Stellar RPC - Lack of information during ExceededLimit simulation errors

[mootz12]

During a simulateTransaction ExceededLimit failure, there is virtually no information shared with the user besides the failure, and no simple way to get more information.

For example, here is the SimulateTransactionErrorResponse for 2 different transactions w/ diagnostic events enabled:

{
  _parsed: true,
  id: undefined,
  latestLedger: 56826563,
  events: [],
  error: 'HostError: Error(Budget, ExceededLimit)\nDebugInfo not available\n'
}

{
  _parsed: true,
  id: undefined,
  latestLedger: 56826569,
  events: [
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] },
    ChildStruct { _attributes: [Object] }
  ],
  error: 'HostError: Error(Contract, #1215)\n' +
    '\n' +
    'Event log (newest first):\n' +
    '   0: [Diagnostic Event] contract:CAJJZSGMMM3PD7N33TAPHGBUGTB43OC73HVIK2L2G6BNGGGYOSSYBXBD, topics:[error, Error(Contract, #1215)], data:"escalating error to VM trap from failed host function call: fail_with_error"\n' +
    ...
}

Given the large number of ExceededLimit error reasons possible, it would be a significant improvement to share what limit was tripped.

Currently, if you run into this, the best way to diagnose what limit it being reached is to download a (mainnet/testnet) snapshot with all the relevant contracts (not easy if you don't know the contract you are interacting with well), and write a Rust test to determine what resources were used.

There are also differences in snapshot testing vs simulation, so the reproduction might not always be accurate.

[dmkozh]

There is a bunch of env issues, that we need to address, though some of these aren't very straightforward (specifically, adding more logging for the budget errors).

Given the large number of ExceededLimit error reasons possible

FWIW the number of reasons is not that large: budget only covers CPU and memory (and I've yet to see the memory limit being hit). But this unfortunately is also the error we return for exceeding the shadow budget - hitting the shadow budget limit for regular transactions has never been intended. Without that caveat it would've been safe to say that it must be CPU instructions exceeded even without looking at diagnostics.

[mootz12]

Ah, I thought exceeding other limits (e.g. read entries) would also throw this error

[dmkozh]

The other limits are not enforced in simulation at the moment, which is also an issue that we'll hopefully address eventually (though clear error messaging will be much easier to achieve there).