4 ms·
It's really interesting to see that you're using Cassandra for this. IIRC, Cassandra was created by Facebook for their messaging, and realized that eventual con
by maktouch 10y ago
It's really interesting to see that you're using Cassandra for this. IIRC, Cassandra was created by Facebook for their messaging, and realized that eventual consistency was a bad model for chat, so they moved to HBase instead. (source: http://highscalability.com/blog/2010/11/16/facebooks-new-real-time-messaging-system-hbase-to-store-135.html http://highscalability.com/blog/2010/11/16/facebooks-new-rea...)
The tombstone issue was really interesting ! Thanks for sharing.
- Vishnevskiy 10y agoThey did make the original and the core model is the same, but Cassandra in 2017 is quite different from what they open sourced in features, usability and stability. I guess everyone makes their own tradeoffs though. This has been working wonderfully for us.
- jjirsa 10y agoYou can have strong consistency in Cassandra - ING gave a talk at Cassandra summit 2015 on their multi-DC Strongly consistent use cases Cassandra let's you choose - per query - how many replicas must ack the query. Strong consistency is just a query parameter away.
- nvarsj 10y agoNeed to be careful about the wording, "strong consistency". I dislike that datastax uses that in their documentation, because it's misleading and really confuses people. There is no commit protocol in place - Cassandra is still an AP db under the hood, so even having multiple replicas acknowledge doesn't mean the data is consistent. For that you need paxos or something similar. This becomes very obvious if you are doing updates from multiple sources to the same key.
- jjirsa 10y agoYou do realize that Cassandra implements paxos and has a CAS system right?