~4.8M car files - users that hit the new relay since it was started. I rigged up a node cluster (12 workers) that runs through the car files and dumps the follow graph into 12 different CSVs. Throughput bounces btwn 200-400ps https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ragtjsm2j2vknwkz3zp4oxrd&cid=bafkreib3qnlh4bydtfrxmiu4f76ygmqsuzelu2m7vihnwyvfawdg4cs6cy