Conversation
…pends still validate ErgoNodeViewHolder's Declined branch discarded the pool returned by ErgoMemPool.process, so the prefilter's rejected-id cache never reached a running node. Install the returned pool when it differs by reference, which is a no-op for every pre-existing declining path. Also: single-pass input resolution at the prefilter anchor, a clearer rejection message, and an end-to-end case covering a conformant owner spend with checkReemissionRules enabled.
Invariant (h) drives the token-preserving rejection through a real ErgoNodeViewHolder, using a test-only subclass whose InjectState message reuses the existing protected updateNodeView to install a synthetic token-bearing state. It asserts the id is cached both in the published ChangedMempool reader and in a freshly queried GetDataFromCurrentView. A companion control pins the reference-inequality guard as a no-op on a declining path that returns the same pool. Removing only the guarded install makes (h) fail while the control still passes.
Send InjectState and the transaction through the same probe to avoid an Akka per-pair FIFO race that could let the tx overtake the state injection onto the missing-UTXO path. Assert the intended decline reason in both cases, and assert minimalFeeAmount == 0 as an explicit precondition, since the fee gate precedes the prefilter. Behaviour unchanged; all four mutation checks (predicate-false, delete-install, unconditional-install, d3 burn) confirmed against these tests.
Test that proofsForTransactions() does not modify the UTXO set tree along a growing chain
Require wallet-owned change addresses
…overy Preserve prepared UTXO snapshot state on restart
Replay the three reviewed commits from PR #2512 on the release branch. Preserve the seven approved file contents and the checkpoint, startup and synchronization observation tests without importing unrelated master changes.
Fix Digest state recovery checkpoint root
Block assembly is bounded by maxBlockSize and maxBlockCost, both enforced
by CandidateGenerator.correctLimits. The mempool API returned `size` but
not `cost`, so external block builders could enforce only the size limit
and had to recompute or guess per-transaction cost, even though the node
already measured it on mempool admission.
Add a `cost` field to the mempool transaction JSON, sourced from
UnconfirmedTransaction.lastCost. It is null when the node never measured
the cost: digest state, or a transaction returned to the pool by a
rollback and not yet re-validated by CleanupWorker.
Affected endpoints:
GET /transactions/unconfirmed
GET /transactions/unconfirmed/byTransactionId/{txId}
POST /transactions/unconfirmed/byErgoTree
byTransactionId used ErgoMemPoolReader.modifierById, which returns
Option[ErgoTransaction] and drops the cost. Add unconfirmedById to the
reader, returning the UnconfirmedTransaction wrapper; modifierById is now
defined in terms of it. byErgoTree kept the wrapper out of its collect
blocks, so those become filters. Deduplication is unchanged:
UnconfirmedTransaction.equals is defined on the transaction id.
The digest-state branch of byErgoTree previously serialized a bare
ErgoTransaction, which did not match the ErgoTransactionWithInputBoxes
schema the spec declares for that response. It now goes through the same
JSON builder, so `cost` is never silently absent.
The field is additive and optional, so clients ignoring unknown keys are
unaffected. Confirmed-transaction schemas are untouched, as lastCost only
exists for unconfirmed transactions.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ting Count pending inbound connections toward the limit
Expose transaction validation cost in mempool transaction JSON
…20260913 Retry cached block sections after remote header processing
Reemission mempool filter now excludes only transactions with reemission token in outputs.
# Conflicts: # src/main/resources/api/openapi.yaml # src/main/resources/application.conf # src/test/scala/org/ergoplatform/nodeView/wallet/ErgoWalletServiceSpec.scala
Discard stale transaction rejections after candidate retry
Retry candidate generation in case of failed proofs
Fixes for nipopows + utxo set snapshot bootstrapping
odiseusme
left a comment
odiseusme
left a comment
There was a problem hiding this comment.
Reviewed the candidate at 2046cfc. Two blockers, both in candidate generation; I found no additional blockers in the production diff, subject to the assessment limits below.
Blocker 1 - the candidate does not compile. CandidateGenerator.scala:762:93: EliminateTransactions(retryEliminate) where retryEliminate is already an EliminateTransactions (found EliminateTransactions, required Seq[ModifierId]). Reproduced locally with sbt compile. The retry-success path at :751 passes it correctly.
Blocker 2 - the change contradicts its own comment. The emission-only fallback is reached only when both proof attempts failed, so the second collection's classifications were never validated against a state that produced a usable candidate. At f002151d2 this branch read EliminateTransactions(Seq.empty) under the comment "Both collections produced failed proofs; their rejections may be stale." At the head the comment is unchanged but the code attempts to propagate retryEliminate; removing only the extra wrapper would retain potentially stale rejections. I'd keep the empty list on the second-failure fallback, and keep the retry's classifications on retry success, as in #2559. @a-shannon reports a rollback regression failing here once the type error is fixed locally; that matches, though I have not run it myself.
Non-blocking, from reading the rest of the diff:
- Mempool JSON (
TransactionsApiRoute.scala:133,321):costis new, and the digest-state ErgoTree query now serializes with resolved inputs so digest mode matches the declaredErgoTransactionWithInputBoxesschema. Both are visible to API consumers and deserve a release note. - Wallet change addresses (
ErgoWalletService.scala:402,ErgoWalletState.scala:85-98): a change address must now be tracked and signable, and updates require an unlocked wallet. Previously persisted external change addresses stop being used and change goes to the root address. That is a behaviour change for existing operators and needs a release note beyond the OpenAPI text. - NiPoPoW (
ErgoNodeViewHolder.scala:320-330): applying a proof now callssetHeadersChainSynced()whenutxoBootstrapis on. This starts snapshot discovery earlier; it does not itself establish a change in the proof-validation trust boundary. - Inbound accounting (
NetworkController.scala:185-198,245-267) looks correct to me: the pending slot covers onlyConnectedtoConfirmConnection, that decision is local, dead sockets are caught byTerminated, and post-confirmation handlers are bounded byhandshakeTimeout.
Not assessed: crash-consistency ordering between snapshot state and history writes, digest replay from the recovered checkpoint, and whether cached or in-flight body sections remain correctly gated during snapshot bootstrap.
|
Both blockers are resolved at Local on The non-blocking items in my earlier comment still stand, in particular release notes for the mempool JSON |
6 participants
Uh oh!
There was an error while loading. Please reload this page.