fix: correct FSI before throwing "Not enough space" to break poisoned-cache loop

[Copilot]

The in-memory Free Space Index (_fsi) is updated immediately (non-transactionally) when DeleteCore compacts a page. A rolled-back delete reverts the physical page via WAL but leaves the FSI with an inflated free-space value. Every subsequent insert is routed to that stale page, hits the physical capacity limit, and throws — without ever correcting the FSI — creating an unrecoverable loop.

Changes

• DocumentCollection.InsertIntoPage — call _fsi.Update(pageId, actualFreeSpace) before throwing "Not enough space". The error message now captures the stale FSI value (pre-correction) for diagnostics:

  _fsi.TryGetFreeBytes(pageId, out var staleFsi);
  _fsi.Update(pageId, freeSpace);                  // break the loop
  throw new InvalidOperationException(
      $"Not enough space: need {requiredSpace}, have {freeSpace} | ... | FSI={staleFsi}");

• DynamicCollection.InsertIntoPage — same fix.

• Regression test InsertAsync_SucceedsAfterRollbackInducedFsiDesync — populates pages with large documents, induces desync via delete + rollback, then asserts that a subsequent insert succeeds within two attempts (first attempt corrects the FSI; second allocates a fresh page).

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• av-build-tel-api-v1.avaloniaui.net
  • Triggering command: /usr/share/dotnet/dotnet dotnet exec --runtimeconfig /home/REDACTED/.nuget/packages/avalonia.buildservices/11.3.2/tools/netstandard2.0/runtimeconfig.json /home/REDACTED/.nuget/packages/avalonia.buildservices/11.3.2/tools/netstandard2.0/Avalonia.BuildServices.Collector.dll (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[Copilot]

Pull request overview

Addresses a rollback-induced Free Space Index (FSI) desynchronization where a stale (overly optimistic) in-memory FSI entry can cause repeated inserts to be routed to an actually-full page, resulting in a “Not enough space” failure loop.

Changes:

• Update FSI with the page’s actual free space before throwing “Not enough space” in DocumentCollection.InsertIntoPage.
• Apply the same “correct FSI before throw” behavior in DynamicCollection.InsertIntoPage.
• Add a regression test that reproduces the rollback-induced FSI desync and asserts inserts succeed within two attempts.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
tests/BLite.Tests/InsertBulkTests.cs	Adds regression test reproducing the poisoned-cache loop scenario.
src/BLite.Core/DynamicCollection.cs	Updates FSI before throwing on insufficient page space to break the loop.
src/BLite.Core/Collections/DocumentCollection.cs	Captures stale FSI value for diagnostics and corrects FSI before throwing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This fix was applied to DynamicCollection.InsertIntoPage as well, but the regression test added in this PR only exercises the typed DocumentCollection path. Since DynamicCollection has its own test suite (e.g., tests/BLite.Tests/DynamicCollectionTests.cs), consider adding a corresponding rollback-induced FSI desync test for DynamicCollection to prevent regressions in that codepath too.

[Copilot]

staleFsi will be 0 when TryGetFreeBytes returns false, which changes the diagnostic from the previous FSI=n/a to a misleading FSI=0. Consider preserving the prior behavior by capturing the bool return and formatting FSI as n/a when the page isn't tracked in the FSI yet.

Suggested change

	_fsi.TryGetFreeBytes(pageId, out var staleFsi);
	_fsi.Update(pageId, freeSpace);
	throw new InvalidOperationException($"Not enough space: need {requiredSpace}, have {freeSpace} | PageId={pageId} | SlotCount={header.SlotCount} | Start={header.FreeSpaceStart} | End={header.FreeSpaceEnd} | FSI={staleFsi}");
	var hasStaleFsi = _fsi.TryGetFreeBytes(pageId, out var staleFsi);
	var fsiDiagnostic = hasStaleFsi ? staleFsi.ToString() : "n/a";
	_fsi.Update(pageId, freeSpace);
	throw new InvalidOperationException($"Not enough space: need {requiredSpace}, have {freeSpace} | PageId={pageId} | SlotCount={header.SlotCount} | Start={header.FreeSpaceStart} | End={header.FreeSpaceEnd} | FSI={fsiDiagnostic}");

[Copilot]

The exception filter uses ex.Message.StartsWith("Not enough space") without a StringComparison, which makes the match culture-sensitive and a bit brittle. Use the StartsWith(string, StringComparison.Ordinal) overload (or another non-message-based discriminator if available) to keep the test stable across environments.

Suggested change

	catch (InvalidOperationException ex) when (ex.Message.StartsWith("Not enough space"))
	catch (InvalidOperationException ex) when (ex.Message.StartsWith("Not enough space", System.StringComparison.Ordinal))