mirror of
https://github.com/featurebasedb/featurebase.git
synced 2026-08-28 10:54:59 +00:00
4467c18baa
The special case of Starting allowed us to make sure every node in a
cluster waited for the whole cluster to come up, but caused problems
later if a node died and came back. We drop the Starting state for
clusters, treating a STARTING node as equivalent to an UNKNOWN (or
DOWN) node for purposes of cluster state, so clusters will go from
Down to Degraded to Normal as nodes come up. We now wait for the
Normal state during initial bringup. We would previously have accepted
Degraded, if you could reach it, for instance if a node came up and
then went down again before another node finished starting, but I'm
pretty sure that was unintentional.
This solves a problem where while a node was down, we'd accept
queries that we could handle in a degraded state, but then we'd
*stop* accepting them when the node started coming back up.
(cherry picked from commit
|
||
|---|---|---|
| .. | ||
| disco.go | ||
| hasher.go | ||
| node.go | ||
| noder.go | ||
| snapshot.go | ||