SIGTRAP in Obsidian renderer at node::sqlite::UserDefinedFunction::xDestroy during c-ares finalization (request: experimentalNativeSqlite opt-out)

[apptb]

Summary

Claudian v2.0.8 / v2.0.9 triggers repeated Obsidian renderer crashes (Obsidian Helper (Renderer)) on macOS via a node:sqlite finalization race. Three crashes today in a 42-minute window (08:09, 08:29, 08:51 EDT). Patching out the node:sqlite import to force JSONL-direct session reads has fully stopped the crashes (0 in 70+ minutes since patch landed).

Request: please add a setting like experimentalNativeSqlite: false so users can opt out of node:sqlite while a real fix lands.

Environment

• macOS 26.2 (25C56), Apple Silicon
• Obsidian Helper (Renderer) 0.14.8 (Electron-based)
• Claudian: confirmed on v2.0.8 and v2.0.9
• Claude Code CLI: ~/.local/bin/claude (managed install)
• Vault has 299 session JSONL files in ~/.claude/projects/<encoded-vault-path>/

Crash signature

exception: EXC_BREAKPOINT  signal: SIGTRAP  code: 0x0000000000000001
=== Faulting thread 0 ===
  Electron Framework  ares_dns_rr_get_ttl
  Electron Framework  ares_dns_rr_get_ttl
  Electron Framework  ares_dns_rr_get_ttl
  Electron Framework  ares_dns_rr_get_ttl
  Electron Framework  node::sqlite::UserDefinedFunction::xDestroy(void*)
  Electron Framework  v8::PropertyDescriptor::set() const
  Electron Framework  v8::PropertyDescriptor::set() const
  Electron Framework  v8::String::ContainsOnlyOneByte() const
  Electron Framework  v8::Promise::Resolver::GetPromise()
  ...

The c-ares DNS resolver and node:sqlite UserDefinedFunction destructor stack-on each other during V8 GC, leading to a V8 invariant violation. Pattern is consistent across all three crash dumps.

Reproduction

Hard to reduce, since it depends on:

• Open Obsidian + Claudian
• Have a Claude Code session active that performs network I/O (any tool call that hits external services)
• During GC of a Claude Code session that accessed sqlite session storage
• Crash hits when sqlite UDF destructor is invoked while c-ares is mid-callback

In our environment with active Claude Code work it reliably hits within a working session — three crashes in ~40 minutes.

Workaround applied locally

In .obsidian/plugins/claudian/main.js, the runtime check that does import("node:sqlite") is replaced so it falls through to the JSONL-direct read path that already exists:

• Patch byte-exact: count("node:sqlite") = 0 in main.js after patch
• Plugin loads cleanly (obsidian dev:errors reports none)
• 299 sessions still discoverable + readable (verified via parsed JSONL)
• 0 renderer crashes for 70+ minutes since patch
• BRAT auto-update from v2.0.8 → v2.0.9 clobbered the patch within ~2 minutes; the patch had to be re-applied to v2.0.9 and the plugin pinned in BRAT

I've also installed a launchd canary (com.pkm.claudian-canary, runs every 15 min) that greps for node:sqlite and warns if a future plugin update reintroduces it.

Ask

A user-facing toggle to disable node:sqlite-backed session storage and fall back to JSONL-direct reads. Suggested name: experimentalNativeSqlite: false in plugin settings, or an env var like CLAUDIAN_DISABLE_NATIVE_SQLITE=1.

This would let affected users opt out without forking the plugin or maintaining post-update patches.

Happy to provide the three full .ips dumps if useful.

[github-actions]

This issue has been inactive for 7 days and will be closed in 7 days if no further activity occurs.