set annotation_id as primary key in annotation_tag table

[swtch1]

This corrects a failure to migrate/create the annotation_tag table on MariaDB with innodb_force_primary_key=1 which is required for Galera clustering.

setting annotation_id as primary key in table to satisfy innodb_force_primary_key=1

Related to #12971

[CLAassistant]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Joshua Thornton seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[swtch1]

Seeing I'm getting a failure in test:

• /go/src/github.com/grafana/grafana/pkg/services/sqlstore/annotation_test.go
  Line 64:
  Expected: nil
  Actual: 'UNIQUE constraint failed: annotation_tag.annotation_id'

And in Postgres integration:

• /go/src/github.com/grafana/grafana/pkg/services/sqlstore/annotation_test.go
  Line 64:
  Expected: nil
  Actual: 'pq: duplicate key value violates unique constraint "annotation_tag_pkey"'

Not being familiar with all the uses of the database here I'm not exactly sure where concentrate next. Could I get some direction?

[marefr]

annotation_id and tag_id probably needs to be primary key. Not sure if that's supported by Grafana migrator. Not sure this change will be allowed for existing annotation_tag tables which could be the next potential problem.

[torkelo]

Yea, this won’t work, need a new column if a primary key is needed

[swtch1]

If I setup an ID field with autoincrement is that they way to go?

[torkelo]

yes

[marefr]

We still need a migration path for that change though I suppose. Basically rename table, create new table, migrate data and drop old table - think that's how we usually do it and it should be examples of that among our migration code.

[bergquist]

Creating a new table and migrating data won't help if still try to create the table without PK first.

In this case, I think we need to both modify the current migration (won't run if it's already present) and create rename/copy data into a new table with PK to get full consistency.

here is an example of how to create create a new table and migrate the old data https://github.com/grafana/grafana/blob/master/pkg/services/sqlstore/migrations/dashboard_mig.go#L208

[torkelo]

Closing this as it needs new id column

[thestevenbell]

I'm facing this same issue while using MariaDB, fresh install using docker. Grafana v6.0.1 and 6.0.2. Is there are work around?

msg="Connecting to DB" logger=sqlstore dbtype=mysql
msg="Starting DB migration" logger=migrator
msg="Executing migration" logger=migrator id="Create 
msg="Executing migration failed" logger=migrator error="Error 1173: This table type requires a primary 
msg="Exec failed" logger=migrator error="Error 1173: ey" sql="CREATE TABLE IF NOT EXISTS `annotation_tag` LL\n, `tag_id` BIGINT(20) NOT NULL\n) ENGINE=InnoDB mb4_unicode_ci;"
msg="Server shutdown" logger=server reason="Service rror 1173: This table type requires a primary key"