Merge branch 'main' into rocksdb_pageserver

pageserver/src/lib.rs

@@ -25,7 +25,30 @@ pub struct PageServerConf {
     pub gc_period: Duration,
 }
 
-// Zenith Timeline ID is a 32-byte random ID.
+/// Zenith Timeline ID is a 128-bit random ID.
+///
+/// Zenith timeline IDs are different from PostgreSQL timeline
+/// IDs. They serve a similar purpose though: they differentiate
+/// between different "histories" of the same cluster.  However,
+/// PostgreSQL timeline IDs are a bit cumbersome, because they are only
+/// 32-bits wide, and they must be in ascending order in any given
+/// timeline history.  Those limitations mean that we cannot generate a
+/// new PostgreSQL timeline ID by just generating a random number. And
+/// that in turn is problematic for the "pull/push" workflow, where you
+/// have a local copy of a zenith repository, and you periodically sync
+/// the local changes with a remote server. When you work "detached"
+/// from the remote server, you cannot create a PostgreSQL timeline ID
+/// that's guaranteed to be different from all existing timelines in
+/// the remote server. For example, if two people are having a clone of
+/// the repository on their laptops, and they both create a new branch
+/// with different name. What timeline ID would they assign to their
+/// branches? If they pick the same one, and later try to push the
+/// branches to the same remote server, they will get mixed up.
+///
+/// To avoid those issues, Zenith has its own concept of timelines that
+/// is separate from PostgreSQL timelines, and doesn't have those
+/// limitations. A zenith timeline is identified by a 128-bit ID, which
+/// is usually printed out as a hex string.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub struct ZTimelineId([u8; 16]);