Since the update is done semi-automatically by the means of AWS RDS, does AWS takes any responsibility for that? Unless this use case is warned against in migration notes, I would think they ought to.
I think it happened right after the alter table, but it was discovered after the upgrade. It's normal to have more eyes on the system after a DB upgrade, and also common to blame the DB for post-upgrade problems.
Turns out this time the DB was to blame, but not because of the upgrade.
If the alter table in the blog post is not a simplified version of what was executed (barring changing column names, of course), that means the table had no primary key before the migration, which is a problem on its own.
To be honest, I don't even know if there's a safe way out of that situation in a replication setup, but one plan I would have tried to test in that situation is:
- switch binlog_format to ROW (and never look back ...)
- run a noop alter table to rebuild the table and hope that with ROW format, the rows get inserted in the same order (hope really hard please, with feeling)
- run the alter table
Fortunately, recent versions of MySQL have ROW as the default binlog_format.
I agree the root cause here is the lack of a primary key to begin with. But as far as I know, DDL is always replicated as just a statement, regardless of session binlog_format. So I believe the only real fix here is the general approach suggested in the manual [1], i.e. create a new empty table that has the auto_increment PK added and then populate it from the old table.
Per the article, the replicas are configured the same. MIXED is just dynamically selecting the format automatically, and it seems to be the case that AUTOINCREMENT has fundamentally broken semantics under replication.
And instead of erroring out when replicating, it instead chooses ROW format and plays a game of complete nonsense.
The user error is in not sufficiently reading the docs, but it seems to me MySQL is going out of its way to wrap the noose
No, the binlog format is actually irrelevant here; the post author was just incorrect about that part. ALTER TABLE always gets replicated as a statement, regardless of the binlog format.
Auto_increment only has broken replication semantics in the very specific situation the author encountered: a table has data, but no primary key (and also no alternative unique index which could serve as the clustered index key) and then an attempt is made to alter the table to add an auto_increment primary key.
Basically that form of ALTER TABLE statement is telling the database to add a sequential ID to each existing row, but without providing any deterministic way that those numbers should be assigned. So each replica may choose a different numbering, causing the problem experienced by the author.
It's a foot gun, but not a common one in production at any real scale where you'd have a replica in the first place. InnoDB tables really should always have primary keys, and there are various ways to ensure that happens (sql_require_primary_key option, generated invisible primary key option, external linters, etc).
Author here. I didn't make it clear in the article but the table in question did have a primary key, which was migrated to an unique key, and then the new auto-incremental primary key was added.
As I mentioned in the article, the old ID field was then used in the update statement for the other tables.
I'm updating the article now to make that part clear.
Agreed, and SQL is as important to fully understand, as when, for example, writing C. An inept usage of it, can lead to complete and total disaster.
You wouldn't want a junior to write (unreviewed) an internet connected daemon. Or to try to meet a complex RFC spec. Yet SQL? Why not?!
I think one of the greatest disservices people have done, is to abstract away SQL in frameworks. It certainly lets juniors more safely work with databases, but it really has reduced the general SQL knowledge out there. I see many senior programmers, with almost no exposure. Never touched it.
That's kinda fun (not in a fun way).
Since the update is done semi-automatically by the means of AWS RDS, does AWS takes any responsibility for that? Unless this use case is warned against in migration notes, I would think they ought to.
I don't understand why this issue only appeared after the upgrade, not immediately after the migration to add the column?
Versions are different during the upgrade. That's how upgrade is done.
You spin up replica, upgrade it to the target engine version, ensure it's up to date and switch over. [1]
[1] https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/USER_...
I think it happened right after the alter table, but it was discovered after the upgrade. It's normal to have more eyes on the system after a DB upgrade, and also common to blame the DB for post-upgrade problems.
Turns out this time the DB was to blame, but not because of the upgrade.
If the alter table in the blog post is not a simplified version of what was executed (barring changing column names, of course), that means the table had no primary key before the migration, which is a problem on its own.
To be honest, I don't even know if there's a safe way out of that situation in a replication setup, but one plan I would have tried to test in that situation is: - switch binlog_format to ROW (and never look back ...) - run a noop alter table to rebuild the table and hope that with ROW format, the rows get inserted in the same order (hope really hard please, with feeling) - run the alter table
Fortunately, recent versions of MySQL have ROW as the default binlog_format.
I agree the root cause here is the lack of a primary key to begin with. But as far as I know, DDL is always replicated as just a statement, regardless of session binlog_format. So I believe the only real fix here is the general approach suggested in the manual [1], i.e. create a new empty table that has the auto_increment PK added and then populate it from the old table.
[1] https://dev.mysql.com/doc/refman/9.7/en/replication-features...
Mixing binlog formats across replicas sounds like user error to me.
Per the article, the replicas are configured the same. MIXED is just dynamically selecting the format automatically, and it seems to be the case that AUTOINCREMENT has fundamentally broken semantics under replication.
And instead of erroring out when replicating, it instead chooses ROW format and plays a game of complete nonsense.
The user error is in not sufficiently reading the docs, but it seems to me MySQL is going out of its way to wrap the noose
No, the binlog format is actually irrelevant here; the post author was just incorrect about that part. ALTER TABLE always gets replicated as a statement, regardless of the binlog format.
Auto_increment only has broken replication semantics in the very specific situation the author encountered: a table has data, but no primary key (and also no alternative unique index which could serve as the clustered index key) and then an attempt is made to alter the table to add an auto_increment primary key.
Basically that form of ALTER TABLE statement is telling the database to add a sequential ID to each existing row, but without providing any deterministic way that those numbers should be assigned. So each replica may choose a different numbering, causing the problem experienced by the author.
It's a foot gun, but not a common one in production at any real scale where you'd have a replica in the first place. InnoDB tables really should always have primary keys, and there are various ways to ensure that happens (sql_require_primary_key option, generated invisible primary key option, external linters, etc).
Author here. I didn't make it clear in the article but the table in question did have a primary key, which was migrated to an unique key, and then the new auto-incremental primary key was added.
As I mentioned in the article, the old ID field was then used in the update statement for the other tables.
I'm updating the article now to make that part clear.
Or the use of AUTO_INCREMENT, a feature which is already a footgun if implemented badly (as it clearly was)
Agreed, and SQL is as important to fully understand, as when, for example, writing C. An inept usage of it, can lead to complete and total disaster.
You wouldn't want a junior to write (unreviewed) an internet connected daemon. Or to try to meet a complex RFC spec. Yet SQL? Why not?!
I think one of the greatest disservices people have done, is to abstract away SQL in frameworks. It certainly lets juniors more safely work with databases, but it really has reduced the general SQL knowledge out there. I see many senior programmers, with almost no exposure. Never touched it.