features:

1. genconf generates the default config file for users.
2. migrate command to initialize persistence component (for now, DB support).
3. basic resource managing commands (CURD) + watch. Maybe, this would build on the client SDK ...
4. health diagnosis (TODO). metrics and cluster infos.

For now, cache is held by the core.Core, but the cache component is mainly used in the raft.FSM. To close up all of the cache component entries, so move it into FSMWrapper maybe make it cleraly to read.

Background

cassemd must get leader address so that it could join to cluster successfully, this is not easy to use and is unreasonable.

Solution

• forward join request to the leader while a slave node receives a join request
• allow cassemd to receive an array of node address so that it could try to join one by one.

• Redesign the concepts of config container.
• Focus on the edge problems in the situation of config center.
• The reuse of pair maybe not the main issue on which the config center should take care.

• feat: adjust Repository constructor API
• feat: independent convertor from repository
• feat: reduce authorizer from server but in coordinator
• refactor: pull away raft from core which makes core as daemon rather than a component.
• refactor: merge user repo, repo and policy adapter.

Background

Need a distributed persistence component to support what helps cassem become a system satisfied CP and provide available as much as possible. On another way, DO NOT want to import another dependency to make the save procedure complicated.

Risk

What is a risk is that would changes the cassem software architecture, even makes HTTP API changes.

Thoughts

bbolt would be used to replace mysql.

now cache components are implemented by go built-in map, it's too simple to manage caches. So that cassemd needs a new version of the cache component to manage caches.

The duty of the cache component includes:

// IStore required ICache to implement persist and restore functions so that the cache could be
// save into file and restore from that.
type IStore interface {
	// Persist serializes cache into []byte data.
	Persist() ([]byte, error)

	// Restore apply data to override all the cache data, data is coming from Persist.
	Restore(data []byte) error
}

// ICache represents the proxy to operate cache data,
// also need to replace data while cache size is over it's limits.
type ICache interface {
	IStore

	// Set returns wasSet means this key needs to synchronous, err means set failed.
	Set(key string, v []byte) SetResult

	// Get
	Get(key string) ([]byte, error)

	// Del if any error is returned, core.Core would not trigger synchronous of the cache.
	Del(key string) SetResult
}

// SetResult represents what operations would be caused by ICache.Set, operations include:
//
// 1. NeedSync tells users that them should trigger setting apply.
// 2. NeedDeleteKey should be set while Cache-Replacing happened or core need to delete cache by itself.
//
type SetResult struct {
	err           error
	NeedSync      bool
	NeedDeleteKey string
}

Now, cassemdb bootstrapped with fixed configuration of cassemdb cluster and it's members. But it should be another way:

1. root node1 bootstrap or recover the cluster.
2. node-2 to node-n join to the cluster which contains node1 only.