Adversary #492
Adversary #492
Conversation
| "doc:build:sdk": "jsdoc -c jsdoc.json cli/lib/nf3.mjs" | ||
| "doc:build:sdk": "jsdoc -c jsdoc.json cli/lib/nf3.mjs", | ||
| "build-adversary": "node test/adversary/transpile-adversary.mjs", | ||
| "adversary-test": "mocha --timeout 0 --bail --exit test/adversary.test.mjs" |
There was a problem hiding this comment.
We should call build-adversary inside the adversary-test so that we don't have people run the test against "outdated" adversaries
There was a problem hiding this comment.
Adversary code is required at the stage of ./setup-nightfall and ./start-nightfall. Only after these two are completed do we run the tests. If there is no source code for adversary by ./start-nightfall , it complains (don't remember if it fails, my memory says it does).
We could call build-adversary in the test. It would work if some adversary code already exists during build and start stage and this version of adversary is started. After this, nodemon will create the new code again during test and restart the adversary container.
Considering we won't use nodemon in non dev deployments, I would recommend we get the developers used to calling build-adversary at the beginning of staging and prod deployments. Because they have to remember calling ./start-nightfall with -a, as well as ./setup-nightfall with adversary service too.
PS: We don't push nightfall-adversary to repo. It is part of .gitignore file
| async function makeBlock(proposer, number = TRANSACTIONS_PER_BLOCK) { | ||
| logger.debug('Block Assembler - about to make a new block'); | ||
| // pick a random number between 0 to 6 (error length - 1) | ||
| const errorIndex = Math.floor(Math.random() * 0); |
There was a problem hiding this comment.
Not sure what errorIndex does here - it seems to always be 0. count is always 0 as well.
Does this mean that getMostProfitableTransactions always makes valid transactions?
There was a problem hiding this comment.
It's a global variable. So it is incremented with each block proposed.
Non valid blocks get created as it increments.
This is a test run from github action
There was a problem hiding this comment.
ah ok - so errorIndex is always zero because we are not doing the random thing for now
This PR generates nigthfall adversary. Nightfall adversary is meant to be an adversarial block proposer that creates a combination of good and bad blocks. This is built for testing purposes. It is created using generative metaprogramming by copying nightfall-optimist code and injecting the code that creates bad blocks and bad transactions during block creation and transactions selection respectively. It creates blocks of type
Test
This functionality is tested by
adversary-testjob in check-PRs github action.If the test is a success it means that a user has the correct balance considering the blocks proposed and rollbacks from bad blocks that contain their transaction. Bad block types to be generated are defined in files
test/adversary/adversary-code/database.mjsandtest/adversary/adversary-code/block.mjs. A challenger has spotted these bad blocks and successfully challenged them.Removed
neg-http.mjsbecause this implements the same functionality a lot cleanerNote
The nightfall adversary code is written such that it picks bad block or bad transaction type in the order as defined in
database.mjsandblock.mjsfiles. This order avoids the adversary from randomly pickingDuplicateNullifierwhen there are no spent transactions whose nullifiers can be duplicatedDuplicateTransactionwhen there are no transactions to duplicateIncorrectLeafCountwhen there is no prior blockIncorrectTreeRootwhen there are no 2 prior blocksThis logic will later be replaced such that type of bad block to be generated will be picked automatically and randomly by nightfall adversary. This update depends on having a script that seeds the deployment with the first two L2 blocks and at least two transfers in the second blocks. This is being handled in the issue