Conditionally run migrations per-player

[cckelly]

This may not be the best forum to ask a question on implementation but I couldn't find a better place so I apologize in advance if it's misguided.

I'm looking at using umzug for lazily running migrations once per player. I want to be able to add migrations, and when a player connects to my server, I want to run a migration for them if we haven't already. My initial thought is we need to be able to dynamically modify the params that get passed to the list of migrations as a function that returns the connected user's id and then run the migrations against that (injecting into WHERE clauses), but how can we mark the migration as having ran for that player and never run them again? Do I need a custom storage method to store this info in a separate db table? Any guidance on how to implement this with umzug would be extremely helpful.

[mmkal]

Interesting use case! Yes, I'd think you'll need a custom storage. They're not difficult to write. I don't know what your stack is, but here's a working example I threw together with pg-mem as a pseudo-db. You can swap it out with whatever you want though:

const umzug = require("umzug@3.0.0-beta.15")
const pgMem = require('pg-mem')

const db = pgMem.newDb()

db.public.query('create table migrations(player_id text, name text)')

class PlayerStorage {
  constructor(playerId) {
    this.playerId = playerId
  }
  
  async executed() {
    const result = db.public.query(`
      select name from migrations where player_id = '${this.playerId}'`
    )
    return result.rows.map(r => r.name)
  }
  
  async logMigration({name}) {
    db.public.query(
        `insert into migrations (player_id, name) values ('${this.playerId}', '${name}')`
    )
  }
  
  async unlogMigration({name}) {
    db.public.query(
        `delete from migrations where player_id = '${this.playerId}' and name = '${name}'`
    )
  }
}

const getMigrator = playerId => {
    return new umzug.Umzug({
        migrations: [
            {
                name: 'm1',
                up: async ({ context }) => console.log('m1 running for ' + context.playerId),
                down: async () => {}
            },
            {
                name: 'm2',
                up: async ({ context }) => console.log('m2 running for ' + context.playerId),
                down: async () => {}
            },
        ],
        context: { playerId }, // you could pass `db` in here, assuming your migrations interact with the db
        storage: new PlayerStorage(playerId),
        logger: console,
    })
}

(async () => {
    await getMigrator('player1').up()
    console.log('____')
    await getMigrator('player1').up()
    console.log('____')
    await getMigrator('player2').up()
})()

If you run this (say by pasting into https://npm.runkit.com/umzug, where I tried it out), you'll see that no migrations are run for the second time await getMigrator('player1').up() is called.

---

It's hard to give more guidance without knowing more about the context. But maybe the important pieces of the example above are:

• Creating a custom storage
• Passing in whatever data is necessary via context: { ... } in the Umzug constructor params
• Creating a new Umzug instance on the fly is fine. There's no particular need to have a singleton migrator instance

[cckelly]

Neat! This is exactly what I was looking for. I'm using postgres actually, so this is fairly close to what I need as it is.

[cckelly]

Is there any particular reason we can't use 2.x if the beta isn't production ready yet?

[cckelly]

@mmkal going a small step further here, is it possible to conditionally return migrations that the migrator should run?