Dynamic dispatch forces heap allocation and adds extra overhead.
Dynamic casting as in the ones below, lacks compile-time type safety:
https://github.com/solana-labs/solana/blob/eeb622c4e/quic-client/src/lib.rs#L172-L175https://github.com/solana-labs/solana/blob/eeb622c4e/udp-client/src/lib.rs#L52-L55
The commit removes all instances of Any, Box<dyn ...>, and Arc<dyn ...>,
and instead uses generic and associated types.
There are only two protocols QUIC and UDP; and the code which has to
work with both protocols can use a trivial thin enum wrapper.
With respect to connection-cache specifically:
* connection-cache/ConnectionCache is a single protocol cache which
allows to use either QUIC or UDP without any build dependency on the
other protocol.
* client/ConnectionCache is an enum wrapper around both protocols and
can be used in the code which has to work with both QUIC and UDP.
Co-authored-by: Tyera Eulberg <tyera@solana.com>
tpu-client/tpu_connection_cache is refactored out the module and moved to connection-cache/connection_cache and the logic in client/connection_cache is consolidated to connection-cache/connection_cache as well. client/connection_cache only has a thin wrapper which forward calls to connection-cache/connection_cache and deal with constructions of quic/udp connection cache for clients using them both.2.
The TpuConnection is refactored to ClientConnection to make it generic and functions renamed to be proper for other workflows. eg. tpu_addr -> server_addr, send_transaction --> send_data and etc...
The enum dispatch is removed so that we can make the bulk of code of quic and udp agnostic of each other. The client is possible to load quic or udp only into its runtime.
The generic type parameter in the tpu-client/tpu_connection_cache is removed in order to create both quic and udp connection cache and use the object to send transactions with multiple branching when sending data. The generic type parameters and associated types are dropped in other types in order to make the trait "object safe" for this purpose.
I have annotated the code explaining the reasoning and the refactoring source -> destination.
There is no functional changes
bench-tps has been performed for rpc-client, thin-client and tpu-client. And it is found the performance number largely match the ones before the refactoring.
* Add run parent directory for accounts files
* fix test test_concurrent_snapshot_packaging
* review comments. renamed the path setup function
* Addressed most of the review comments
* remove explict type def for map result
* handle create_accounts_run_and_snapshot_dirs error with expect
* update with more review comments
* minor fixes from review comments
* simplify account_filename option assignment
* handle error from create_accounts_run_and_snapshot_dirs
* use then instead of then_some for lazy evaluation
* Clean up files in the old account_path before trasitioning to the new run path
* try_exist and accounts_dir removing extra
* sync rmdir, is_dir check
* handle the account_path not deletable case
* Add fully-reproducible online tracer for banking
* Don't use eprintln!()...
* Update programs/sbf/Cargo.lock...
* Remove meaningless assert_eq
* Group test-only code under aptly named mod
* Remove needless overflow handling in receive_until
* Delay stat aggregation as it's possible now
* Use Cow to avoid needless heap allocs
* Properly consume metrics action as soon as hold
* Trace UnprocessedTransactionStorage::len() instead
* Loosen joining api over type safety for replaystage
* Introce hash event to override these when simulating
* Use serde_with/serde_as instead of hacky workaround
* Update another Cargo.lock...
* Add detailed comment for Packet::buffer serialize
* Rename sender_overhead_minimized_receiver_loop()
* Use type interference for TraceError
* Another minor rename
* Retire now useless ForEach to simplify code
* Use type alias as much as possible
* Properly translate and propagate tracing errors
* Clarify --enable-banking-trace with better naming
* Consider unclean (signal-based) node restarts..
* Tweak logging and cli
* Remove Bank events as it's not needed anymore
* Make tpu own banking tracer thread
* Reduce diff a bit..
* Use latest serde_with
* Finally use the published rolling-file crate
* Make test code change more consistent
* Revive dead and non-terminating test code path...
* Dispose batches early now that possible
* Split off thread handle very early at ::new()
* Tweak message for TooSmallDirByteLimitl
* Remove too much of indirection
* Remove needless pub from ::channel()
* Clarify test comments
* Avoid needless event creation if tracer is disabled
* Write tests around file rotation and spill-over
* Remove unneeded PathBuf::clone()s...
* Introduce inner struct instead of tuple...
* Remove unused enum BankStatus...
* Avoid .unwrap() for the case of disabled tracer...
* introduce workspace.package
* introduce workspace.dependencies
* read version from root cargo.toml
* pass check when version = { workspace = true }
* don't bump version when version = { workspace = true }
* including workspace Cargo.toml when bump version
* programs/sbf use workspace inheritance
* fix increasing cargo version ignore program/sbf/Cargo.toml
* Add run parent directory for accounts files
* fix test test_concurrent_snapshot_packaging
* review comments. renamed the path setup function
* Addressed most of the review comments
* remove explict type def for map result
* handle create_accounts_run_and_snapshot_dirs error with expect
* update with more review comments
* minor fixes from review comments
* simplify account_filename option assignment
* handle error from create_accounts_run_and_snapshot_dirs
* use then instead of then_some for lazy evaluation
find_packet_sender_stake_stage::FindPacketSenderStakeReceiver is quite
verbose to include in function arguments, and type name is descriptive
enough that it doesn't need to be qualified with the crate name in every
instance.
Currently, the cleanup service counts the number of shreds in the
database by iterating the entire SlotMeta column and reading the number
of received shreds for each slot. This gives us a fairly accurate count
at the expense of performing a good amount of IO.
Instead of counting the individual slots, use the live_files()
rust-rocksdb entrypoint that we expose in Blockstore. This API allows us
to get the number of entries (shreds) in the data shred column family by
reading file metadata. This is much more efficient from IO perspective.
When broadcasting shreds, turbine excludes the slot leader from the
random shuffle. Doing so, shreds should never loopback to the leader.
If shreds reaching retransmit stage are from the node's own leader slots
they should not be retransmited to any nodes.
* First draft of ingesting duplicate proofs in Gossip into blockstore.
* Add more unittests.
* Add more unittests for bad cases.
* Fix lint errors for tests.
* More linter fixes for tests.
* Lint fixes
* Rename get_entries, move location of comment.
* Some renaming changes and comment fixes.
* Fix compile warning, this enum is not used.
* Fix lint errors.
* Slow down cleanup because this could potentially be expensive.
* Forgot to reset cleanup count.
* Add protection against attackers when constructing chunk map when
we ingest Gossip proofs.
* Use duplicate shred index instead of get_entries.
* Rename ClusterInfoDuplicateShredListener and fix a few small problems.
* Use into_shreds to piece together the proof.
* Remove redundant code.
* Address a few small errors.
* Discard slots too advanced in the future.
* - Use oldest proof for each pubkey
- Limit number of pubkeys in each slot to 100
* Disable duplicate shred handling for now.
* Revert "Disable duplicate shred handling for now."
This reverts commit c3fcf403876cfbf90afe4d2265a826f21a5e24ab.
* Increase turbine propagation const
Value is used as a delay threshold for issuing shred repairs and analysis is showing we are overly aggressive in requesting repairs. Shreds show up via turbine before the repair completes the vast majority of the time
* Use Duration type for MAX_TURBINE_PROPAGATION
Store non-vote transaction counts that are now recorded by the banks
into the `blockstore`.
`SamplePerformanceService` now populates `PerfSampleV2` with counts from
the banks.
Problem
The plugins need to know when all transactions for a block have been all notified to serve getBlock request correctly. As block and transaction notifications are sent asynchronously to each other it will be difficult.
Summary of Changes
Include the executed transaction count in block notification which can be used to check if all transactions have been notified.
The commit adds an associated SignedData type to Shred trait so that
merkle and legacy shreds can return different types for signed_data
method.
This would allow legacy shreds to point to a section of the shred
payload, whereas merkle shreds would compute and return the merkle root.
Ultimately this would allow to remove the merkle root from the shreds
binary.
The commit allocates 2% of slots to running experiments with different
turbine fanouts based on the slot number.
The experiment is feature gated with an additional feature to disable
the experiment.
* Plumb dumps from replay_stage to repair
When dumping a slot from replay_stage as a result of duplicate or
ancestor hashes, properly update repair subtrees to keep weighting and
forks view accurate.
* add test
* pr comments
* Support bi-directional quic communication, use the same endpoint for the quic server and client
This is needed for supporting using quic for repair
* Added comments on the bi-directional communication tests
* Removed some debug logs
* clippy issue
The num_repair field is only blockstore insertion metric being updated
outside of Blockstore::insert() call chain; move the update to insert()
with the rest of the fields in BlockstoreInsertionMetrics struct.
* Add dump_node to update stake for heaviest subtrees
Additionally refactor subtrees to store children as a hashset
* Add a more complicated forks test
* chose -> choose
* remove is_dumped flag and reuse latest_invalid_ancestor instead
* Update cost model to use requested_cu instead of estimated cu #27608
* remove CostUpdate and CostModel from replay/tvu
* revive cost update service to send cost tracker stats
* CostModel is now static
* remove unused package
Co-authored-by: Tao Zhu <tao@solana.com>
* Move ConnectionCache back to solana-client, and duplicate ThinClient, TpuClient there
* Dedupe thin_client modules
* Dedupe tpu_client modules
* Move TpuClient to TpuConnectionCache
* Move ThinClient to TpuConnectionCache
* Move TpuConnection and quic/udp trait implementations back to solana-client
* Remove enum_dispatch from solana-tpu-client
* Move udp-client to its own crate
* Move quic-client to its own crate
In the quic server handle_connection, when we timed out in receiving the chunks, we loop forever to wait for the chunk. If the client never provide another chunk, the server can hopelessly wait for that chunk and wasting server resources. Instead WAIT_FOR_CHUNK_TIMEOUT_MS is introduced to bound this to 10 seconds at maximum. The stream will be dropped if it times out.
The manual Blockstore compaction that was being initiated from
LedgerCleanupService has been disabled for quite some time in favor of
several optimizations.
Co-authored-by: Ryo Onodera <ryoqun@gmail.com>
* Split out voting and banking threads in banking stage
Additionally this allows us to aggressively prune the buffer for voting threads
as with the new vote state only the latest vote from each validator is
necessary.
* Update local cluster test to use new Vote ix
* Encapsulate transaction storage filtering better
* Address pr comments
* Commit cargo lock change
* clippy
* Remove unsafe impls
* pr comments
* compute_sanitized_transaction -> build_sanitized_transaction
* &Arc -> Arc
* Move test
* Refactor metrics enums
* clippy
https://github.com/solana-labs/solana/pull/27193
added hash domain to ping-pong protocol.
For backward compatibility responses both with and without domain were
generated and accepted.
Now that all clusters are upgraded, this commit enforces the hash domain
by removing the response without the domain.
We have transactions counted in replay-slot-end-to-end-stats, but that
metric is broken down to report things per thread.
So, report total_transactions for the entire slot (all threads) in
replay-slot-stats.
#### Summary of Changes
Removes the constant default for ShredStorageType::RocksFifo
as the shred storage size is either user-specified or derived
from --limit-ledger-size in #27459.
- Batch filtering invalid transactions (fail to sanitize, too old or already processed) before forwarding
- Combine packet filtering and forwarding to share sanitized transactions
- `iter_desc` is no longer needed, remove it;
- Add a method to share the logic of removing packets from buffer after they were removed from MinMaxHeap
- Add test coverage for forward_packet_batches_by_accounts
- rebase, resolve conflicts
Separate storage for voting and transaction threads:
- Voting threads utilize a shared reference in order to dedup extraneous
votes
- Transactions have thread local storage like before
* Add structure to collect and coalesce vote packets
Will be used in banking stage to throw out extraneous vote packets
before processing
* pr comments
* Update inner lock to arc to improve performance
--tpu-enable-udp is introduced. And when this is on, the transaction receive and transaction forward is enabled using udp.
Except for a few tests which was hard-coded sending transactions using udp, most tests are being done with udp based tpu disabled.
* Plumb priority_fee_cache into rpc
* Add PrioritizationFeeCache api
* Add getRecentPrioritizationFees rpc endpoint
* Use MAX_TX_ACCOUNT_LOCKS to limit input keys
* Remove unused cache apis
* Map fee data by slot, and make rpc account inputs optional
* Add priority_fee_cache to rpc test framework, and add test
* Add endpoint to jsonrpc docs
* Update docs/src/developing/clients/jsonrpc-api.md
* Update docs/src/developing/clients/jsonrpc-api.md
In kin-sim, we found that bounded channel causes halt for account
background services. As the number of accounts grows, the time for
pruning and cleaning increases, which would leads to longer intervals
between the pruning of deaded bank slots. With 1.7B accounts, we will
exceed the 10K bounded channel threshold that causes halt of account
back ground services. Without pruning, the node will eventually run out
of memory.
* Check overflow on vote tx compaction boundary
Check for overflow during the conversion between VoteStateUpdate and
CompactVoteStateUpdate.
* Try removing clippy supress
Tenets:
1. Limit thread names to 15 characters
2. Prefix all Solana-controlled threads with "sol"
3. Use Camel case. It's more character dense than Snake or Kebab case
* Create a new function cleanup_accounts_paths, a trivial change
* Remove account files asynchronously
* Update and simplify the implementation after the validator test runs.
* Fixes after testing on the dev device
* Discard tokio. Use thread instead
* Fix comments format
* Fix config type to pass the github test
* Fix failed tests. Handle the case of non-existing path
* Final cleanup, addressing the review comments
Avoided OsString.
Made the function more generic with "impl AsRef<Path>"
Co-authored-by: Jeff Washington <jeff.washington@solana.com>
In order to maintain backward compatibility, for now the responding node
will hash the token both with and without domain so that the other node
will accept the response regardless of its upgrade status.
Once the cluster has upgraded to the new code, we will remove the legacy
domain = false case.
A change included in
https://github.com/solana-labs/solana/pull/20480
was that when the root node in turbine broadcast tree is down, the
leader will broadcast the shred to all nodes in the first layer.
The intention was to mitigate the impact of dead nodes on shreds
propagation, because if the root node is down, then the entire cluster
will miss out the shred.
On the other hand, if x% of stake is down, this will cause 200*x% + 1
packets/shreds ratio at the broadcast stage which might contribute to
line-rate saturation and packet drop.
To avoid this bandwidth saturation issue, this commit reverts that logic
and always broadcasts shreds from the leader only to the root node.
As before we rely on erasure codes to recover shreds lost due to staked
nodes being offline.
Given the 32:32 erasure recovery schema, current implementation requires
exactly 32 data shreds to generate coding shreds for the batch (except
for the final erasure batch in each slot).
As a result, when serializing ledger entries to data shreds, if the
number of data shreds is not a multiple of 32, the coding shreds for the
last batch cannot be generated until there are more data shreds to
complete the batch to 32 data shreds. This adds latency in generating
and broadcasting coding shreds.
In addition, with Merkle variants for shreds, data shreds cannot be
signed and broadcasted until coding shreds are also generated. As a
result *both* code and data shreds will be delayed before broadcast if
we still require exactly 32 data shreds for each batch.
This commit instead always generates and broadcast coding shreds as soon
as there any number of data shreds available. When serializing entries
to shreds:
* if the number of resulting data shreds is less than 32, then more
coding shreds will be generated so that the resulting erasure batch
has the same recovery probabilities as a 32:32 batch.
* if the number of data shreds is more than 32, then the data shreds are
split uniformly into erasure batches with _at least_ 32 data shreds in
each batch. Each erasure batch will have the same number of code and
data shreds.
For example:
* If there are 19 data shreds, 27 coding shreds are generated. The
resulting 19(data):27(code) erasure batch has the same recovery
probabilities as a 32:32 batch.
* If there are 107 data shreds, they are split into 3 batches of 36:36,
36:36 and 35:35 data:code shreds each.
A consequence of this change is that code and data shreds indices will
no longer align as there will be more coding shreds than data shreds
(not only in the last batch in each slot but also in the intermediate
ones);
This change sets the receive_window for non-staked node to 1 * PACKET_DATA_SIZE, and maps the staked nodes's connection's receive_window between 1.2 * PACKET_DATA_SIZE to 10 * PACKET_DATA_SIZE based on the stakes.
The changes is based on Quinn library change to support per connection receive_window tweak at the server side. quinn-rs/quinn#1393
#### Problem
LedgerCleanupService requires compactions to propagate & digest range-delete tombstones
to eventually reclaim disk space.
#### Summary of Changes
This PR makes LedgerCleanupService::cleanup_ledger delete any file whose slot-range is
older than the lowest_cleanup_slot. This allows us to reclaim disk space more often with
fewer IOps. Experimental results on mainnet validators show that the PR can effectively
reduce 33% to 40% ledger disk size.
* Keypair: implement clone()
This was not implemented upstream in ed25519-dalek to force everyone to
think twice before creating another copy of a potentially sensitive
private key in memory.
See https://github.com/dalek-cryptography/ed25519-dalek/issues/76
However, there are now 9 instances of
Keypair::from_bytes(&keypair.to_bytes())
in the solana codebase and it would be preferable to have a function.
In particular since this also comes up when writing programs and can
cause users to either start messing with lifetimes or discover the
from_bytes() workaround themselves.
This patch opts to not implement the Clone trait. This avoids automatic
use in order to preserve some of the original "let developers think
twice about this" intention.
* Use Keypair::clone
Prior to this change, long running commands like `solana-ledger-tool
verify` would OOM due to AccountsDb cleanup not happening.
Co-authored-by: Michael Vines <mvines@gmail.com>
* Concurrent replay slots
* Split out concurrent and single bank replay paths
* Sub function processing of replay results for readability
* Add feature switch for concurrent replay
* Only take the latest vote for each validator in gossip
Since the new vote updates are no longer incremental, there
is no value in storing intermediate votes.
* Address pr feedback
* Handle potential downgrade path, FullTowerVote -> Incremental
* Rename sent to bank -> gossip slot
* Handle downgrade case properly
* Only downgrade for newer votes and feature flag, ignore incremental votes otherwise
* Update test
* Use client certs in QUIC to get peer's stake
* fixes to cert processing
* integrate the code
* clippy
* more cleanup
* sort cargo deps
* test fixes
* info -> debug
Shreds have different workload and traffic pattern from TPU vote and
transaction packets. Some of recent changes to SigVerifyStage are not
suitable or at least optimal for shreds sig-verify; e.g. random discard,
dedup with false positives, discard excess by IP-address, ...
SigVerifier trait is meant to abstract out the distinctions between the
two pipelines, but in practice it has led to more verbose and convoluted
code.
This commit discards SigVerifier implementation for shreds sig-verify
and instead provides a standalone stage for verifying shreds signatures.
With recent patches, window-service recv-window does not do much other
than redirecting packets/shreds to downstream channels.
The commit removes window-service recv-window and instead sends
packets/shreds directly from sigverify to retransmit-stage and
window-service insert thread.
- Forward packets by prioritization in desc order
- Add support of cost-tracking by transaction requested compute units
- Hook up account buckets to forwarder
- Add metrics for forwardable batches count
- Remove redundant invalid packets filtering at end of slot since forwarder will do the same when batch forwardable packets
- Add bench test for forwarding
* Remove UseQuic type
Move to storing the UdpSocket on ConnectionCache and accepting a bool
* Remove use_quic from ConnectionCache constructor
Replace with separate with_udp constructor to force callers to choose
Fully deserializing shreds in window-service before sending them to
retransmit stage adds latency to shreds propagation.
This commit instead channels through the payload and relies on only
partial deserialization of a few required fields: slot, shred-index,
shred-type.
* allow initial hash calc to occur in bg
* validator_initialized -> startup_verification_complete
* add infos for leader and vote
* rework snapshot for startup verification
* change to assert
slot_stats are submitted at a different cadence from the rest of
RetransmitStats. Current code erroneously clears slot_stats before
submitting any metrics.
Shred slot and parent are not verified until window-service where
resources are already wasted to sig-verify and deserialize shreds.
This commit moves above verification to earlier in the pipeline in fetch
stage.
Shreds are dropped in window-service if the slot leader is the node
itself:
https://github.com/solana-labs/solana/blob/cd2878acf/core/src/window_service.rs#L181-L185
However this is done after wasting resources verifying signature on
these shreds, and requires a redundant 2nd lookup of the slot leader.
This commit instead discards such shreds in sigverify stage where we
already know the leader for the slot.
Following commits will skip shreds deserializaton before retransmit, and
so we will only have a ShredId and not a fully deserialized shred to
obtain the shuffling seed from.
* Make sure to root local slots even with hard fork
* Address review comments
* Cleanup a bit
* Further clean up
* Further clean up a bit
* Add comment
* Tweak hard fork reconciliation code placement
In order to preserve current behavior, the threshold is set to the
current value of the argument to IndexedParallelIterator::with_min_len.
Follow up commits will recalibrate this threshold to optimize
performance on mainnet-beta.
Shreds arriving at a node for retransmit tend to belong to the same slot
(or a just a couple of different slots). Slot leader and cluster nodes
are common for the shreds of the same slot, and so the common work to
look up these values can be factored out.
This commit first group-bys shreds by slot to factor out that common
lookup work.
* Define shuffle to prep using same shuffle for multiple slices
* Determine transaction indexes and plumb to execute_batch
* Pair transaction_index with transaction in TransactionStatusService
* Add new ReplicaTransactionInfoVersion
* Plumb transaction_indexes through BankingStage
* Prepare BankingStage to receive transaction indexes from PohRecorder
* Determine transaction indexes in PohRecorder; add field to WorkingBank
* Add PohRecorder::record unit test
* Only pass starting_transaction_index around PohRecorder
* Add helper structs to simplify test DashMap
* Pass entry and starting-index into process_entries_with_callback together
* Add tx-index checks to test_rebatch_transactions
* Revert shuffle definition and use zip/unzip
* Only zip/unzip if randomize
* Add confirm_slot_entries test
* Review nits
* Add type alias to make sender docs more clear
In prepration of
https://github.com/solana-labs/solana/pull/25807
which reworks erasure batch sizes, this commit:
* adds a helper function mapping the number of data shreds to the
erasure batch size.
* adds ProcessShredsStats to Shredder::entries_to_shreds in order to
replace and remove entries_to_data_shreds from the public interface.
Shred versions are not verified until window-service where resources are
already wasted to sig-verify and deserialize shreds.
The commit verifies shred-version earlier in the pipeline in fetch stage.
Jeff Biseda
Ashwin Sekar
Andrew Fitzgerald
steviez
behzad nouri
Illia Bobyr
Wen
Pankaj Garg
Lijun Wang
Xiang Zhu
dependabot[bot]
Brooks
Will Hickey
carllin
joeaba
Kirill Fomichev
Ryo Onodera
Yihau Chen
Brennan Watt
Kevin Ji
Trent Nelson
Jeff Washington (jwash)
Brian Anderson
Tao Zhu
HaoranYi
Jason Davis
Jon Cinque
Maximilian Schneider
Michael Vines
Tyera
Dmitri Makarov
sakridge
haoran
Justin Starry
Yueh-Hsuan Chiang
Christian Kamm
apfitzge
Tyera Eulberg
Jeff Washington (jwash)
janlegner
Richard Patel
Boqin Qin(秦 伯钦)
github-actions[bot]
Jack May
Nicholas Clarke
Nick Rempel
Steven Luscher