fix(microvm-run): avoid db.sqlite backup restart loop under concurrent writes

[devinrsmith]

The nix store DB snapshot taken before VM boot used a plain .backup command, which the SQLite docs specify will restart from page 1 whenever the source database is modified by another connection mid-copy. On an actively used host, nix-daemon writes to db.sqlite frequently enough (builds, substitutions, GC, etc.) that the backup could be restarted continuously, observed as the sqlite3 process pegging a CPU core indefinitely — the bottleneck was repeated re-copying, not disk throughput.

Fix by opening an explicit read transaction on the same connection before running .backup. This pins a consistent snapshot up front, so external commits during the backup no longer invalidate already- copied pages and trigger a restart. Also add a 30s busy-timeout as a defensive measure in case another process holds a competing lock at the moment the transaction starts.

Ref: https://sqlite.org/forum/info/cca839708d74a20014f7188b86a19b267602d497bfa90ec1d1e79111a5b24adb

Summary by CodeRabbit

• Bug Fixes
  • Improved VM startup reliability when copying the host Nix store database.
  • Reduced backup failures by taking a consistent snapshot before the copy begins.
  • Added a longer wait period and a quick consistency check to help prevent intermittent restore issues.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 68d259af-fcf0-47f1-b24e-81e2a826e639

📥 Commits

Reviewing files that changed from the base of the PR and between 9caa9b4 and 1bb4576.

📒 Files selected for processing (1)

• flake.nix

---

📝 Walkthrough

Walkthrough

Modifies flake.nix's VM runner script to wrap the SQLite Nix store database backup in an explicit read transaction with a timeout and consistency check, replacing the previous single-step .backup invocation, with updated explanatory comments.

Changes

SQLite Backup Mitigation

Layer / File(s)	Summary
Transaction-pinned backup sequence flake.nix	Comment updated to explain that pinning a read transaction prevents .backup restarts during concurrent writes; backup command now sets .timeout 30000, wraps .backup in BEGIN;/COMMIT;, and runs SELECT count(*) FROM sqlite_master; beforehand, with the destination path quoted.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main fix: preventing db.sqlite backup restart loops in microvm-run under concurrent writes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[systemstart]

LGTM, thanks for contributing.

[systemstart]

Added in v0.10.0