Most aggregate functions support multi-stage distributed execution in which on every node we perform a local aggregation, then we route all the intermediate results to all of the nodes (according to the grouping columns), and then we perform a final aggregation that gives us the result. For some aggregate functions the decomposition into local and final stages is trivial - e.g. in order to calcula

Now insert and query share the resource ( Max Process Count control) 。 When the query with high TPS，the insert will get error (“error: too many process”). I think separator the resource for Insert and Query will makes sense. Ensure enough resource for insert。It looks like Use Yarn， Insert and Query use the different resource quota。
Or the simple way , Can we set Ratio for Insert and

The test case SchemaBuilderTest is not make sense. It is better to redesign it for testing.

The network option setting code uses the default option code, which interprets the bytes passed to the option as a binary encoded 8 byte integer. The only way to set an integer value is to set the environment variable to the 8 byte binary encoding.

Hello Philip!
I think there is an issue with this part of the code of rqlite (store/store.go).

func (s *Store) Database(leader bool) ([]byte, error) {
	if leader && s.raft.State() != raft.Leader {
		return nil, ErrNotLeader
	}
	// Ensure only one snapshot can take place at once, and block all queries.
	s.mu.Lock()
	defer s.mu.Unlock()

	f, err := ioutil.TempFile("", "rqlilte-snap-

Right now there's no way to determine if a tablet server has bootstrapped and is ready without calling the RPC method IsTabletServerReady. It would be useful if this was exposed in yb-ts-cli so we can determine when deploying if it's safe to continue the playbook and start running DDL commands and/or remove a leadership blacklist.

Some tests use the following pattern

try {
    ... do something ...
    fail();
}
catch (IllegalStateException e) {
    assertEquals(e.getMessage(), "the expected message");
}

this is not very readable and can be expressed more tersely with assertThatThrownBy
also, in case the message not as expected, the actual failure gets lots, so it's harder to diagnose the problem

Currently we don't have any mechanism to limit the maximum number of clients that could be handled simultaneously.

This feature should be designed properly. Here is some clue: https://redis.io/topics/clients#maximum-number-of-clients

Now that we support more than 1 data type, we should include the values data type in cache nodes. This give us a way of easily identifying what type of data a key points to e.g. string, queue (and more types in the future)

Currently a cache node object consists of a Key, a Value and a TTL. We should also add a field specifying the type of the Value