HTND v2.18.3 [OPTIONAL]
This release eliminates pruning point movement network stall, improves RPC security, memory management on low RAM systems, and removes some incorrectly handled edge cases that could have resulted in node crashes from malformed inputs.
Incompatible behaviour
RPC now binds to loopback only by default. You MUST use —rpcallowip (see below) to continue to permit non loopback RPC connections.
Summary of Changes
- Huge reduction in network pause at pruning: The pruning-point move runs off the block-processing path, the network no longer stalls for 20-80 seconds and pruning point (not supported with LevelDB - PebbleDB feature only)
- Improved resource handling: Memory and Pebble cache defaults better account for available system resources on low memory footprints, while respecting an explicitly configured GOGC. Effective storage settings are logged. 6GiB is currently the minimum RAM a mainnet node can be realistically run on. 8GiB recommended to support utxoindex, and 16GiB for busy nodes.
- Control over RPC access: RPC now binds to loopback only by default. Node Operators can authorise RPC clients by IP CIDR - use
—rpcallowip=192.168.1.0/24,219.88.72.13(to allow that RFC1918 C class network and hoosatworld). If you rely on remote RPC connections you MUST specify —rpcallowip to open up the wildcard listener. - Network and consensus fixes: Improved handling of relayed blocks and peer connections, limited oversized peer rejection messages, and corrected edge cases involving header timestamps, transaction scripts, and empty Merkle roots.
- Safer shutdown and storage: Fixed several database-close and iterator error cases, improved shutdown coordination, and ensured data is flushed when closing without a write-ahead log.
- Fix macOS arm64 build to use deterministic HoohashV110
- Two cases are removed which could results in a node "lonely/solo mining" and running of in it's own fork which could only be recovered by a clean resync.
- P2P Peers no longer isolate themelsevs under transcation load exceeding 5K messages/route.
- IBD peers are no longer automatically disconnected when perceived as slow, when the slowness could be originating from our own node
Rollback procedure
The new pruning journal and history worker run on every network, mainnet included, with no switch (docs/pruning-worker.md). This changes the on disk format of the DAG.
To go back to 2.18.2: stop the node, run the 2.18.3 binary with the same flags plus --prepare-downgrade, wait for a clean exit, then start 2.18.2.
NB Archival history cannot be re-synced from peers. Archival operators should keep a datadir copy before upgrading
or downgrading.