This post has been de-listed
It is no longer included in search results and normal feeds (front page, hot posts, subreddit posts, etc). It remains visible only via the author's post history.
In a distributed database made of thousands of partitions(e.g DynamoDB, Cassandra etc), how do they do consistent backups across all partitions?
Imagine a batch write request went to 5 partitions and the system returned success to caller.
Now even though these items or even partitions were unrelated, a backup should include all writes across partitions or none.
How do distributed databases achieve it?
I think doing a costly 2 Phase-Commit is not possible.
Do they rely on some form of logical clocks and lightweight co-ordination(like agreeing on logical clock)?
Subreddit
Post Details
- Posted
- 11 months ago
- Reddit URL
- View post on reddit.com
- External URL
- reddit.com/r/databasedev...