gaspersic/solana
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Cost model 1.7 (#20188)

Browse Source 
* Cost Model to limit transactions which are not parallelizeable (#16694)

* * Add following to banking_stage:
  1. CostModel as immutable ref shared between threads, to provide estimated cost for transactions.
  2. CostTracker which is shared between threads, tracks transaction costs for each block.

* replace hard coded program ID with id() calls

* Add Account Access Cost as part of TransactionCost. Account Access cost are weighted differently between read and write, signed and non-signed.

* Establish instruction_execution_cost_table, add function to update or insert instruction cost, unit tested. It is read-only for now; it allows Replay to insert realtime instruction execution costs to the table.

* add test for cost_tracker atomically try_add operation, serves as safety guard for future changes

* check cost against local copy of cost_tracker, return transactions that would exceed limit as unprocessed transaction to be buffered; only apply bank processed transactions cost to tracker;

* bencher to new banking_stage with max cost limit to allow cost model being hit consistently during bench iterations

* replay stage feed back program cost (#17731)

* replay stage feeds back realtime per-program execution cost to cost model;

* program cost execution table is initialized into empty table, no longer populated with hardcoded numbers;

* changed cost unit to microsecond, using value collected from mainnet;

* add ExecuteCostTable with fixed capacity for security concern, when its limit is reached, programs with old age AND less occurrence will be pushed out to make room for new programs.

* investigate system performance test degradation  (#17919)

* Add stats and counter around cost model ops, mainly:
- calculate transaction cost
- check transaction can fit in a block
- update block cost tracker after transactions are added to block
- replay_stage to update/insert execution cost to table

* Change mutex on cost_tracker to RwLock

* removed cloning cost_tracker for local use, as the metrics show clone is very expensive.

* acquire and hold locks for block of TXs, instead of acquire and release per transaction;

* remove redundant would_fit check from cost_tracker update execution path

* refactor cost checking with less frequent lock acquiring

* avoid many Transaction_cost heap allocation when calculate cost, which
is in the hot path - executed per transaction.

* create hashmap with new_capacity to reduce runtime heap realloc.

* code review changes: categorize stats, replace explicit drop calls, concisely initiate to default

* address potential deadlock by acquiring locks one at time

* Persist cost table to blockstore (#18123)

* Add `ProgramCosts` Column Family to blockstore, implement LedgerColumn; add `delete_cf` to Rocks
* Add ProgramCosts to compaction excluding list alone side with TransactionStatusIndex in one place: `excludes_from_compaction()`

* Write cost table to blockstore after `replay_stage` replayed active banks; add stats to measure persist time
* Deletes program from `ProgramCosts` in blockstore when they are removed from cost_table in memory
* Only try to persist to blockstore when cost_table is changed.
* Restore cost table during validator startup

* Offload `cost_model` related operations from replay main thread to dedicated service thread, add channel to send execute_timings between these threads;
* Move `cost_update_service` to its own module; replay_stage is now decoupled from cost_model.

* log warning when channel send fails (#18391)

* Aggregate cost_model into cost_tracker (#18374)

* * aggregate cost_model into cost_tracker, decouple it from banking_stage to prevent accidental deadlock. * Simplified code, removed unused functions

* review fixes

* update ledger tool to restore cost table from blockstore (#18489)

* update ledger tool to restore cost model from blockstore when compute-slot-cost

* Move initialize_cost_table into cost_model, so the function can be tested and shared between validator and ledger-tool

* refactor and simplify a test

* manually fix merge conflicts

* Per-program id timings (#17554)

* more manual fixing

* solve a merge conflict

* featurize cost model

* more merge fix

* cost model uses compute_unit to replace microsecond as cost unit
(#18934)

* Reject blocks for costs above the max block cost (#18994)

* Update block max cost limit to fix performance regession (#19276)

* replace function with const var for better readability (#19285)

* Add few more metrics data points (#19624)

* periodically report sigverify_stage stats (#19674)

* manual merge

* cost model nits (#18528)

* Accumulate consumed units (#18714)

* tx wide compute budget (#18631)

* more manual merge

* ignore zerorize drop security

* - update const cost values with data collected by #19627
- update cost calculation to closely proposed fee schedule #16984

* add transaction cost histogram metrics (#20350)

* rebase to 1.7.15

* add tx count and thread id to stats (#20451)
each stat reports and resets when slot changes

* remove cost_model feature_set

* ignore vote transactions from cost model

Co-authored-by: sakridge <sakridge@gmail.com>
Co-authored-by: Jeff Biseda <jbiseda@gmail.com>
Co-authored-by: Jack May <jack@solana.com>
This commit is contained in:
Tao Zhu
2021-10-06 15:11:41 -05:00
committed by Trent Nelson
parent a4df784e82
commit db85d659b9
40 changed files with 3208 additions and 266 deletions
Download Patch File Download Diff File Expand all files Collapse all files

279
core/src/execute_cost_table.rs Normal file
View File

@@ -0,0 +1,279 @@
/// ExecuteCostTable is aggregated by Cost Model, it keeps each program's
/// average cost in its HashMap, with fixed capacity to avoid from growing
/// unchecked.
/// When its capacity limit is reached, it prunes old and less-used programs
/// to make room for new ones.
use log::*;
use solana_sdk::pubkey::Pubkey;
use std::{collections::HashMap, time::SystemTime};
// prune is rather expensive op, free up bulk space in each operation
// would be more efficient. PRUNE_RATIO defines the after prune table
// size will be original_size * PRUNE_RATIO.
const PRUNE_RATIO: f64 = 0.75;
// with 50_000 TPS as norm, weights occurrences '100' per microsec
const OCCURRENCES_WEIGHT: i64 = 100;
const DEFAULT_CAPACITY: usize = 1024;
#[derive(Debug)]
pub struct ExecuteCostTable {
capacity: usize,
table: HashMap<Pubkey, u64>,
occurrences: HashMap<Pubkey, (usize, SystemTime)>,
}
impl Default for ExecuteCostTable {
fn default() -> Self {
ExecuteCostTable::new(DEFAULT_CAPACITY)
}
}
impl ExecuteCostTable {
pub fn new(cap: usize) -> Self {
Self {
capacity: cap,
table: HashMap::with_capacity(cap),
occurrences: HashMap::with_capacity(cap),
}
}
pub fn get_cost_table(&self) -> &HashMap<Pubkey, u64> {
&self.table
}
pub fn get_count(&self) -> usize {
self.table.len()
}
// instead of assigning unknown program with a configured/hard-coded cost
// use average or mode function to make a educated guess.
pub fn get_average(&self) -> u64 {
if self.table.is_empty() {
0
} else {
self.table.iter().map(|(_, value)| value).sum::<u64>() / self.get_count() as u64
}
}
pub fn get_mode(&self) -> u64 {
if self.occurrences.is_empty() {
0
} else {
let key = self
.occurrences
.iter()
.max_by_key(|&(_, count)| count)
.map(|(key, _)| key)
.expect("cannot find mode from cost table");
*self.table.get(&key).unwrap()
}
}
// returns None if program doesn't exist in table. In this case,
// client is advised to call `get_average()` or `get_mode()` to
// assign a 'default' value for new program.
pub fn get_cost(&self, key: &Pubkey) -> Option<&u64> {
self.table.get(&key)
}
pub fn upsert(&mut self, key: &Pubkey, value: u64) -> Option<u64> {
let need_to_add = self.table.get(key).is_none();
let current_size = self.get_count();
if current_size == self.capacity && need_to_add {
self.prune_to(&((current_size as f64 * PRUNE_RATIO) as usize));
}
let program_cost = self.table.entry(*key).or_insert(value);
*program_cost = (*program_cost + value) / 2;
let (count, timestamp) = self
.occurrences
.entry(*key)
.or_insert((0, SystemTime::now()));
*count += 1;
*timestamp = SystemTime::now();
Some(*program_cost)
}
// prune the old programs so the table contains `new_size` of records,
// where `old` is defined as weighted age, which is negatively correlated
// with program's age and
// positively correlated with how frequently the program
// is executed (eg. occurrence),
fn prune_to(&mut self, new_size: &usize) {
debug!(
"prune cost table, current size {}, new size {}",
self.get_count(),
new_size
);
if *new_size == self.get_count() {
return;
}
if *new_size == 0 {
self.table.clear();
self.occurrences.clear();
return;
}
let now = SystemTime::now();
let mut sorted_by_weighted_age: Vec<_> = self
.occurrences
.iter()
.map(|(key, (count, timestamp))| {
let age = now.duration_since(*timestamp).unwrap().as_micros();
let weighted_age = *count as i64 * OCCURRENCES_WEIGHT + -(age as i64);
(weighted_age, *key)
})
.collect();
sorted_by_weighted_age.sort_by(|x, y| x.0.partial_cmp(&y.0).unwrap());
for i in sorted_by_weighted_age.iter() {
self.table.remove(&i.1);
self.occurrences.remove(&i.1);
if *new_size == self.get_count() {
break;
}
}
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_execute_cost_table_prune_simple_table() {
solana_logger::setup();
let capacity: usize = 3;
let mut testee = ExecuteCostTable::new(capacity);
let key1 = Pubkey::new_unique();
let key2 = Pubkey::new_unique();
let key3 = Pubkey::new_unique();
testee.upsert(&key1, 1);
testee.upsert(&key2, 2);
testee.upsert(&key3, 3);
testee.prune_to(&(capacity - 1));
// the oldest, key1, should be pruned
assert!(testee.get_cost(&key1).is_none());
assert!(testee.get_cost(&key2).is_some());
assert!(testee.get_cost(&key2).is_some());
}
#[test]
fn test_execute_cost_table_prune_weighted_table() {
solana_logger::setup();
let capacity: usize = 3;
let mut testee = ExecuteCostTable::new(capacity);
let key1 = Pubkey::new_unique();
let key2 = Pubkey::new_unique();
let key3 = Pubkey::new_unique();
testee.upsert(&key1, 1);
testee.upsert(&key1, 1);
testee.upsert(&key2, 2);
testee.upsert(&key3, 3);
testee.prune_to(&(capacity - 1));
// the oldest, key1, has 2 counts; 2nd oldest Key2 has 1 count;
// expect key2 to be pruned.
assert!(testee.get_cost(&key1).is_some());
assert!(testee.get_cost(&key2).is_none());
assert!(testee.get_cost(&key3).is_some());
}
#[test]
fn test_execute_cost_table_upsert_within_capacity() {
solana_logger::setup();
let mut testee = ExecuteCostTable::default();
let key1 = Pubkey::new_unique();
let key2 = Pubkey::new_unique();
let cost1: u64 = 100;
let cost2: u64 = 110;
// query empty table
assert!(testee.get_cost(&key1).is_none());
// insert one record
testee.upsert(&key1, cost1);
assert_eq!(1, testee.get_count());
assert_eq!(cost1, testee.get_average());
assert_eq!(cost1, testee.get_mode());
assert_eq!(&cost1, testee.get_cost(&key1).unwrap());
// insert 2nd record
testee.upsert(&key2, cost2);
assert_eq!(2, testee.get_count());
assert_eq!((cost1 + cost2) / 2_u64, testee.get_average());
assert_eq!(cost2, testee.get_mode());
assert_eq!(&cost1, testee.get_cost(&key1).unwrap());
assert_eq!(&cost2, testee.get_cost(&key2).unwrap());
// update 1st record
testee.upsert(&key1, cost2);
assert_eq!(2, testee.get_count());
assert_eq!(((cost1 + cost2) / 2 + cost2) / 2, testee.get_average());
assert_eq!((cost1 + cost2) / 2, testee.get_mode());
assert_eq!(&((cost1 + cost2) / 2), testee.get_cost(&key1).unwrap());
assert_eq!(&cost2, testee.get_cost(&key2).unwrap());
}
#[test]
fn test_execute_cost_table_upsert_exceeds_capacity() {
solana_logger::setup();
let capacity: usize = 2;
let mut testee = ExecuteCostTable::new(capacity);
let key1 = Pubkey::new_unique();
let key2 = Pubkey::new_unique();
let key3 = Pubkey::new_unique();
let key4 = Pubkey::new_unique();
let cost1: u64 = 100;
let cost2: u64 = 110;
let cost3: u64 = 120;
let cost4: u64 = 130;
// insert one record
testee.upsert(&key1, cost1);
assert_eq!(1, testee.get_count());
assert_eq!(&cost1, testee.get_cost(&key1).unwrap());
// insert 2nd record
testee.upsert(&key2, cost2);
assert_eq!(2, testee.get_count());
assert_eq!(&cost1, testee.get_cost(&key1).unwrap());
assert_eq!(&cost2, testee.get_cost(&key2).unwrap());
// insert 3rd record, pushes out the oldest (eg 1st) record
testee.upsert(&key3, cost3);
assert_eq!(2, testee.get_count());
assert_eq!((cost2 + cost3) / 2_u64, testee.get_average());
assert_eq!(cost3, testee.get_mode());
assert!(testee.get_cost(&key1).is_none());
assert_eq!(&cost2, testee.get_cost(&key2).unwrap());
assert_eq!(&cost3, testee.get_cost(&key3).unwrap());
// update 2nd record, so the 3rd becomes the oldest
// add 4th record, pushes out 3rd key
testee.upsert(&key2, cost1);
testee.upsert(&key4, cost4);
assert_eq!(((cost1 + cost2) / 2 + cost4) / 2_u64, testee.get_average());
assert_eq!((cost1 + cost2) / 2, testee.get_mode());
assert_eq!(2, testee.get_count());
assert!(testee.get_cost(&key1).is_none());
assert_eq!(&((cost1 + cost2) / 2), testee.get_cost(&key2).unwrap());
assert!(testee.get_cost(&key3).is_none());
assert_eq!(&cost4, testee.get_cost(&key4).unwrap());
}
}