[v2.3] Automatically use optimal Block Engine region #923

esemeniuc · 2025-07-22T02:37:30Z

This pull request introduces several changes aimed at improving the handling of configuration by adding support for dynamic Block Engine endpoint discovery. Whenever the connection with the block engine is broken, we automatically try to connect to the best available region.

Changes:

Implemented dynamic Block Engine endpoint discovery using latency-based selection via a new ping utility
Introduced a disable_block_engine_autoconfig flag in BlockEngineConfig to allow users to bypass automatic endpoint discovery
Replaced RwLock with ArcSwap for shred_receiver_address to avoid acquiring a lock per sent shred batch

Port of #920, but with autoconfig default change to be reverted in near future

core/src/proxy/block_engine_stage.rs

ebin-mathews · 2025-07-23T20:14:24Z

core/src/proxy/block_engine_stage.rs

+        let mut endpoint_discovery = BlockEngineValidatorClient::connect(backend_endpoint.clone())
+            .await
+            .map_err(|e| ProxyError::BlockEngineConnectionError(e.to_string()))?;
+        let endpoints = endpoint_discovery


Curious: Do we really need an "initial" block engine connection to get this info? Shouldn't this be decoupled from block engine itself so that the validator simply fetches this info from an external server and connects to the optimal block engine?

Are we using block engine as a server here for ease of use or is there any other reason this is coupled to block engine?

Trying to see why we even need to provide an initial block engine url here if the idea is to do a self-discovery to connect to the most optimal block engine

@esemeniuc @buffalu @jedleggett

i think its best to avoid fetching from external services for 2 reasons:

you need to connect to the block engine in the first place. don't want to have more complex setup than necessary

validator service to should always be working

how will you get the new non block engine endpoint to the validator? i don't think hard coding it is a better solution

validator/src/cli.rs

only reroute if relayer connected (#123) feat: add client tls config (#121) remove extra val (#129) fix clippy (#130) copy all binaries to docker-output (#131) Ledger tool halts at slot passed to create-snapshot (#118) update program submodule (#133) quick fix for tips and clearing old bundles (#135) update submodule to new program (#136) Improve stake-meta-generator usability (#134) pinning submodule head (#140) Use BundleAccountLocker when handling tip txs (#147) Add metrics for relayer + block engine proxy (#149) Build claim-mev in docker (#141) Rework bundle receiving and add metrics (#152) (#154) update submodule + dev files (#158) Deterministically find tip amounts, add meta to stake info, and cleanup pubkey/strings in MEV tips (#159) update jito-programs submodule (#160) Separate MEV tip related workflow (#161) Add block builder fee protos (#162) fix jito programs (#163) update submodule so autosnapshot exits out of ledger tool early (#164) Pipe through block builder fee (#167) pull in new snapshot code (#171) block builder bug (#172) Pull in new slack autosnapshot submodule (#174) sort stake meta json and use int math (#176) add accountsdb conn submod (#169) Update tip distribution parameters (#177) new submodules (#180) Add buildkite link for jito CI (#183) Fixed broken links to repositories (#184) Changed from ssh to https transfer for clone Seg/update submods (#187) fix tests (#190) rm geyser submod (#192) rm dangling geyser references (#193) fix syntax err (#195) use deterministic req ids in batch calls (#199) update jito-programs revert cargo update Cargo lock update with path fix fix cargo update autosnapshot with block lookback (#201) [JIT-460] When claiming mev tips, skip accounts that won't have min rent exempt amount after claiming (#203) Add logging for sol balance desired (#205) * add logging * add logging * update msg * tweak vars update submodule (#204) use efficient data structures when calling batch_simulate_bundles (#206) [JIT-504] Add low balance check in uploading merkle roots (#209) add config to simulate on top of working bank (#211) rm frozen bank check simulate_bundle rpc bugfixes (#214) rm frozen bank check in simulate_bundle rpc method [JIT-519] Store ClaimStatus address in merkle-root-json (#210) * add files * switch to include bump update submodule (#217) add amount filter (#218) update autosnapshot (#222) Print TX error in Bundles (#223) add new args to support single relayer and block-engine endpoints (#224) point to new jito-programs submod and invoke updated init tda instruction (#228) fix clippy errors (#230) fix validator start scripts (#232) Point README to gitbook (#237) use packaged cargo bin to build (#239) Add validator identity pubkey to StakeMeta (#226) The vote account associated with a validator is not a permanent link, so log the validator identity as well. bugfix: conditionally compile with debug flags (#240) Seg/tip distributor master (#242) * validate tree nodes * fix unit tests * pr feedback * bump jito-programs submod Simplify bootstrapping (#241) * startup without precompile * update spacing * use release mode * spacing fix validation rm validation skip Account for block builder fee when generating excess tip balance (#247) Improve docker caching delay constructing claim mev txs (#253) fix stake meta tests from bb fee (#254) fix tests Buffer bundles that exceed cost model (#225) * buffer bundles that exceed cost model clear qos failed bundles buffer if not leader soon (#260) update Cargo.lock to correct solana versions in jito-programs submodule (#265) fix simulate_bundle client and better error handling (#267) update submod (#272) Preallocate Bundle Cost (#238) fix Dockerfile (#278) Fix Tests (#279) Fix Tests (#281) * fix tests update jito-programs submod (#282) add reclaim rent workflow (#283) update jito-programs submod fix clippy errs rm wrong assertion and swap out file write fn call (#292) Remove security.md (#293) demote frequent relayer_stage-stream_error to warn (#275) account for case where TDA exists but not allocated (#295) implement better retries for tip-distributor workflows (#297) limit number of concurrent rpc calls (#298) Discard Empty Packet Batches (#299) Identity Hotswap (#290) small fixes (#305) Set backend config from admin rpc (#304) Admin Shred Receiver Change (#306) Seg/rm bundle UUID (#309) Fix github workflow to recursively clone (#327) Add recursive checkout for downstream-project-spl.yaml (#341) Use cluster info functions for tpu (#345) Use git rev-parse for git sha Remove blacklisted tx from message_hash_to_transaction (#374) Updates bootstrap and start scripts needed for local dev. (#384) Remove Deprecated Cli Args (#387) Master Rebase improve simulate_bundle errors and response (#404) derive Clone on accountoverrides (#416) Add upsert to AccountOverrides (#419) update jito-programs (#430) [JIT-1661] Faster Autosnapshot (#436) Reverts simulate_transaction result calls to upstream (#446) Don't unlock accounts in TransactionBatches used during simulation (#449) first pass at wiring up jito-plugin (#428) [JIT-1713] Fix bundle's blockspace preallocation (#489) [JIT-1708] Fix TOC TOU condition for relayer and block engine config (#491) [JIT-1710] - Optimize Bundle Consumer Checks (#490) Add Blockhash Metrics to Bundle Committer (#500) add priority fee ix to mev-claim (#520) Update Autosnapshot (#548) Run MEV claims + reclaiming rent-exempt amounts in parallel. (#582) Update CI (#584) - Add recursive submodule checkouts. - Re-add solana-secondary step Add more release fixes (#585) Fix more release urls (#588) [JIT-1812] Fix blocking mutexs (#495) [JIT-1711] Compare the unprocessed transaction storage BundleStorage against a constant instead of VecDeque::capacity() (#587) Automatically rebase Jito-Solana on a periodic basis. Send message on slack during any failures or success. Fix periodic rebase #594 Fixes the following bugs in the periodic rebase: Sends multiple messages on failure instead of one Cancels entire job if one branch fails Ignore buildkite curl errors for rebasing and try to keep curling until job times out (#597) Sleep longer waiting for buildkite to start (#598) correctly initialize account overrides (#595) Fix: Ensure set contact info to UDP port instead of QUIC (#603) Add fast replay branch to daily rebase (#607) take a snapshot of all bundle accounts before sim (#13) (#615) update jito-programs submodule Add 2.0 to daily rebase (#626) Export agave binaries during docker build (#627) Buffer bundles that exceed processing time and make the allowed processing time longer (#611) Publish releases to S3 and GCS (#633) Rebase from different repos (#637) Point SECURITY.md to immunefi (#671) Loosen requirements on tip accounts touchable in BankingStage (#683) Separate out broadcast + retransmit shredstream (#703) Add packet flag for staked node (#705) Add auto-rebase to v2.1 (#739) Fix release github (#745) Move block_cost_limit tracking to BankingStage in preparation for SIMD-0207 (#753) Add precompile checks in BundleStage (#787) Add auto-rebase to v2.2 (#818) Add better error handling around missing transaction signatures for bundle id generation (#860) Remove unwrap from authentication (#861) Revert Jito-Solana WorkingBankEntry changes (#873) BP #885: Add libclang to Dockerfile (#886) Remove the tip distributor code (#888) Rebase: Update anchor to not use deprecated crates Backport #952 to v2.3: Remove trusted relayer packets (#953) Backport #954 to v2.3: Fix shred retransmit (#955)

esemeniuc requested review from 0xEdgar, buffalu, jedleggett and segfaultdoc July 22, 2025 02:37

esemeniuc force-pushed the eric/discovery-service-2-3 branch from e5cc9bf to fcc0c6c Compare July 22, 2025 02:43

esemeniuc changed the title ~~Automatically use optimal Block Engine region~~ [v2.3] Automatically use optimal Block Engine region Jul 22, 2025

esemeniuc marked this pull request as ready for review July 22, 2025 02:59

esemeniuc force-pushed the eric/discovery-service-2-3 branch from fcc0c6c to e185835 Compare July 22, 2025 03:03

jedleggett approved these changes Jul 22, 2025

View reviewed changes