StackExchange.Redis: couple of questions about transactions

Question

I've got several question about using transactions from StackExchange.Redis:

1. Is it allowed to execute commands in transaction which could potentially target different nodes in cluster environment? Say, first command's key has {1} hashtag and second command's key has {2} hashtag.
2. When does transaction's Execute()/ExecuteAsync() returns false? Only when set conditions were not met? Could it return false if there were no conditions set?
3. If some network or internal Redis error occur, will Execute()/ExecuteAsync() throw or just return false? Should I also check commands' tasks (assuming that commands are fully correct and normally are not expected to throw) or they will just be cancelled?

Unfortunately the doc doesn't explain #2 and #3 in details.

Answer 1

I've digged a bit into StackExchange.Redis source code and played with the driver, here are my observations:

• no actual interaction happens when you call op methods on ITransaction instance
• driver starts talking to Redis only when you call Execute()/ExecuteAsync()

The answers seem to be the following:

1. As @royi-mindel mentioned, all keys in transaction must target the same slot. Otherwise you get "Multi-key operations must involve a single slot" exception.
2. Execute()/ExecuteAsync() returns false in two cases: when transaction was discarded because conditions were not met, and when driver failed to queue the command (for example because of server OOM). All command tasks will be marked as canceled. Also Execute()/ExecuteAsync() does not return false if one of the commands failed during execution (for example, because of wrong type operation).
3. In case some network problem occurs, Execute()/ExecuteAsync() will throw an exception and all commands tasks will remain in "waiting for activation" state.

To sum it up, command tasks should only be checked if Execute()/ExecuteAsync() returned true: every task will either contain a result or an error (see Exception property).