When all ZK Servers fail to resolve, erlzk becomes stuck in a loop about erlzk HOT 6 CLOSED

I think we should change resolve_server to return {ok, Result} or {error, Reason} rather than always a bare list. It should return {ok ,[]} only for {error, nxdomain} and return any others errors unchanged.

resolve_servers will need to be updated to handle the new {ok, Result} or {error, Reason} return from resolve_server, and also to return {ok, Result} or {error, Reason} instead of a bare list. It should return any errors from resolve_server immediately.

init will need to be updated to handle the new {ok, Result} or {error, Reason} return from resolve_servers and if {ok, []} or {error, Reason} are returned, it should return {stop, Reason} rather than continuing on to deduplicate and shuffle the list, and attempt to connect.

I think this behaviour is consistent with the Java implementation, and should fix the issue. I'll try to find time to make a pull request for this.

from erlzk.

Thank you for the detailed report.

Your ideas for fixing sounds good for me, and pull request is very welcomed.

from erlzk.

It might be worth tracking what type of error is returned by inet:gethostbyname (transient errors like timeouts, or permanent errors like nxdomain). If all calls to inet:gethostbyname return permanent errors then give up, but if all calls return transient errors then keep retrying, and if any calls succeed in resolving a name then continue.

When I made the changes related to DNS resolution, retrying etc in #9 I tried to make the behaviour of erlzk similar to the Java library for ZooKeeper, so it might be worth reading through that to see if it gives any hints. If I have any time I'll take a look.

from erlzk.

@GlenWalker That would be very nice, and thank you for your previous improvements to erlzk again.

from erlzk.

Thanks! We should probably also tag this as v0.6.4 too, since I updated package.exs and erlzk.app.src to 0.6.4

from erlzk.

@GlenWalker Yes, done. Really appreciated for your contribution.

from erlzk.