CASSGO-6 'Unable to discover cluster nodes with an empty rack name' issue fix #1785
Conversation
joao-r-reis
changed the title
tengu-alt
force-pushed
the
empty-rack-fix
branch
from
55e1d51 to
ed0c71d
Compare
| if err != nil { | ||
| return RowData{}, err | ||
| } | ||
| if column.Name == "rack" && column.Keyspace == "system" && (column.Table == "peers_v2" || column.Table == "peers") { |
There was a problem hiding this comment.
This seems like a hack, plus it could break backwards compatibility for existing users of RowData (outside of gocql) that read from those tables.
Could the issue be fixed another way, for example improving the API/adding new API for all users?
There was a problem hiding this comment.
I agree with addressing #1834 first, @martin-sucha can you provide feedback on #1834 ? I see your original comment on that issue, but I'm not sure if #1834 accurately implements what you were proposing.
tengu-alt
force-pushed
the
empty-rack-fix
branch
from
ed0c71d to
e5b7c46
Compare
| @@ -157,6 +157,7 @@ type HostInfo struct { | |||
| state nodeState | |||
| schemaVersion string | |||
| tokens []string | |||
| isRackNil bool | |||
There was a problem hiding this comment.
We can change rack to be *string, it's a private field. A conversion has to happen in the Rack() function though.
There was a problem hiding this comment.
Sounds good to me!
Let's wait for the feedback that you have asked for.
tengu-alt
force-pushed
the
empty-rack-fix
branch
from
e5b7c46 to
caf3893
Compare
This PR addresses issue #1706.
The driver was unable to discover nodes with an empty rack name because the
isValidPeer()function incorrectly checked for an empty string in the rack value (which is a valid case). The correct validation should ensure the rack name is not null (as a null value indicates an invalid peer).This PR fixes the method of retrieving the rack value from the
system.peersandsystem.peers_v2tables. Now, the rack value is set as a pointer to a string, allowing the value to be checked for null.