databend-query can't distinguish error meta service endpoints？

[lddlww]

databend version:

v0.9.16-nightly-3fcebe7(rust-1.68.0-nightly-2023-01-27T16:29:17.399818523Z)

my databend-query config toml with two error meta service endpoints：

[meta]
endpoints = ["10.5.16.18:9191","10.5.16.18:87:9191","10.5.16.18:49:9191"]
username = "root"
password = "root"
client_timeout_in_second = 60
auto_sync_interval = 60

[log.file]
level = "INFO"
format = "text"
dir = "/data/databend/9000/log"

start databend-query normal using above config without any warn or error

databend-query log like follow:

2023-01-29T10:27:17.955203Z  INFO common_meta_store: src/meta/store/src/lib.rs:124: use remote meta conf=RpcClientConf { endpoints: ["10.5.16.18:9191", "10.5.16.18:87:9191", "10.5.16.18:49:9191"], username: "root", password: "root", tls_conf: None, timeout: Some(60s), auto_sync_interval: None }
2023-01-29T10:27:17.955615Z  INFO worker_loop: common_meta_client::grpc_client: src/meta/client/src/grpc_client.rs:341: MetaGrpcClient::worker spawned

correct meta service endpoints

["10.5.16.18:9191","10.5.16.87:9191","10.5.16.49:9191"]

problem：
when i leaved 10.5.16.18:9191 meta service from correct meta service cluster , the databend-query can't be accessed and mysql client throw ERROR 1815 (HY000): endpoints is empty

[BohuTANG]

Ping @drmingdrmer

[drmingdrmer]

Did you mean leave a command ./target/debug/databend-meta --leave-via?

Let me have a test on my laptop first.

[drmingdrmer]

Ok I see your problem!

[drmingdrmer]

If the databend-meta cluster has only one member "10.5.16.18:9191", then removing it will cause the query to be unable to run.

Such behavior should be forbidden. I'll fix this, then.

And I think ERROR 1815 (HY000): endpoints is empty is a correct error in such a case: there are no reachable endpoints

[lddlww]

Such behavior should be forbidden. I'll fix this, then
thank you very much

And I think ERROR 1815 (HY000): endpoints is empty is a correct error in such a case: there are no reachable endpoints
yes，you are right

[drmingdrmer]

I've merged a fix that forbids removing the last node in a cluster:

• refactor(meta): do not remove the last node from a cluster #9781

[lddlww]

that's great work