Bug Report: Mysqld.GetSchema() race condition when table is dropped throughout inspection

[shlomi-noach]

Overview of the Issue

Mysqld.GetSchema() is susceptible to two race conditions. One leading to a Table X doesn't exist (errno 1146) error, and the other to silent missing PK information.

The race conditions both happen when a table is dropped after the initial call to collectBasicData():

vitess/go/vt/mysqlctl/schema.go

Line 90 in e8dc004

tds, err := mysqld.collectBasicTableData(ctx, dbName, request.Tables, request.ExcludeTables, request.IncludeViews)

1st race condition: a table is dropped right after this query, and before iterating the table for collectSchema()/.GetColumns():

vitess/go/vt/mysqlctl/schema.go

Line 110 in e8dc004

fields, columns, schema, err := mysqld.collectSchema(ctx, dbName, td.Name, td.Type, request.TableSchemaOnly)

in this scenario, we get a Table X doesn't exist (errno 1146), because we actively SELECT from the table that is now dropped.

2nd race condition: a table is dropped right before analyzing primary keys:

vitess/go/vt/mysqlctl/schema.go

Line 131 in e8dc004

colMap, err = mysqld.getPrimaryKeyColumns(ctx, dbName, tableNames...)

In this scenario, there's no error here, because we get PK information from information_schema. But we will have missing data in the table definition (nil value for PrimaryKeyColumns).

Either way, it's valid for a table to be dropped during this time. The fix is to identify that a table has been dropped, and silently ignore such table.

---

This affects anything that attempts to ReloadSchema, which means any VReplication or Online DDL operations.

Reproduction Steps

Binary Version

-

Operating System and Environment details

-

Log Fragments

No response

[shlomi-noach]

The 2nd race condition holds true, but identifying it is a bit more complex. The fact that a table does not appear in INFORMATION_SCHEMA.STATISTICS does not mean the table does not exist - it may exist and have no indexes.