generalize Node.run for invocation from both real node and from ThreadNet tests and refine interface used by real node #2720
Commits on Nov 12, 2020
-
network: parameterize DiffusionApplications record
Adds the following parameters to the DiIffusionApplications record type, so that it can be used both for the real diffusion layer and for a mock layer in the ThreadNet tests. * addresses: real uses RemoteAddress and LocalAddress, ThreadNet uses CoreNodeId * version data: real uses NodeToNodeVersionData and NodeToClientVersionData, ThreadNet uses UnversionedProtocolData * monad: real uses IO, ThreadNet uses IOLike m => m -
consensus: parameterize Node.run over the diffusion layer
The ThreadNet tests will run a mock diffusion layer. It's simpler than the real diffusion layer, and it's connection/disconnection timeline can be scripted according to each generated " test plan " (eg enabling shrinking to an explicitly static topology).
-
-
consensus: decouple Node.run from ChainDB file system
The ThreadNet tests use a mock filesystem, so .run must not directly use Ouroboros.Consensus.Storage.FS.IO.ioHasFS to initialize the ChainDB. Moreover, the tests use fully separate mock FSs for each internal database, for example to better localize reporting of unclosed file handles. So .run should not even directly organize the internal databases as sibling folders.
-
-
consensus: decouple Node.run from randomIO, including ChainSyncTimeout
Moreover, the ThreadNet tests limit the ChainSync delays by construction instead of dynamically.
-
consensus: parameterize Node.run over the address types
The ThreadNet tests uses CoreNodeId whereas the real node uses RemoteAddress (ie SockAddr) and LocalAddress (ie FilePath).
-
-
consensus: move customise* fields of RunNodeArgs to LowLevelRunNodeArgs
The cardano-node invocation currently uses these functions. We therefore add sufficient corresponding fields to StdRunNodeArgs so that the cardano-node invocation can be simplified: it just passes the data without having to first interpret it as these customisation functions.
-
-
consensus: move clock skew from RunNodeArgs to LowLevelRunNodeArgs
It is not an option the real node likely has to override, better to not require them to provide it.
-
consensus: change prefix of LowLevelRunNodeArgs from rn- to llrn-
For consistency.
-
Left-over from alignment in the distant past.
-
consensus: make the protocol versions a low-level argument
In `cardano-node`, the default values for node-to-node and node-to-client versions are used. The only reason for having them in `RunNodeArgs` was to make it possible for the network team to test the node with different versions. They can still do this by using `.runWith` and `stdLowLevelRunNodeArgsIO`, and overriding the fields manually.
-
consensus: don't require NetworkMagic in StdRunNodeArgs
`RunNodeArgs` contains the `ProtocolInfo`, which already contains the `NetworkMagic`. So don't require it in `StdRunNodeArgs` but derive it from the `RunNodeArgs`. This means we have to pass `RunNodeArgs` to `stdLowLevelRunNodeArgsIO`.
