Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
 Peloton Self-Driving Infrastructure #1346
This PR is working toward an infrastructure for self-driving components on Peloton.
In this checkpoint, we presented:
-- did sanity check by printing out latencies: for identical traces of SQL queries, the latencies drop when Timer start/stop is moved into plan_executor as opposed to txn_manager
-- side benefit is that latency of internal transactions (such as those generated by stats aggregation thread) is not recorded, since those don't go through plan_executor. I think this is what we want?
This is not going to add much overhead to the engine as the collector would not be blocked. Actually, the blocking thing is the background aggregating thread as it is CAS on the safe_ flag.
We would like the test the overhead with some TPCC testbench but it seems that the master branch is running slow. We would definitely run TPCC if we get fast enough later, and we claim it's not going to be hard (just change the stats_mode would switch the collection from on/off).