github stacks-network/stacks-core 4.0.6
Release 4.0.6
on GitHub

54 minutes ago

This release contains bugfixes and improvements to the stacks-node and stacks-signer binaries, ensuring more consistent block production.
This release is compatible with chainstate directories from 4.0.x.

This release is recommended for all Stacks node and signer operators.

This release includes the following changes:
https://github.com/stacks-network/stacks-core/blob/4.0.6/CHANGELOG.md

⚠️ Breaking Changes

  • Unused speculative tries from mining and block-proposal validation are no longer persisted. MARF databases automatically migrate to schema 4 on writable open, removing the legacy mined_blocks table and its stored tries. Older binaries cannot open upgraded databases, including in read-only mode; upgrade chainstate inspection and snapshot tooling alongside the node.
  • Signers in this release rely on the node's new GET /v3/blocks/:block_id/view endpoint and the canonical field on /v3/sortitions to release a signed block that a Bitcoin fork has orphaned. Against an older node a signer stays safe but never releases such a block, which can stall a tenure after a reorg. Upgrade the stacks-node before, or together with, its signer.
  • Removed the stackslib::chainstate::stacks::index module; the MARF now lives in the stacks-marf crate. Downstream library consumers should import from stacks_marf instead. There is no functional change.

Added

  • Adds the build-time environment variable CUSTOM_VERSION_STRING and exposes stacks_build_info through node and signer Prometheus metrics, with standard version, STACKS_NODE_VERSION, Git commit, and custom version labels. Custom metadata remains excluded from public version strings.
  • Added GET /v3/blocks/:block_id/view, reporting the tenure and burn view a block hangs from and whether a burnchain fork has orphaned either, and a canonical field on the sortition info returned by /v3/sortitions. Signers use both to tell whether a conflicting block is dead.
  • Consensus tests can start a tenure with a block of chosen transactions through ConsensusChain::append_tenure_start_block, which keeps invalid transactions and returns the block's result like other test blocks. Test-only change.
  • The core MARF types (MARF, MARFOpenOpts, MarfConnection, MarfTransaction) are now re-exported from the root of the stacks-marf crate, so they can be imported as stacks_marf::MARF. The existing stacks_marf::marf:: paths keep working.
  • New [connection_options] settings bound how much inbound p2p work a node does in one network pass: max_inbound_messages_per_pass (default 64) caps the messages handled from any one peer, and max_inbound_processing_ms (default 500) caps the time spent handling inbound messages before the node moves on to inventory sync, block downloads and StackerDB sync. Work that doesn't fit is carried over to the next pass, which starts immediately unless the node is still processing previously downloaded blocks. While a peer's leftover messages are queued, the node stops reading from that peer's socket, so a burst of messages can't overflow the inbox and drop the connection.
  • With the testing feature, the signer binary reads its test flags from the JSON file named by STACKS_SIGNER_TEST_FLAGS_FILE, so a harness running the signer as a separate process can inject faults.
  • Add a single-contract signer-manager-v2.clar reference with accumulated staker payouts, rejected withdrawal retry, per-cycle reward reserves, payout configuration, allowlisting, and signer metadata. The original signer-manager.clar remains available.
  • The signer now shrinks its database file at startup when at least 256 MiB of it is free space (e.g. after pruning) and its live data is at most 1 GiB, so the space goes back to the OS. Set db_shrink = false in the signer config to disable it.
  • Added node.sqlite_vfs configuration option to select the SQLite VFS used for node databases. "unix-excl" holds one exclusive lock per database for the life of the process and keeps the WAL index in memory, removing per-transaction file locking on network and clustered filesystems. The default ("default") keeps current behavior.

Changed

  • A globally defined constant now wins over a local variable or parameter of the same name. Such a case should not actually exist (it would be prevented earlier). This change merely makes the behavior consistent for the unlikely case that it does happen.
  • Prevent deployment of transient contracts from Epoch 4.1 on. This is a legacy construct that is no longer necessary and that only serves to confuse.
  • From Epoch 4.1, Clarity cost and memory charges, in analysis and at runtime, size callable types, and unions of them, by the contract and trait identifiers they hold. Size limits are unchanged.
  • Clarity docs: the define-read-only description now states that the reserved-name exception applies from Epoch 4.1.
  • Calling a user-defined function, whether directly, through contract-call? or from a transaction, copies its arguments less often and holds fewer copies of them while the call runs.
  • ClarityBackingStore::get_cc_special_cases_handler now takes the called contract, and the node only supplies the contract-call special-case handler for calls into PoX contracts.
  • Once a node's Stacks tip is a Nakamoto block and the first Nakamoto reward cycle has ended, it caches the epoch 2.x block inventory data it uses to answer GetBlocksInv requests, instead of querying the chainstate for every request.
  • The burnchains, config, core, net and util_lib modules now imports the MARF from stacks_marf directly instead of through the stackslib::chainstate::stacks::index re-export. No behaviour change.
  • The chainstate code now imports the MARF from stacks_marf directly instead of through the stackslib::chainstate::stacks::index re-export. No behaviour change.
  • stacks-node, clarity-cli, stacks-inspect and marf-squash now depend on stacks-marf directly instead of reaching the MARF through stackslib's re-export, and stacks-node's testing feature enables stacks-marf/testing explicitly. No behaviour change.
  • The last assert!(matches!(..)) in the MARF's tests now uses assert_matches!, and redundant custom failure messages were dropped since assert_matches! already prints the value. Test-only change.
  • The node's Prometheus metrics endpoint is now served by tiny_http instead of async-std/async-h1, removing those dependencies and http-types. The endpoint's responses are unchanged. The configuration docs for the node's prometheus_bind and the signer's endpoint and metrics_endpoint now warn that these servers do not limit request sizes and should only be reachable from trusted hosts.
  • The Clarity v2 parser now reads tokens by reference instead of copying each token every time it is peeked or consumed, making parser::v2::parse about 10% faster on pox-4. Parse results and diagnostics are unchanged.
  • PoxVersions moved to pox-locking and is now the one list of PoX contracts.
  • Upgraded prometheus from 0.9 to 0.14 with its protobuf feature disabled, removing unused protobuf from monitoring_prom builds of the node and signer. Exported metrics are unchanged.
  • Documented PoX-5 lock semantics for locked and unlock_height in the /v2/accounts RPC spec.
  • Upgraded rusqlite from 0.31 to 0.41 for libsqlite3-sys (now 0.39.1). The bundled SQLite version moves from 3.45.0 to 3.54.1.
  • From Epoch 4.1, sortition draws the winning block-commit and the null miner's challenge from separate, domain-separated digests of the sortition hash and VRF seed, instead of reusing one digest for both.
  • The Clarity sugar expander and definition sorter no longer deep-clone nested expressions once per enclosing list or tuple, so both passes are linear rather than quadratic in nesting depth (the sugar expander pass is about 7x faster on pox-4). The definition sorter also no longer copies the whole contract before sorting. The expanded AST is unchanged.
  • Retain contract deployment interfaces in transaction receipts without retaining the full contract analysis graph.
  • TupleTypeSignature and ListTypeData now cache their value size at construction instead of recomputing it on every call, and TypeSignature::size() reads that cache for tuples and lists rather than walking the type. The cached field is not serialized: it is recomputed on deserialization, so the wire format is unchanged.

Fixed

  • In Clarity 7, variable shadowing (i.e. a local binding with the same name as a top-level definition) is prevented during contract analysis in all cases. Previously, some cases would have been missed, only breaking (or leading to unexpected behavior) at runtime.
  • Signers now release a signed conflict whose burn view a Bitcoin reorg has orphaned. A tenure that extends across a reorged burn block holds two blocks at the same height differing by burn view alone; without this a signer could be barred from that height for the rest of the tenure.
  • Allow Nakamoto blocks with exactly 2 MiB of transaction data in mining and signer proposal validation starting in Epoch 4.1, preserving the exclusive size limit in earlier epochs.
  • Fixed a node syncing from genesis stalling at the Epoch 3.0 handoff: blocks downloaded just before the switch were dropped, and the Nakamoto tenures among them were never fetched again.
  • From Epoch 4.1, a contract can call contract-call? through a trait it defines while it deploys, at the top level. Before, the deploy failed with NoSuchContract (#6831).
  • Fixed the miner's handling of a transaction whose receipt (its result plus its print events) is too large to fit into a block. The execution cost of a transaction the miner rejects is no longer left charged to the block being assembled; the Nakamoto miner drops a transaction whose receipt alone reaches 75% of the per-block receipt limit from the mempool as problematic instead of retrying it on every block; and a transaction whose receipt no longer fits in the room left in the current block is skipped and retried in a later block.
  • Signers no longer pre-commit to a block at a height in a tenure where they have already committed to a different block. A sibling the signer has signed bars the pre-commit permanently; one it has only pre-committed to bars it for pre_commit_exclusivity_timeout_secs (default 30 seconds), or until enough weight rejects that sibling for reasons that will not be reconsidered, whichever comes first. Added signer database schema version 22, which creates the blocks_consensus_hash_height index on blocks (consensus_hash, stacks_height).
  • A signer's own signature over a block never goes stale against a sibling: another block at the same height in the same tenure is refused however much time has passed. Only a burnchain fork releases it, by orphaning the tenure's sortition or the block's burn view. Signatures at other heights or in other tenures are released by staleness, a reorg permit, or (on a node running this release) the same burnchain-fork checks; a signed conflict is no longer released merely because the node's Stacks chain does not reach it.
  • Signers now post a block received in a miner's BlockPushed message to their node only if the block's signer signatures are valid for the signer's reward set.
  • Signers now reload the signer set's state machine updates from StackerDB when their reward cycle starts, so a signer whose node reaches the cycle's first burn block after its peers' nodes no longer rejects every proposal with NoSignerConsensus until the next burn block.
  • The signer now exits with an error naming the problem when it cannot open its database, for example one written by a newer release, instead of panicking.
  • Signers will now attempt to defer proposals received before the global state machine has reached consensus amongst the signer set, rather than immediately rejecting with NoSignerConsensus. This should improve block production at Bitcoin block boundaries.
  • Signers now take a block proposal's burn height and reward cycle from the sortition that elected the block's tenure, and ignore proposals that misstate them, rather than trusting the values the miner sends alongside the block.
  • A block proposal for a tenure that the signer's node has not yet processed is rejected with the transient NoSortitionView reason, so the miner re-proposes it once the signer has caught up.
  • Update signer response processing and database lookups.
  • Signers and miners no longer decode the legacy replay_transactions field of inbound state machine updates, and only decode state machine update slots loaded at startup when they hold a state machine update.

Docker images have been published to GitHub Container Registry:

stacks-core

docker pull ghcr.io/stacks-network/stacks-core:4.0.6@sha256:c92c10b8ad4875655312044439bb0d8dddae415668749b6646384a5df5597a9b
docker pull ghcr.io/stacks-network/stacks-core:4.0.6-alpine@sha256:2e3d22baa1f5c5ee3d7d9bb143323de1744b29fab74ac7f57322c62d6b111a5e

stacks-signer

docker pull ghcr.io/stacks-network/stacks-signer:4.0.6@sha256:49d76711acf7e858b6434114043ecee079aac015392e94f87613358b86dd3dda
docker pull ghcr.io/stacks-network/stacks-signer:4.0.6-alpine@sha256:edfe3bbaa3ac121b86274b4a05d3abbeecd050bda24003a7155e956335be1eed
Check out latest releases or
releases around stacks-network/stacks-core 4.0.6

Don't miss a new stacks-core release

NewReleases is sending notifications on new releases.

Get notifications