- Add TradeConfig::with_swqos_cores_from_end(bool) to use last N CPU cores for SWQOS, reducing contention with main thread and default tokio workers.
- Add recommended_sender_thread_core_indices(swqos_count) to get the same last-N core indices for with_dedicated_sender_threads (recommended combo for lower latency).
- Document core affinity and latency in async_executor and with_dedicated_sender_threads.
Made-with: Cursor
- Rename wait_transaction_confirmed to wait_tx_confirmed (params, lib, docs, examples)
- SDK timing: use [SDK][provider] style with fixed-width labels; add newline before block
- Move print_sdk_timing_block to common::sdk_log; unify SWQOS_LABEL_WIDTH
- SWQOS submitted/failed: aligned [Provider] labels via log_swqos_submitted/submission_failed
- Extract short error message from JSON (message/data) or quoted string; prefix with 'error: '
- Format elapsed as 'X.XXXX ms' or 'X.XXXX µs' (4 decimals, space before unit); use comma before error
- Add SwqosType::as_str() for zero-allocation log labels; only parse JSON when input starts with '{'
Made-with: Cursor
- Bump version to 3.6.4 in Cargo.toml
- Update version references in README.md and README_CN.md
- Add RELEASE_v3.6.4.md with English release notes
- All workspace examples verified to compile (cargo build --workspace)
- Remove RELEASE_v3.6.3.md and docs/ASYNC_EXECUTOR_REVIEW.md
- Minor updates in types, lib, transaction_builder, async_executor, executor, params
Made-with: Cursor
- transaction_pool: add PARALLEL_SENDER_COUNT (18), ensure prefill >= 18 so
multi-channel build never serializes; prefill 64 with max(PREFILL, 18)
- async_executor: document that pool prefill must match sender thread count
- executor: do not sort submit_timings (avoids any extra work); comment that
log order is completion order
- docs: add ASYNC_EXECUTOR_REVIEW.md
- remove RELEASE_NOTES_v3.6.2.md; minor example/types/lib/params updates
Made-with: Cursor
The 32 worker tasks in swqos_worker_loop use yield_now().await when
the queue is empty, which busy-spins across all tokio threads. Replace
with tokio::sync::Notify so workers sleep until jobs are enqueued.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pump/PumpSwap buy: track_volume OptionBool = Some(true) only when is_cashback_coin, else Some(false)
- PumpSwap sell cashback: use get_user_volume_accumulator_quote_ata(quote_mint) instead of WSOL-only ATA
- Sync idl/pump_amm.json, idl/pump_fees.json from pump-public-docs
Made-with: Cursor
Replace full eager preallocation in the SWQOS serializer with a bounded\nprewarm and lazy growth strategy.\n\n- Prewarm only a small hot set of buffers instead of allocating all\n 10,000 upfront\n- Keep the same max pool capacity and buffer size to preserve the\n original buffer-pool design in steady state\n- Stop zero-filling each buffer during initialization; reserve capacity\n and grow data on demand\n- Return temporary buffers in Base64Encoder::serialize_and_encode to\n avoid pool depletion\n- Add tests for bounded prewarm and lazy allocation/return behavior\n\nWhy:\nThe previous Lazy initialization allocated and touched ~2.44 GiB\n(10,000 x 256 KiB) on first use, which can block the first buy submit\npath and produce a large cold-start latency spike.\n\nValidation:\n- cargo fmt --all\n- cargo check --lib\n- cargo check (SolBot workspace with path dependency)
- Log build_instructions / before_submit once; per-channel submit, confirmed, total (all from grpc_recv_us)
- confirmed = per-channel submit-to-confirm duration; total = batch start-to-confirm when need_confirm
- No logging between execute_parallel and poll_any_transaction_confirmation to reduce latency
- Clone submit_timings only when log_enabled; all logging after result is ready
- Use provider name (Jito, Helius, etc.) in log lines instead of generic SWQOS
- Timing format: .4 ms; simulate (dry-run) label for simulate mode
- execute_parallel returns 4-tuple with submit_timings for per-channel timing
Made-with: Cursor
- Add Helius Sender client (helius.rs) with dual routing and swqos_only mode
- Helius min tip: 0.0002 SOL default, 0.000005 SOL when swqos_only=true
- Add TradeConfig.check_min_tip (default false) to skip min-tip validation for lower latency
- Thread check_min_tip through SwapParams and execute_parallel; only call min_tip_sol when enabled
- Add SWQOS_ENDPOINTS_HELIUS and HELIUS_TIP_ACCOUNTS in constants
- Extend SwqosConfig/SwqosType for Helius; add Helius to get_endpoint and get_swqos_client
- Update trading_client, shared_infrastructure and middleware_system examples
Made-with: Cursor
- Add pumpfun_copy_trading and pumpfun_sniper_trading to workspace members
- Add grpc_recv_us field to all TradeBuyParams/TradeSellParams in examples
- Switch address_lookup and nonce_cache to sol-parser-sdk, use event is_cashback_coin
- Remove invalid rustls call from pumpfun examples for successful build
- All 18 examples pass cargo check --workspace
Made-with: Cursor
- Bump version to 3.5.4; update README and README_CN
- Astralane: irisb binary API (no Base64), POST /irisb with query api-key&method; constants use /irisb endpoints; ping POST getHealth
- BlockRazor: Send Transaction v2 (plain Base64 body, Content-Type text/plain, auth in URI only); ping POST /v2/health; use default_http_client_builder
- SWQOS: all clients use default_http_client_builder (nextblock, temporal, zeroslot, astralane, lightspeed, jito, flashblock); remove unused Duration imports
Made-with: Cursor
In v3.5.0, confirmation was split out of execute_parallel into
poll_transaction_confirmation, but it only polled sig[0]. When
multi-channel submit is used, each channel produces a different
signature and only one lands on-chain. If the landed sig is not
sig[0], the poll times out after 15s and reports failure despite
a successful on-chain transaction.
Add poll_any_transaction_confirmation() that passes all signatures
to get_signature_statuses in a single RPC call per poll, returning
the first one that confirms. The executor now uses this instead of
polling a single signature.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Make serialize_transaction_and_encode sync and route through buffer pool
(serialization::serialize_transaction_sync) to cut allocs on hot path
- Build single-submit body with format! instead of json!+to_string() in bloxroute
- Add serialize_transactions_batch_sync; bloxroute batch uses it and format! for entries
- Fix first-submit cold start: immediate first ping, then 30s keepalive interval
- Fix high latency after ~5 min: pool_max_idle_per_host=4, pool_idle_timeout=300s
(BlockRazor, Temporal, Node1, Astralane, Stellium) so submit reuses ping connection
- Ping: 1.5s timeout, consume response body so connection returns to pool
- poll_transaction_confirmation: avoid cloning status.value[0], use ref
- Translate all swqos comments to English
Co-authored-by: Cursor <cursoragent@cursor.com>
- PumpFunParams/BondingCurve: from_trade/from_dev_trade take is_cashback_coin parameter
- pumpfun_copy_trading and pumpfun_sniper_trading use sol-parser-sdk for gRPC, pass e.is_cashback_coin
- address_lookup/nonce_cache call sites updated; README EN/CN Cashback and example notes
Co-authored-by: Cursor <cursoragent@cursor.com>
Previously Soyas returned early on successful send without:
- Any logging
- Polling for transaction confirmation
Now aligned with other SWQOS clients (zeroslot, lightspeed, etc.):
- Proper error handling with reconnect retry
- Always poll for transaction confirmation
- Log confirmation success/failure with elapsed time
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Change use_exact_sol_amount from bool to Option<bool>
- Default to true (via unwrap_or) for PumpFun and PumpSwap DEXes
- Option is ignored for other DEXes (Raydium, Bonk, Meteora)
- Update all examples to use None (which defaults to true for applicable DEXes)
- Update README documentation
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Pass through the new field from TradeBuyParams to SwapParams so users
can opt-in to using buy_exact_sol_in and buy_exact_quote_in instructions.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add BUY_EXACT_SOL_IN_DISCRIMINATOR for PumpFun
- Add BUY_EXACT_QUOTE_IN_DISCRIMINATOR for PumpSwap
- Add use_exact_in_instruction field to SwapParams
- When enabled, uses exact input instructions that spend exactly the
specified SOL/quote amount with slippage applied to output tokens
- Defaults to false for backwards compatibility
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Introduce TradingInfrastructure to allow multiple wallets to share the same
RPC client and SWQOS clients, reducing initialization overhead and memory usage.
Changes:
- Add InfrastructureConfig struct for wallet-independent configuration
- Add TradingInfrastructure struct to hold shared RPC/SWQOS clients
- Refactor TradingClient to use Arc<TradingInfrastructure>
- Add from_infrastructure() for fast client creation from shared infrastructure
- Add from_infrastructure_with_wsol_setup() for async WSOL ATA setup
- Maintain backwards compatibility with existing TradingClient::new() API
Benefits:
- First wallet: Full initialization (~200-500ms)
- Subsequent wallets (same config): Fast creation (~1-2ms)
- Memory: Single set of RPC/SWQOS clients shared across wallets
- Add SWQOS_BLACKLIST constant for disabling specific providers
- Add SwqosConfig::is_blacklisted() method to check blacklist status
- Skip blacklisted providers during client initialization with warning
- Remove NextBlock references from README, README_CN and examples
- NextBlock is blacklisted by default, can be enabled by removing from SWQOS_BLACKLIST
Change the default BloxRoute endpoint from germany.solana.dex.blxrbdn.com
to global.solana.dex.blxrbdn.com for better routing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When using multiple SWQOS channels with durable nonce, if one channel's
transaction lands on-chain but fails (e.g., ExceededSlippage), the nonce
is consumed and other channels cannot succeed.
Previously, the SDK would wait for all channels to timeout (~15s) before
returning, and would often return a generic "timeout" error instead of
the actual on-chain error.
This fix:
- Adds `landed_on_chain` field to TaskResult to track if tx landed
- Adds `is_landed_error()` to detect on-chain failures vs timeouts
- Adds `landed_failed_flag` to ResultCollector for early exit signaling
- Returns immediately with the actual error when a tx lands but fails
This makes behavior consistent with the success case - both return
immediately once the nonce is consumed.
Benefits:
1. Get the actual error (e.g., ExceededSlippage) instead of timeout
2. Return in ~1s instead of ~15s
3. Enable faster retry logic for callers
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Fixed a critical bug in find_by_base_mint and find_by_quote_mint functions
where pool data decoding failed because the 8-byte discriminator was not
being skipped. This caused all PumpSwapParams::from_mint_by_rpc calls to
fail with "No valid pool decoded" error.
Changes:
- Modified find_by_base_mint to skip discriminator: pool_decode(&acc.data[8..])
- Modified find_by_quote_mint to skip discriminator: pool_decode(&acc.data[8..])
- Added data length check to prevent out-of-bounds access
- Aligned with fetch_pool function which correctly uses &account.data[8..]
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Fixed incorrect get() index from total_size + 1 to proper instruction index
- Use enumerate() to get correct instruction index for prefetch optimization
- Prevents array bounds errors and ensures correct next instruction prefetch
Bug: Using total_size (byte count) as array index for instruction lookup
Solution: Use enumerate() iterator index for proper instruction access
Wood