VersionLoader doesn't recreate the version table after it is dropped while the loader is alive
Maintainers usually reply within 1 day
@seragAhmed is already working on this.
Since Nov 13, 2026.
Assessment
- Difficulty
- 4/5
- Estimated time
- 1-2 days
- Newbie friendliness
- 54/100
Research direction
Start with src/FluentMigrator.Runner.Core/VersionLoader.cs, especially LoadVersionInfo() and RemoveVersionTable(), and trace how the creation flags interact with preview mode and HasCreatedVersionTable. Reproduce with SQLite using the same runner scope: migrate up, roll back to version 0, then migrate up again. Done means the version table is recreated safely after removal, including when another process dropped it, while preview behavior and creation reporting remain correct.
Written by the indexing model from the issue text.
Description
Summary
VersionLoader remembers that it has created the version table, and never checks that again. If the version table is later dropped while that VersionLoader instance is still alive, the next MigrateUp() through the same runner skips recreating the table. It then fails when it records the applied migration:
SQLite Error 1: 'no such table: VersionInfo'.
Any of these can drop the table:
RollbackToVersion(0)Rollback(n)that reverts every applied migration- another process doing either of those
This has been the behaviour since 2012. It does not depend on the application locking added in #2432, but that review is where it came up (see below).
How to reproduce
The version table must not exist when the runner's VersionLoader is created, so the loader creates it. Then, in the same dependency-injection scope:
// Empty database, so no VersionInfo table yet
using var scope = serviceProvider.CreateScope();
var runner = scope.ServiceProvider.GetRequiredService<IMigrationRunner>();
runner.MigrateUp(); // VersionLoader creates VersionInfo, then migration 1 is applied
runner.RollbackToVersion(0); // migration 1 reverted, VersionInfo dropped
runner.MigrateUp(); // throws: no such table: VersionInfo
It also fails across two processes. Instance A runs MigrateUp() against an empty database and keeps its scope open. Instance B runs RollbackToVersion(0). A's next MigrateUp() fails the same way.
It works when the version table already existed before the runner was resolved, because then the loader never set its "created" flag. Run MigrateUp() once in a separate scope first, and the sequence above succeeds.
Reproduced on main at cd7ae9e3, with SQLite. Nothing in the code is provider-specific.
Why it happens
VersionLoader (src/FluentMigrator.Runner.Core/VersionLoader.cs) has four flags, one per version-table step:
| Flag | Step it guards |
|---|---|
_versionSchemaMigrationAlreadyRun |
create the schema |
_versionMigrationAlreadyRun |
create the table |
_versionUniqueMigrationAlreadyRun |
add the AppliedOn column and unique index |
_versionDescriptionMigrationAlreadyRun |
add the Description column |
LoadVersionInfo() runs a step only when the object is missing in the database and the flag is not set:
if (!AlreadyCreatedVersionTable && !_versionMigrationAlreadyRun)
{
Runner.Up(VersionMigration);
_versionMigrationAlreadyRun = true;
}
RemoveVersionTable() drops the table, and the schema when OwnsSchema is true, but never clears the flags. So once a loader has created the table, it will not create it again, even after the table is gone. LoadVersionInfo() then sees no table, loads an empty version list, and the runner applies migrations. UpdateVersionInfo() then inserts into a table that does not exist.
The runner's VersionLoader lives for the whole scope: it is created lazily and cached by MigrationRunner. Long-lived scopes therefore hit this, for example a hosted service or a test fixture that reuses one runner.
Why the flags can't simply be deleted
The flags do two jobs that a fix must keep:
- Preview mode. With
--preview/AsGlobalPreview(), the processor logs theCREATE TABLEinstead of running it, so the table never exists. Without a flag, every call toLoadVersionInfo()would emit the version-table DDL again. HasCreatedVersionTableandHasCreatedVersionSchema(added for #2170 in #2405) report whether this loader created those objects.MigrationRunner.LoadVersionInfoIfRequired()uses that to tell the caller a table was created. The existence check alone can't answer that, because the table already exists once the constructor returns.
Possible fixes
- Reset the flags in
RemoveVersionTable(). This is the smallest change and fixes the single-process case. It does not fix the two-process case, because instance B'sRemoveVersionTable()cannot reset instance A's flags. - Trust the database when not in preview. Use the flags only to stop repeated DDL in preview mode, and otherwise act on
AlreadyCreated*alone. Keep "this loader created it" as a separate value forHasCreated*. This fixes both cases. The extra existence checks are already run today on everyLoadVersionInfo(), so there is no new cost.
Either way, the check and the create should stay under the application lock from #2432 (IApplicationLocker). The two-process case is a race unless they do.
Is it a blocker?
No. It needs the version table to be dropped while a runner that created it is still in use, which is rare outside tests and teardown scripts. Most applications run MigrateUp() once per scope and exit. Suggested milestone: 10.0.0, alongside the async runner work, which reshapes VersionLoader anyway. It could also go into a 9.x patch release if the reset-in-RemoveVersionTable() fix is enough.
Background
GitHub Copilot raised this while reviewing #2432 (comment). It was deferred there because #2432 adds an opt-in application lock (#1949), and this bug exists with or without that lock. #2432 did fix a related problem: Rollback, RollbackToVersion and MigrateDown now reload the version list before choosing what to revert. That PR did not change how VersionLoader decides whether to recreate the table.
🤖 Generated with Claude Code
- Dominant language
- C#
- Stars
- 3.5k
- Forks
- 702
- Avg merge
- 2d 15h
- Merged PRs (30d)
- 19
Getting set up
Starts the project's dev container in your browser, under your own GitHub account.
- No Dockerfile or Docker Compose file
- No pull request template
- No contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from fluentmigrator/fluentmigrator
-
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
fluentmigrator/fluentmigrator#2460 ·
Maintainers usually reply within 1 day
-
feature
Difficulty 2/5 1-3 hours Newbie friendliness 63/100
fluentmigrator/fluentmigrator#2434 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 84/100
fluentmigrator/fluentmigrator#2316 · 7 comments ·
Maintainers usually reply within 1 day
-
infrastructure:build-pipeline
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
fluentmigrator/fluentmigrator#1853 ·
Maintainers usually reply within 1 day
-
ADR: Create.Check/Alter.Check/Delete.Check top-level syntaxPossibly taken @jzabroski claimed this 2 days ago. Openfeature
fluentmigrator/fluentmigrator#2458 · 1 reaction · 2 assignees ·
Maintainers usually reply within 1 day
All issues in fluentmigrator/fluentmigrator
Similar issues
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
ersinkoc/phonebox#411 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 85/100
-
area: ui bug triage
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
hand-to-AI
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
CactuseSecurity/firewall-orchestrator#5803 ·
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
Maintainers usually reply within 1 day