2/11/2026 at 11:55:37 AM
CRDT hacker here. The talk says CRDTs need to store a history of fine-grained edits indefinitely to make them work. There are some ways around this requirement. I'd be happy to chat more about this if anyone is interested!by josephg
2/11/2026 at 2:03:24 PM
I’m new to CRDTs, but could we clean up tombstones if all replicas acknowledge? Not too sure, but could we persist snapshots of state at certain “compaction” points?by sanufar
2/11/2026 at 5:05:15 PM
You can do this by (1) extending your CRDT into a CTM (see https://braid.org/time-machines) and then (2) use the antimatter algorithm (braid.org/antimatter), or something similar, for acknowledgements.The antimatter algorithm allows peers to learn where the rest of the network has caught up to, without a central actor, or relying on consensus, across arbitrary P2P connection & disconnection events in the network. It even allows subnetworks, after a partition, to prune the history that they generate while partitioned, while still holding onto the necessary older history to reconnect with the other half of the partition.
by toomim
2/11/2026 at 7:31:45 PM
Yep. As well as what Toomim said, eg-walker (another text crdt algorithm) doesn’t need or use tombstones. With eg-walker we just store the history of all changes (kinda like git). Some changes are deletes. The changes are only used for merging old stuff, so if you’re happy to set a cutoff (or have some coordination like antimatter) you can delete the operation history and all the overhead that goes along with it.by josephg