pos: shrink NodeConfig and secure-storage to what's actually used#3458
Merged
peilun-conflux merged 6 commits intoConflux-Chain:masterfrom Apr 27, 2026
Conversation
…ogger, LeaderReputationConfig) Four never-read pieces of the Diem-era NodeConfig tree: - `MetricsConfig` + `NodeConfig.metrics`: unused since the diem_metrics observability layer was removed in Conflux-Chain#3451; the only remaining reference was a `set_data_dir` propagation that wrote to a field no code reads. - Outer `TestConfig` + `NodeConfig.test`: the struct has a single `#[serde(skip)]` field and is never accessed by the conflux binary. (Not to be confused with `SafetyRulesTestConfig`, which is still used.) - `SafetyRulesConfig.logger`: safety-rules never reads it; the top-level `NodeConfig.logger` is the only logger actually configured. - `LeaderReputationConfig`: struct defined in consensus_config.rs but never referenced from any code path. Note: `NodeConfig` uses `deny_unknown_fields`, so any operator YAML that happens to contain `metrics:` or `test:` at the top level will now fail to parse. The mainnet PoS template (hard_fork_tool.py) and both Python test-framework YAML writers (tests/test_framework/util.py and integration_tests/test_framework/util/__init__.py) do not emit these keys, so no known config is affected. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… SecureBackend
Two changes that reduce the "operator thinks they can configure this" surface
of the PoS YAML:
1. Lift always-overwritten NodeConfig fields to constructor args.
Five fields were populated from the main TOML / network at startup and
immediately written over whatever the YAML said, making them
footguns — operators could set them and silently have them ignored:
- SafetyRulesConfig.test
- SafetyRulesConfig.export_consensus_key (always `true` in production)
- SafetyRulesConfig.vrf_private_key
- SafetyRulesConfig.vrf_proposal_threshold
- ConsensusConfig.chain_id
Replaced with a new PosRuntimeKeys struct threaded through
start_pos_consensus -> setup_pos_environment -> start_consensus ->
EpochManager::new. `create_safety_rules` gains explicit author /
consensus_private_key / vrf_private_key / export_consensus_key args.
Drops the SafetyRulesTestConfig struct (its last user was the lifted
`test` field; `execution_key` was never read).
2. Collapse SecureBackend + drop NamespacedStorage.
SecureBackend was a two-variant enum but the InMemoryStorage variant
was never constructed from config (tests use InMemoryStorage::new()
directly). Replace SafetyRulesConfig.backend with OnDiskStorageConfig
and delete the enum.
NamespacedStorage (the Storage-wrapper implementing on-disk namespacing
for multi-tenant secure_storage.json) was only reachable via
OnDiskStorageConfig.namespace, which defaulted to None and was never set
by any template. Delete the module + the Storage::NamespacedStorage
variant + the .namespace field.
Storage still has an InMemoryStorage variant because test code
instantiates Storage::from(InMemoryStorage::new()); that path is kept.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PosRuntimeKeys bundled per-node identity/secrets (author + BLS + VRF private keys) with chain-wide consensus parameters (chain_id + vrf_proposal_threshold). These have distinct lifecycles and different security properties, and the codebase already separates them elsewhere — ValidatorSigner / ValidatorVerifier / EpochState are the existing Diem idiom for this split. Split into two types threaded through `start_pos_consensus → setup_pos_environment → start_consensus → EpochManager::new`: PosNodeKeys — author, consensus_private_key, vrf_private_key PosChainParams — chain_id, vrf_proposal_threshold EpochManager's internal field layout is left flat; the typed boundary at function signatures is what carries the semantic distinction. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… backend
Before: `Storage` was an enum over `InMemoryStorage` and `OnDiskStorage`,
dispatched via `#[enum_dispatch]`. `InMemoryStorage` existed purely so
tests and benches could skip filesystem I/O. But the perf delta is
immaterial — unit tests do a handful of ops (ms), and the
`benches/safety_rules.rs` in-memory baseline isn't exercised in CI
(`cargo bench --no-run` only). Keeping a second backend to serve use
cases that don't care about the saving was scaffolding.
Delete the enum, the in-memory backend, the `NamespacedStorage` wrapper
(already unused after the SecureBackend collapse earlier in this PR),
the `#[enum_dispatch]` attributes on the `KVStorage`/`CryptoStorage`
traits, and the `enum_dispatch` crate dep. `OnDiskStorage` keeps its
pre-refactor shape: a JSON file on disk that is the source of truth,
read on every `get` and rewritten atomically on every `set`. The hot
per-vote read (`SAFETY_DATA`) is cached one layer up by
`PersistentSafetyStorage::cached_safety_data`, which is unchanged.
Tests and benches switch to `OnDiskStorage::new(TempPath / NamedTempFile)`.
The `benches/safety_rules.rs::InMemory` baseline is removed — with only
one backend, a separate in-memory baseline has nothing to compare
against. `PersistentSafetyStorage::{replace_with_suffix, save_to_suffix}`
lose their match-on-`Storage::OnDiskStorage` boilerplate and the
runtime `"unsupported secure storage type"` error branch that went
with it.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0c4d749 lifted the consensus / VRF private keys out of NodeConfig and stored them bare on EpochManager. But safety_rules, ProposalGenerator, VrfProposer, RoundManager, and the force_* handlers all take owned keys, so the stored form still needs to be cloneable. BLSPrivateKey's Clone impl is behind the diem-crypto `cloneable-private-keys` feature, only activated transitively via pos-types's fuzzing dev-dependency. CI's `cargo clippy --release --all` (no --all-targets) doesn't activate dev-deps, so the feature is off and cfxcore fails with five "no method named `clone` found for struct `BLSPrivateKey`" errors in epoch_manager.rs. Wrap the stored keys in ConfigKey<K> like they were before. ConfigKey's Clone and private_key() round-trip via BCS serde on the inner key, so they don't require K: Clone — the same trick NodeConfig used. Owned-key consumers switch from `.clone()` to `.private_key()`; the RoundManager pass-through becomes a bare `.clone()` on the ConfigKey. Same per-call cost as before the refactor (identical BCS round-trip), and the avoided alternative — turning on cloneable-private-keys in release builds — would silently hand a freely-cloneable private key to anyone in cfxcore, which is exactly what ConfigKey was designed to prevent. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
peilun-conflux
force-pushed
the
cleanup/pos-shrink-nodeconfig-and-storage
branch
from
122779a to
27c3ee6
Compare
ef96de3 removed enum_dispatch from safety-rules and added diem-temppath; the main workspace Cargo.lock was updated in that commit but the tool sub-workspace Cargo.lock files weren't, causing CI's --locked builds in tools/consensus_bench and tools/evm-spec-tester to fail. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pana
approved these changes
Apr 27, 2026
Member
Pana
left a comment
Pana
left a comment
There was a problem hiding this comment.
@Pana reviewed 31 files and all commit messages.
Reviewable status:complete! all files reviewed, all discussions resolved (waiting on ChenxingLi).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Some
NodeConfigfields were silently overwrittenConsensusConfig.chain_idand fourSafetyRulesConfigfields (test,export_consensus_key,vrf_private_key,vrf_proposal_threshold) were read from YAML and immediately replaced with values from the main Conflux TOML / network at startup. Operators could set them and get them silently ignored — a footgun. Those values now thread through as runtime arguments via two new structs (PosNodeKeysfor per-node identity/secrets,PosChainParamsfor chain-wide consensus params), matching the Diem separation ofValidatorSignervsValidatorVerifier/EpochState.Secure-storage collapses to one backend
SecureBackendwas a multi-variant enum inherited from Diem. Vault was deleted in603c1d1d3,NamespacedStorageis unused, andInMemoryStorageexisted only so tests could skip file I/O — butPersistentSafetyStorage::cached_safety_dataalready owns the one hot read (SAFETY_DATAper vote), and the test speedup is immaterial.OnDiskStoragebecomes the sole backend;Storageenum,enum_dispatchdep, the_ => Err("unsupported")branches inreplace_with_suffix/save_to_suffix, and theNamespacedStoragemodule all go away.NodeConfigusesdeny_unknown_fields, so anypos_config.yamlsetting one of the removed fields will now fail to load:metrics,testconsensus.chain_idsafety_rules.{logger, test, export_consensus_key, vrf_private_key, vrf_proposal_threshold}Audited
run/pos_config/pos_config.yamlon both themainnetandtestnetbranches — neither sets any of these fields.This change is