* 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>
* Add separate vote processing tpu port
* Add feature to send to tpu vote port
* Add vote rejecting sigverify mode
* use packet.meta.is_simple_vote_tx in place of deserialization
* consolidate code that identifies vote tx atcommon path for cpu and gpu
* new key for feature set
* banking forward tpu vote
* add tpu vote port to dockerfile and other review changes
* Simplify thread id compare
* fix a test; updated cluster_info ABI change
Co-authored-by: Tao Zhu <tao@solana.com>
* removes Slot from TransmitShreds (#19327)
An earlier version of the code was funneling through stakes along with
shreds to broadcast:
https://github.com/solana-labs/solana/blob/b67ffab37/core/src/broadcast_stage.rs#L127
This was changed to only slots as stakes computation was pushed further
down the pipeline in:
https://github.com/solana-labs/solana/pull/18971
However shreds themselves embody which slot they belong to. So pairing
them with slot is redundant and adds rooms for bugs should they become
inconsistent.
(cherry picked from commit 1deb4add81)
# Conflicts:
# core/benches/cluster_info.rs
# core/src/broadcast_stage.rs
# core/src/broadcast_stage/broadcast_duplicates_run.rs
# core/src/broadcast_stage/fail_entry_verification_broadcast_run.rs
# core/src/broadcast_stage/standard_broadcast_run.rs
* removes backport merge conflicts
Co-authored-by: behzad nouri <behzadnouri@gmail.com>
* removes packet-count metrics from retransmit stage
Working towards sending shreds (instead of packets) to retransmit stage
so that shreds recovered from erasure codes are as well retransmitted.
Following commit will add these metrics back to window-service, earlier
in the pipeline.
(cherry picked from commit bf437b0336)
# Conflicts:
# core/src/retransmit_stage.rs
* adds packet/shred count stats to window-service
Adding back these metrics from the earlier commit which removed them
from retransmit stage.
(cherry picked from commit 8198a7eae1)
* removes erroneous uses of Arc<...> from retransmit stage
(cherry picked from commit 6e413331b5)
# Conflicts:
# core/src/retransmit_stage.rs
# core/src/tvu.rs
* sends shreds (instead of packets) to retransmit stage
Working towards channelling through shreds recovered from erasure codes
to retransmit stage.
(cherry picked from commit 3efccbffab)
# Conflicts:
# core/src/retransmit_stage.rs
* returns completed-data-set-info from insert_data_shred
instead of opaque (u32, u32) which are then converted to
CompletedDataSetInfo at the call-site.
(cherry picked from commit 3c71670bd9)
# Conflicts:
# ledger/src/blockstore.rs
* retransmits shreds recovered from erasure codes
Shreds recovered from erasure codes have not been received from turbine
and have not been retransmitted to other nodes downstream. This results
in more repairs across the cluster which is slower.
This commit channels through recovered shreds to retransmit stage in
order to further broadcast the shreds to downstream nodes in the tree.
(cherry picked from commit 7a8807b8bb)
# Conflicts:
# core/src/retransmit_stage.rs
# core/src/window_service.rs
* removes backport merge conflicts
Co-authored-by: behzad nouri <behzadnouri@gmail.com>
* windows: Make solana-test-validator work (#20099)
* windows: Make solana-test-validator work
The important changes to get this going on Windows:
* ledger lock needs to be done on a file instead of the directory
* IPC service needs to use the Windows pipe naming scheme
* always disable the JIT
* file logging not possible yet because we can't redirect stderr,
but this will change once env_logger fixes the pipe output target!
* Integrate review feedback
(cherry picked from commit 567f30aa1a)
# Conflicts:
# validator/src/bin/solana-test-validator.rs
# validator/src/lib.rs
# validator/src/main.rs
* Fix merge conflicts
Co-authored-by: Jon Cinque <jon.cinque@gmail.com>
Broadcast stage and retransmit stage should arrange nodes on turbine
broadcast tree in exactly same order. Additionally any changes to this
ordering (e.g. updating how unstaked nodes are handled) requires feature
gating to keep the cluster in sync.
Current implementation is scattered out over several public methods and
exposes too much of implementation details (e.g. usize indices into
peers vector) which makes code changes and checking for feature
activations more difficult.
This commit encapsulates turbine peer computations into a new struct,
and only exposes two public methods, get_broadcast_peer and
get_retransmit_peers, for call-sites.
(cherry picked from commit 04787be8b1)
Co-authored-by: behzad nouri <behzadnouri@gmail.com>
* Move gossip modules to solana-gossip
* Update Protocol abi digest due to move
* Move gossip benches and hook up CI
* Remove unneeded Result entries
* Single use statements
For all code paths (gossip push, pull, purge, etc) that remove or
override a crds value, it is necessary to record hash of values purged
from crds table, in order to exclude them from subsequent pull-requests;
otherwise the next pull request will likely return outdated values,
wasting bandwidth:
https://github.com/solana-labs/solana/blob/ed51cde37/core/src/crds_gossip_pull.rs#L486-L491
Currently this is done all over the place in multiple modules, and this
has caused bugs in the past where purged values were not recorded.
This commit encapsulated this bookkeeping into crds module, so that any
code path which removes or overrides a crds value, also records the hash
of purged value in-place.
It is crucial that VersionedCrdsValue::insert_timestamp does not go
backward in time:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/crds.rs#L67-L79
Otherwise methods such as get_votes and get_epoch_slots_since will
break, which will break their downstream flow, including vote-listener
and optimistic confirmation:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L1197-L1215https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L1274-L1298
For that, Crds::new_versioned is intended to be called "atomically" with
Crds::insert_verioned (as the comment already says so):
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/crds.rs#L126-L129
However, currently this is violated in the code. For example,
filter_pull_responses creates VersionedCrdsValues (with the current
timestamp), then acquires an exclusive lock on gossip, then
process_pull_responses writes those values to the crds table:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L2375-L2392
Depending on the workload and lock contention, the insert_timestamps may
well be in the past when these values finally are inserted into gossip.
To avoid such scenarios, this commit:
* removes Crds::new_versioned and Crd::insert_versioned.
* makes VersionedCrdsValue constructor private, only invoked in
Crds::insert, so that insert_timestamp is populated right before
insert.
This will improve insert_timestamp monotonicity as long as Crds::insert
is not called with a stalled timestamp. Following commits may further
improve this by calling timestamp() inside Crds::insert, and/or
switching to std::time::Instant which guarantees monotonicity.
Number of parity coding shreds is always less than the number of data
shreds in FEC blocks:
https://github.com/solana-labs/solana/blob/6907a2366/ledger/src/shred.rs#L719
Data shreds are batched in chunks of 32 shreds each:
https://github.com/solana-labs/solana/blob/6907a2366/ledger/src/shred.rs#L714
However the very last batch of data shreds in a slot can be small, in
which case the loss rate can be exacerbated.
This commit expands the number of coding shreds in the last FEC block in
slots to: 64 - number of data shreds; so that FEC blocks are always 64
data and parity coding shreds each.
As a consequence of this, the last FEC block has more parity coding
shreds than data shreds. So for some shred indices we will have a coding
shred but no data shreds. This should not cause any kind of overlapping
FEC blocks as in:
https://github.com/solana-labs/solana/pull/10095
since this is done only for the very last batch in a slot, and the next
slot will reset the shred index.
* Track transaction check time separately from account loads
* banking packet process metrics
* Remove signature clone in status cache lookup
* Reduce allocations when converting packets to transactions
* Add blake3 hash of transaction messages in status cache
* Bug fixes
* fix tests and run fmt
* Address feedback
* fix simd tx entry verification
* Fix rebase
* Feedback
* clean up
* Add tests
* Remove feature switch and fall back to signature check
* Bump programs/bpf Cargo.lock
* clippy
* nudge benches
* Bump `BankSlotDelta` frozen ABI hash`
* Add blake3 to sdk/programs/Cargo.lock
* nudge bpf tests
* short circuit status cache checks
Co-authored-by: Trent Nelson <trent@solana.com>
In several places in gossip code, the entire crds table is scanned only
to filter out nodes' contact infos. Currently on mainnet, crds table is
of size ~70k, while there are only ~470 nodes. So the full table scan is
inefficient. Instead we may maintain an index of only nodes' contact
infos.
Based on run-time profiles, the majority time of new_pull_requests is
spent building bloom filters, in hashing and bit-vec ops.
This commit builds crds filters in parallel using rayon constructs. The
added benchmark shows ~5x speedup (4-core machine, 8 threads).
* Save/restore Tower
* Avoid unwrap()
* Rebase cleanups
* Forcibly pass test
* Correct reconcilation of votes after validator resume
* d b g
* Add more tests
* fsync and fix test
* Add test
* Fix fmt
* Debug
* Fix tests...
* save
* Clarify error message and code cleaning around it
* Move most of code out of tower save hot codepath
* Proper comment for the lack of fsync on tower
* Clean up
* Clean up
* Simpler type alias
* Manage tower-restored ancestor slots without banks
* Add comment
* Extract long code blocks...
* Add comment
* Simplify returned tuple...
* Tweak too aggresive log
* Fix typo...
* Add test
* Update comment
* Improve test to require non-empty stray restored slots
* Measure tower save and dump all tower contents
* Log adjust and add threshold related assertions
* cleanup adjust
* Properly lower stray restored slots priority...
* Rust fmt
* Fix test....
* Clarify comments a bit and add TowerError::TooNew
* Further clean-up arround TowerError
* Truly create ancestors by excluding last vote slot
* Add comment for stray_restored_slots
* Add comment for stray_restored_slots
* Use BTreeSet
* Consider root_slot into post-replay adjustment
* Tweak logging
* Add test for stray_restored_ancestors
* Reorder some code
* Better names for unit tests
* Add frozen_abi to SavedTower
* Fold long lines
* Tweak stray ancestors and too old slot history
* Re-adjust error conditon of too old slot history
* Test normal ancestors is checked before stray ones
* Fix conflict, update tests, adjust behavior a bit
* Fix test
* Address review comments
* Last touch!
* Immediately after creating cleaning pr
* Revert stray slots
* Revert comment...
* Report error as metrics
* Revert not to panic! and ignore unfixable test...
* Normalize lockouts.root_slot more strictly
* Add comments for panic! and more assertions
* Proper initialize root without vote account
* Clarify code and comments based on review feedback
* Fix rebase
* Further simplify based on assured tower root
* Reorder code for more readability
Co-authored-by: Michael Vines <mvines@gmail.com>
filter_crds_values checks every crds filter against every hash value:
https://github.com/solana-labs/solana/blob/ee646aa7/core/src/crds_gossip_pull.rs#L432
which can be inefficient if the filter's bit-mask only matches small
portion of the entire crds table.
This commit shards crds values into separate tables based on shard_bits
first bits of their hash prefix. Given a (mask, mask_bits) filter,
filtering crds can be done by inspecting only relevant shards.
If CrdsFilter.mask_bits <= shard_bits, then precisely only the crds
values which match (mask, mask_bits) bit pattern are traversed.
If CrdsFilter.mask_bits > shard_bits, then approximately only
1/2^shard_bits of crds values are inspected.
Benchmarking on a gce cluster of 20 nodes, I see ~10% improvement in
generate_pull_responses metric, but with larger clusters, crds table and
2^mask_bits are both larger, so the impact should be more significant.
