TransactionState: add TransactionCost #34881
Conversation
apfitzge
force-pushed
the
transaction_state_container-transaction_cost
branch
from
329bc51
to
62d5ed2
Compare
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## master #34881 +/- ##
=======================================
Coverage 81.7% 81.7%
=======================================
Files 826 826
Lines 223413 223441 +28
=======================================
+ Hits 182614 182681 +67
+ Misses 40799 40760 -39 |
| let cu_limit = transaction_state | ||
| .transaction_priority_details() | ||
| .compute_unit_limit; | ||
| let cost = transaction_state.transaction_cost().sum(); |
There was a problem hiding this comment.
Just a quick check: for Vote, its cu_limit is always 0, but has value of 3428 for cost. By using different CUs, will it impact how vote's scheduled?
There was a problem hiding this comment.
vote's don't go through the scheduler - they have their own independent threads still.
There was a problem hiding this comment.
this PR realy has no impact on its' own. The CUs are tracked, but not used for load-balancing currently because they are so unreliable. This is intended to be used later on, but right now its' just tracked, so changing from cu_limit to cost shouldn't impact anything.
There was a problem hiding this comment.
Gotcha, good to know it is for the purpose of load-balancing, not counting towards the limits.
Also worth noting that transaction_cost includes requested_cu_limit, if used. So that unreliability also in transaction_cost atm.
There was a problem hiding this comment.
Right, a major difference here is that for simple txs which we know the cost of, it will not include "requested_cus". For example a simple transfer. So for purposes of prioritizing by reward/cu we want to use this number
There was a problem hiding this comment.
that would de-prioritize txs with large requested_cu but actually not-so-largge_actual-cu, if signature fee is included in reward/cu calculation, would it?
Problem
Summary of Changes
This will allow 2 subsequent improvements:
Fixes #