Starting many nodes at once throw exception address already in use about scalecube-services HOT 9 CLOSED

It is doing check that port is available and if yes bind to this port, but since there 50 nodes doing it in parallel there is a race condition between the check and binding.

For test purpose I recommend to either start each node on specific port or at least configure starting port different for each node in order to avoid contention for the ports at all.

Here is a transport config example how to start each node on different fixed port:

TransportConfig conf = TransportConfig.builder()
  .port(TransportConfig.DEFAULT_PORT + nodeIndex)
  .portAutoIncrement(false)
  .build()

from scalecube-services.

i understand there are ways to workaround this issue but i think that its better to locate port until success
users might fall on taken ports even if they are not taken by cluster code it self and still would like to start a node successfully - not only because of this specific case but also by the desire to run several nodes under same container or just by mistake try to connect on already used port.

from scalecube-services.

Just for clarity, if you try to connect on already used port it will not fail with port already used exception since it will select next available one, the issue is specific for the case when you start many cluster nodes on same machine at same moment of time and all of them trying to bind to the same port range. So outside of the code snippet above it would pretty difficult to reproduce it otherwise.

from scalecube-services.

ok so can it be same situation that someone take my port at this exact moment of time?
my process starts and someone take the port... is it really matters who?

from scalecube-services.

It's possible only in case if they start practically in the same moment of time like in the range of some nano seconds or so.

from scalecube-services.

yep - exactly.
so need to fail-over and not fail

from scalecube-services.

Yes, but it's practically impossible outside of the code snippet above where you start several nodes in the for-loop concurrently.

from scalecube-services.

i didnt understand why its not possible
i start just one node and someone in some rare moment of time just take this port...

from scalecube-services.

Fixed in PR: #117

from scalecube-services.