Open
Description
Description
I think there's an issue where if the beacon node is killed or shutdown ungracefully before it has first written the head* and fork choice to disk, then it will be unable to resume.
I suspect dealing with this 100% reliably would be quite difficult, unless we are able to wrap the creation of the database, and all of its initial entries in one big atomic transaction. Incremental improvements may be better, but would also require more thought.
This issue is similar to #1680 and #1686
*NB: The head is actually just the head tracker and genesis block these days (see #1784)