v0.3.1

• Disable parallel head processing by default and add an option to control it

v0.3.0

• Add DAG-repair support

Failures fetching or merging nodes can mark the datastore as dirty.

A dirty datastore is repaired by a process that rewalks the nodes from the
heads which gets triggered on a configurable interval.

This means we keep track of processed nodes in a new datastore
namespace. Nodes that upgrade will need to reprocess the full DAG
on boot.

This also means that go-ds-crdt no longer needs a DAGSyncer with a HasBlock()
method and can use directly an ipld.DAGService().

• Improved DAG processing logic: logic to avoid reprocessing of branches
  during sync has been improved.
• Avoid broadcasting 0 heads when there's nothing to broadcast.
• Update dependencies on the globaldb example
• Enable parallel processing of new heads

Improved branch-pruning logic allows better processing of tipsets in parallel.

v0.2.4

• Improve branch pruning when syncing
• Update dependencies
• Log some infos every 5 minutes

v0.2.3

• Performance improvements to the List() operation: We have introduced a bloom
  filter over the tombstone set so that keys that have never been deleted do not
  need to be checked against the tombstones. Real-world testing gives around 4x
  performance improvement.

• Be less noisy: We have changed some Info messages to Debug, as they re-occur
  too frequently on datastores with heavy usage.

v0.2.1

• Add context to the DAGSyncer interface HasBlock() method.

v0.2.0

• Add context to methods to comply with the latest version of the Datastore interface
• Dependency upgrades

v0.1.22

• Chore: dependency upgrades

v0.1.21

• Chore: dependency upgrades

v0.1.20

• Dependency upgrades
• Use Github actions for testing

v0.1.17

Changes:

• Update dependencies.