PSS: Use interfaces to configure state stores #37899
Conversation
SarahFrench
added
the
no-changelog-needed
…tead of an instance of the `backend.Backend` interface.
…e, instead of directly using methods related to RPCs. This requires changing where we call `SetStateStoreChunkSize`.
To summarize, we don't really need the `SetStateStoreChunkSize` method, and instead methods like `(*GRPCProvider).ConfigureStateStore` in the `plugin6` package can directly inspect the negotiation process that passes through that code and pull out the chunk size. However that means that that code would also need to implement validation. And that's just `(*GRPCProvider).ConfigureStateStore`; what about all the test mocks that are used in different types of test? They would all need to be implemented similarly to GRPCProvider to be good, useful mocks, and then a lot of things that fulfil the provider.Interface interface are coupled to each other. Instead, it's easier to have validation in the `grpcClient` struct's methods in the `remote` package, as that code is common to all scenarios. That code needs a method to 'reach into' the provider.Interface value, so we use the `SetStateStoreChunkSize` method.
SarahFrench
force-pushed
the
pss/use-interface-configure-store
branch
from
97dd6bf to
a38298d
Compare
I'm surprised that removing this doesn't break E2E tests of PSS that use grpcwrap, but I think there's `plugin6` code that runs in that situation, so maybe chunking is handled elsewhere.
| if cs, ok := p.provider.(providers.StateStoreChunkSizeSetter); ok { | ||
| cs.SetStateStoreChunkSize(p.typeName, int(chunkSize)) | ||
| } else { | ||
| // TODO: Remove |
There was a problem hiding this comment.
It feels like leaving this in would be helpful, any objections?
There was a problem hiding this comment.
We could also avoid the _, ok notation above and let it panic that way. I think it's merely a question of how likely this is ever to happen and be exposed to the end user.
Personally, I like panic to be a direct side effect of a problem rather than intention where it is possible (i.e. I'd rather let the casting panic) and leaving it to Go to express itself but I do not hold that opinion too strongly. 🤷🏻
There was a problem hiding this comment.
That makes sense - I'll push a change for this
| if cs, ok := p.provider.(providers.StateStoreChunkSizeSetter); ok { | ||
| cs.SetStateStoreChunkSize(p.typeName, int(chunkSize)) | ||
| } else { | ||
| // TODO: Remove |
There was a problem hiding this comment.
We could also avoid the _, ok notation above and let it panic that way. I think it's merely a question of how likely this is ever to happen and be exposed to the end user.
Personally, I like panic to be a direct side effect of a problem rather than intention where it is possible (i.e. I'd rather let the casting panic) and leaving it to Go to express itself but I do not hold that opinion too strongly. 🤷🏻
…using grpcwrap package This was removed, incorrectly, in #37899
…using grpcwrap package This was removed, incorrectly, in #37899
…using grpcwrap package This was removed, incorrectly, in #37899
2 participants
I realised that the code for calling RPC methods had leaked through the code further than intended, so this PR updates calling code in
Metamethods and ensures that configuring a state store is only done via thebackend.Backendinterface.Target Release
1.15.x
Rollback Plan
Changes to Security Controls
Are there any changes to security controls (access controls, encryption, logging) in this pull request? If so, explain.
CHANGELOG entry