Closed ekmartin closed 6 years ago
I think we actually specifically want batch_put
here, because otherwise we generate a single potentially very large update (e.g., if you used -a 10000000
). Maybe the answer is to do multi_put
inside of the batch_put
... I'm surprised to hear that multi_put
is significantly faster than batch_put
, as they should both just fill up TCP buffers...
Closing this now that batch_put
doesn't hang.
I think ultimately we'll want batch_put
+ multi_put
. if a batch_put
runs for too long without checking for acks, it could totally generate a deadlock. The domain of the base will fill the TCP buffer trying to send more acks, but that buffer will never drain because the sender will block on the domain not reading from the base input. The alternative is to make batch_put
occasionally read acks.
Doesn't matter too much, but saves a little waiting time.