Failed to initialize memory_tree schema: disk I/O error: Error code 1546: Error attempting to truncate file

[ElioNeto]

Original issue tinyhumansai#2410 by @tiny-sentry[bot] on 2026-05-21T04:38:42Z

---

Sentry Issue: TAURI-RUST-EV

Failed to initialize memory_tree schema: disk I/O error: Error code 1546: Error attempting to truncate file

[ElioNeto]

Comment by @sunilkumarvalmiki on 2026-05-21T08:08:36Z

I took a first pass at reproducing this on current main (c204a53, crate version 0.54.4) from a clean Windows worktree.

Findings:

• Current src/openhuman/memory/tree/store.rs already treats SQLite extended code 1546 (SQLITE_IOERR_TRUNCATE) as a transient cold-start condition, together with 14 (CANTOPEN) and 4874 (IOERR_SHMMAP).
• with_connection now serializes first-time schema initialization per DB path and caches the initialized connection, so concurrent workers should not all race through PRAGMA journal_mode=WAL / schema setup on a cold workspace.
• The jobs worker also classifies 1546 as transient and backs off instead of reporting it as a hard failure.

Verification I ran:

cargo test --manifest-path Cargo.toml --lib with_connection_serialises_concurrent_schema_init -- --nocapture
cargo test --manifest-path Cargo.toml --lib is_transient_cold_start_classifies_known_extended_codes -- --nocapture
cargo test --manifest-path Cargo.toml --lib is_sqlite_io_transient_matches_ioerr_truncate -- --nocapture
$env:OPENHUMAN_WORKSPACE = "<fresh temp workspace>"
$env:RUST_LOG = "debug"
.\target\debug\memory-tree-init-smoke.exe 64
.\target\debug\memory-tree-init-smoke.exe 64

Result:

• All three focused tests passed.
• Fresh cold-start smoke: 64/64 ok, 0 failed.
• Warm re-run smoke: 64/64 ok, 0 failed.

I could not reproduce the schema-init failure on current main. This may already be covered by the current connection-cache / transient-I/O handling, or the Sentry event may be from an older build or a production filesystem state that the local smoke does not recreate. If this is still firing after c204a53 / 0.54.4, the next useful data point would be the affected app version/build and whether the event happens during first launch, after a crash/restart, or on an existing memory_tree/chunks.db with stale -wal / -shm side files.

[ElioNeto]

Comment by @sanil-23 on 2026-05-21T19:51:07Z

Fixed by tinyhumansai#2455 (fix(memory): run memory_tree on TRUNCATE journal instead of WAL, merged 2026-05-21).

The root cause is that the memory_tree SQLite store (chunks.db) was the only tree DB running on a WAL journal. WAL's -wal/-shm side-files fail at cold-start init under concurrent worker bootstrap — surfacing on Windows as SQLITE_IOERR_TRUNCATE (code 1546, "Error attempting to truncate file"), which is exactly this report.

tinyhumansai#2455 switches the store to the TRUNCATE rollback journal (with synchronous=FULL), removing the -wal/-shm files entirely; existing WAL databases migrate in place on first open with committed data preserved.

Note: the fix is on main but not yet in a tagged release (current is v0.54.0), so it will reach users with the next release cut. Closing as resolved by tinyhumansai#2455.