Saved Unfunded Channels Can Crash Eclair on Restart

Nodes running Eclair v0.14.0 and earlier can crash repeatedly on restart when unfunded channel requests saved to disk exhaust memory as they are reloaded at startup. ACINQ, the Eclair developer, issued fixes in July and a broader security release in September to address related defects.

The flaw let a remote peer open many channel requests that were recorded locally but never funded on the Bitcoin blockchain. Inconsistent checks of temporary and final channel identifiers caused Eclair to undercount pending channels and retain large numbers of unfunded channel records. Those stored records consume both memory and database space even though the attacker does not need to broadcast a funding transaction or spend BTC on-chain.

Researcher Erick Cestari demonstrated the persistent-crash scenario in a local regtest environment. In his test, a node running Eclair v0.14.0 exhausted a 4 GB Java virtual machine heap after about 47 minutes and 43 seconds while accumulating roughly 217,623 rows in the channel database. The initial crash left the bogus channel records on disk; when the node restarted, it reloaded the same records and exhausted memory again. Cestari described increasing the Java heap or manually removing the fake channel entries as ways to restore service.

ACINQ merged a patch on July 17 that strengthened duplicate-channel checks and released Eclair v0.14.1 on July 29 to address the denial-of-service weaknesses in the channel handling. A separate issue disclosed as LNF-2026-0003 involved a channel-opening race that could spawn orphaned channel processes consuming CPU or memory; that race condition recovered on disconnect or restart without persistent database buildup. Eclair v0.14.3, released Sept. 14, fixed additional fund-loss vulnerabilities, and ACINQ recommends upgrading to v0.14.3 for the most complete set of fixes.

The reported flaws affect node availability rather than causing direct on-chain loss of funds. The demonstration does not document live exploitation or the number of publicly reachable nodes that remained unpatched. Operators running Eclair v0.14.0 or earlier should upgrade to the latest release, check Java heap settings and channel databases, and be prepared to increase heap limits or clean stored channel records if a node cannot start normally.