Observe after access

[cajoy]

Add a new hook called loaded. It should be triggered whenever model data from the database is returned back to the caller (i.e. find but also create or prototype.updateAttributes) and the context should provide only ctx.instance, no ctx.data. For "batch" load like find, the hook should be fired multiple times, once for each model instance loaded.

---

The original issue description

I see there is couple of observers such as "access", "create" and etc.

What I need is "after find".

The reason is: I need to encrypt/decrypt data in database.

So basically before save I can encrypt data with the key.
But I can't decrypt data after find.

I've tried with getter/setter but it doesn't work.

[clark0x]

the getter should work. Have you defined the encrypt property in model's json file?

[cajoy]

I can't find any configuration for encrypt property in models json file.

So one of solutions is to use "before save" to encrypt and "getter" to decrypt... What may work. But

When I post the form, data comes as plain text decrypted. And for example I call validation on this property before saving it will try to decrypt property what is not encrypted what will cause an issue.

Example: Post form to save data

POST secureField - not encrypted
Model validation
"getter" secureField will try to decrypt (what will cause an issue)

Example: Get/Update from database

Model query
secureField is encrypted
"getter" secureField will decrypt the field
save model
observer "before save" will encrypt field
success

This still seems like something that makes sense to have...and not just because every other operation hook has a before/after.

For example, if you merge one instance of a model into the other, then having an 'after access' hook would allow you to put a "redirectId" object in the old instance so that outdated references to the old model can be forwarded to the new model.

Right now this can be worked around in remote hooks by creating a wrapper module for the model, but is a problem when you start dealing with model relations and other things more tightly integrated into Loopback itself.

[bajtos]

The reason is: I need to encrypt/decrypt data in database.
So basically before save I can encrypt data with the key. But I can't decrypt data after find.

IMO, the hooks are not the right tool for transparent encryption of model properties (data), because the instance passed to "before save" hooks is then returned in the server response. The HTTP client will thus receive encrypted data, unless you implement also an "after save" hook that would decrypt them.

Could you please open a new issue in loopback requesting the feature for an easy way how to specify a different database representation of the model data?

For example, if you merge one instance of a model into the other, then having an 'after access' hook would allow you to put a "redirectId" object in the old instance so that outdated references to the old model can be forwarded to the new model.

This looks like reasonable usage of the new hook to me. Perhaps we can call the new hook "loaded"?

Just note that making database requests from such hook will incur performance overhead.

Ideally, LoopBack should provide a way how to fetch all related data (like your "redirectId") in a single request, either via SQL JOIN or using an appropriate NoSQL alternative. Since we don't support that yet, adding "loaded"/"after access" hook is a reasonable workaround.

[raymondfeng]

+1 to add after access hook. That could be used by ACLs or custom logic to further transform the responses from a DB query.

[jmls]

bugger. I need this functionality. gutted it's not in already.

[fabien]

+1 to after access as well

[raymondfeng]

+1 to have after access.

[fabien]

Do we want to supply ctx.instance (obj?) and cx.data (array?) for this hook?

[drish]

+1 to after access

[bajtos]

IMO, the hook should be called loaded, it should be triggered whenever model data from the database is returned back to the caller (i.e. find but also create or prototype.updateAttributes) and the context should provide only ctx.instance, no ctx.data. For "batch" load like find, the hook should be fired multiple times, once for each model instance loaded.

[fabien]

@bajtos fully agreed 👍