Invalidating cache with update commands

[filiptronicek]

If the database gets updated, the associated cache keys should be invalidated (either by updating or deleting).

[Asjas]

I added Basic Cache Invalidation in version 1.1.0.

I'm basically searching the Redis cache for all keys that matches a model when a Prisma mutation action happens. I then delete all those keys from Redis.

model in this case would be User, Post or similar.

const keys = await redis.keys(`${model}:*`);
const deletedKeys = await redis.del(keys);

So if create or updateMany is called on the User model, all cached keys that refer to the User model will be deleted.

[Asjas]

I know most Prisma actions (queries and mutations) usually include an id field in the args or next return value.

I don't know if this is true for all actions or all databases that are supported by Prisma, but that could be something that would allow me to create a more refined cache where only cached queries that match a model AND that match an id field can be deleted.

I'm curious what other people's opinion on this would be?

[f-o-w-l]

I know most Prisma actions (queries and mutations) usually include an id field in the args or next return value.

I don't know if this is true for all actions or all databases that are supported by Prisma, but that could be something that would allow me to create a more refined cache where only cached queries that match a model AND that match an id field can be deleted.

I'm curious what other people's opinion on this would be?

It makes sense for me to call it a primary key for each model. Models are basically tables, right? If so, then there should only be one primary key per model. Disclaimer, I haven't used Prisma, I'm currently learning it. Referencing #62, I think it would be great to have this kind of API:

prismaClient.$use(
  createPrismaRedisCache({
    models: [
      {
        model: "User",
        primaryKey: "id",
      },
      {
        model: "Post",
        cacheTime: 900,
        primaryKey: "postId",
        excludeCacheMethods: ["findMany", "aggregate"],
      }
    ],
    defaultCacheTime: 300,
    redis,
    defaultExcludeCacheMethods: ["findMany"],
  }),
);

The docs could mention that if a primaryKey isn't specified, then the full model is deleted from cache after any mutation, otherwise the mutation will only delete queries that match the model and primaryKey. I think it makes sense to fully flush the model cache by default instead of checking for the "id" property specifically by default. I like the idea of default fallbacks not just for the cacheTime, I can't think of any downside to allowing excludeCacheMethods to be specified per model. What do you think?

[Asjas]

v2.0.0 has been released.