[rmcdaniel]

Part of #325. A native node starting on an empty MySQL or MariaDB database no longer refuses while another node is still bootstrapping it.

Found by

A multi-node smoke run (#549) started three native nodes together on an empty MySQL database. One node exited with 409 unsupported_rust_schema.

Cause

• MySQL commits DDL on its own.
• One node created dw_server_schema. Before it recorded its rust-initializing marker, another node's read-only preflight ran. That preflight runs before the bootstrap lock, saw an empty marker table and refused.
• A node killed in the same window left a database that no node could start on.
• PostgreSQL bootstraps inside one locked transaction and SQLite inside one file transaction, so neither has this window.

Change

• Preflight now accepts an empty marker table when it is the only table, and the node then waits for the bootstrap lock.
• Under the lock, a missing marker is recorded before the migration runs. This also recovers a node killed between creating the table and recording the marker.

Checks

• The real-kill qualification gains a marker-table phase. It kills a bootstrapping child right after the child creates the marker table, while it holds the bootstrap lock.
• A concurrent start must still be waiting one second later, and must then finish once the child is gone.
• Without the preflight change, the concurrent start refused at once.
• All 15 ignored MySQL qualification tests pass locally on MySQL 8.0 and MariaDB 10.11, on this branch merged with main after Fingerprint a stopped PHP source database for backup proof #523.

🤖 Generated with Claude Code

[rmcdaniel]

Merged main after #566 (755a68d). The port-log conflict was resolved by keeping both entries. Cancelled the superseded run 39538868737 for the previous head.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud