@apollo/client InMemoryCache for React

Hi,

In my current project, I want to loop through the ApolloClient cache in order to list the key and pick the ones I want to clear for specific clients.

I am using lodash’s forEach function and doing the following

forEach(client.cache.data.data, (entry, key) => {

});

Is this the best approach? Probably not, but I couldn’t find any documentation that explains how I could achieve this the best way possible.

The reason behind this is because on my GraphQL server, I am sending to the client the _id of all the data that is being cleared and need to be evicted from the connected clients.

Since the id looks like TypeName:ID, I wanted to loop through the cache and remove all entries that match the _id in the keys instead of having to return the TypeName:ID.

Hope you can point me in the right direction, thanks!

Hi @Kheang_Hok_Chin :wave: thanks for posting! I haven’t seen this approach before. Can you describe a bit more about your use case? A few questions come to mind:

  1. What does the server know that makes it the source of truth for client-side cache invalidation?
  2. What sort of performance or scale issues have you been seeing?
  3. How large do your client-side caches get?

Thanks for your comment.

I use mongodb for a datasource in my GraphQL server and cache some of the queries.
I clear them when a mutation occurs.
On the client side, I need to be able to know what to clear from the cache when that happens.

At first I thought that I could simply use cache.evict({ id: "" }) to clear a specific result and use cache.gc() but it didn’t seem to produce the result I needed.

So instead I followed this approach here that I mentioned that also allows me to scale.

On the client side, I listen on a socket connection for a list of queries to clear when a mutation is executed on GraphQL’s side.

// loop through queries and clear the cache for each
forEach(get(cacheEvict, "queries"), fieldName => {
	client.cache.evict({ id: "ROOT_QUERY", fieldName });
});

This is less specific than I want it to be but it works for my use case until I find an even more precise solution that would work better. Since I had no examples to rely upon for complex scenarios, this is the one I came up with.