[sql-33] session: add migration code from kvdb to SQL #1051
Conversation
ViktorTigerstrom
added
the
no-changelog
ViktorTigerstrom
force-pushed
the
2025-04-migrate-sessions
branch
from
449a2cc to
1828ec4
Compare
In preparation for upcoming migration tests from a kvdb to an SQL store, this commit updates the NewTestDB function to return the Store interface rather than a concrete store implementation. This change ensures that migration tests can call NewTestDB under any build tag while receiving a consistent return type.
When a session has a MacaroonRecipe set, which has a nil value set for either the `Permissions` or the `Caveats` field, the kvdb session store would return that field as nil, while the sql prior to this commit would return an empty array. When the we migrate the session store from kvdb to sql, that will cause the kvdb and the migrate session store to differ when comparing them, which will cause the migration to fail. This commit fixes that by ensuring that the sql session store returns a nil value for those fields in that scenario, rather than an empty array.
This commit introduces the migration logic for transitioning the sessions store from kvdb to SQL. Note that as of this commit, the migration is not yet triggered by any production code, i.e. only tests execute the migration logic.
ViktorTigerstrom
force-pushed
the
2025-04-migrate-sessions
branch
from
1828ec4 to
a868a93
Compare
| macRecipe = &MacaroonRecipe{} | ||
| if perms != nil { | ||
| macRecipe.Permissions = unmarshalMacPerms(perms) | ||
| } | ||
| if caveats != nil { | ||
| macRecipe.Caveats = unmarshalMacCaveats(caveats) |
There was a problem hiding this comment.
Let me know if you think we should rather fix the kvdb store implementation rather than the sql one. But as we've previously opted for not updating the store behaviour, and make the sql one mimic the behaviour until we've completed the migration, I opted for this version here.
| if !session.RevokedAt.IsZero() { | ||
| err = tx.SetSessionRevokedAt( | ||
| ctx, | ||
| makeSetRevokedAtParams(sqlId, session.RevokedAt), | ||
| ) | ||
| if err != nil { | ||
| return err | ||
| } | ||
| } |
There was a problem hiding this comment.
Could potentially update the insert session sql query to also take an optional RevokedAt sql.NullTime param, to not have to do this as a separate step. I opted for this version to keep things as is as much as possible, but if you think that'd be better, I'll update to do that instead :)
| { | ||
| "randomized sessions", | ||
| randomizedSessions, | ||
| }, |
There was a problem hiding this comment.
Opted for a test that uses the "standard" rand lib here, as @bitromortac had a preference for that in #1047 (comment). But can also look into adding a test that uses rapid if you'd prefer.
| // We always have a caveat.Id, but the rest are | ||
| // randomized if they exist or not. | ||
| macCaveat.Id = randomBytes(rand.Intn(10) + 1) |
There was a problem hiding this comment.
Let me know if this is not true in for real world data, as migrations WILL fail if we don't set a caveat.Id when the caveats param of WithMacaroonRecipe is not nil.
| // "dest" state, without checking if the state transition is legal. | ||
| // | ||
| // NOTE: this function should only be used for testing purposes. | ||
| func (db *BoltStore) shiftStateUnsafe(_ context.Context, id ID, |
There was a problem hiding this comment.
Not fully sure if we add test functions like this one that's a struct member to the _test file, or if we add it in the actual struct file?
This PR introduces the migration logic for transitioning the sessions store from kvdb to SQL.
Note that as of this PR, the migration is not yet triggered by any production code, i.e. only tests execute the migration logic.
Part of #917