Auto-invalidate or auto-update cached relations

[bajtos]

Consider the following code snippet from loopback-workspace:

// Project hasMany Models

project.models(function(err, list) { // (1)
  // print all models
  var modelName = // let the user choose one model
  addPermission(modelName, saveToFiles);
}

function addPermission(modelName, cb) { 
  // extract from project.addPermission()
  project.models( // (2)
    {where: { name: modelName }, limit: 1 },
    function(err, list) {
      list[0].options.acls.push({/* ACL config */});
      list[0].save(cb); // (3)
  });
}

function saveToFiles(cb) {
  // extract from project.saveToFiles
  project.models(function(err, list) { // (4)
    // build models.json from list
    // etc.
    cb();
  });
}

The line (1) fills "project.models" cache with the list of all related models. The line (2) queries the related models and returns a new copy of the model, which is saved back to the datasource in (3). Finally line (4) lists all models again, but receives stale data.

Since the modification was made through project.models, it should be possible to improve the caching mechanism, so that either:

1. The query on related models returns existing objects from the cache, so that any updates are automatically propagated to the cache too.
2. The save function of models returned by the query updates the cached record.
3. The save function of models returned by the query invalidates the whole relation cache and forces the next "find-all" query to fetch fresh data from the datasource.

Workaround

Always pass 'true' as the first parameter of "find-all" queries. In other words, changing the line (4) to project.models(true, function(err, list) { fixes the problem.

Although the workaround is trivial, tracking the problem down to the related-models-cache is not straightforward.

[ritch]

The relations cache seems to be pretty useless or entirely broken. It looks like it is never invalidated. You have to grab a new instance to get up to date data.

I'd like to get rid of it in favor of all DAO methods hitting the connector API every time. I'd like to move the optimization to the connector level instead of the model instance level. A connector wide cache would improve all hits to a dataSource not just model relations.

[bajtos]

I am closing this issue, since no user was asking for this enhancement and it clearly was not a priority for us so far. We can reopen it again later when there is more demand.

[kidwm]

Hey, I just encountered the same problem, and turn to delay the models listing query for 1 second as a workaround.

[superkhau]

I am closing this issue, since no user was asking for this enhancement and it clearly was not a priority for us so far. We can reopen it again later when there is more demand.

@bajtos I think you forgot to close the issue. Closing.