Possible memory leak when driver fails to connect due to failed auth

[vlymar]

I was rotating my clickhouse password and there was a period of time during which my application didn't have the new db password, so connections were failing. At the same exact time, we saw a huge memory leak, up until the application was restarted with the correct password. Same pattern was reproduced a second time in another environment. There were corresponding logs about auth failures, presumably from ch/ecto reattempting to establish the connection. I'm 90% sure that during this time all read and write ops were disabled, so i think ecto or the driver itself was just trying to re-establish connections for the pool.

I'm not certain if this is the ch driver, ecto, or something else, so apologies if it's not in the driver. Just wanted to flag it in case anyone has intuition on the cause.

Thanks!

[ruslandoga]

👋 @vlymar

Thank you for the report! There seems to be a socket leak.

$ docker run -d -p 8123:8123 --ulimit nofile=262144:262144 clickhouse/clickhouse-server:23.3.7.5-alpine
$ iex

iex> Mix.install [:ch]
iex> Ch.start_link(username: "wrong", password: "wrong", pool_size: 20)
iex> length(Port.list())
32
# 10:37:05.447 [error] Ch.Connection (#PID<0.255.0>) failed to connect: ** (Ch.Error) Code: 516. DB::Exception: wrong: Authentication failed: password is incorrect, or there is no user with such name. (AUTHENTICATION_FAILED) (version 23.3.7.5 (official build))
# etc.
iex> length(Port.list())
72

I think I forgot to close the socket on error in connect.

[vlymar]

ah, no problem. thanks for opening the fix PR!

[ruslandoga]

I've published the fix as v0.2.3 https://diff.hex.pm/diff/ch/0.2.2..0.2.3

Thank you for the report once again @vlymar!