*: support extra dBFT stage #118
Commits on Jul 17, 2024
-
internal: move tx64 to internal/consensus package
It will be reused by anti-MEV related dBFT interfaces implementations. Signed-off-by: Anna Shaleva <[email protected]>
Commits on Jul 30, 2024
-
Ref. #112. Signed-off-by: Anna Shaleva <[email protected]>
Commits on Jul 31, 2024
-
*: add testing infrastructure for Anti-MEV dBFT extension
Add custom PreBlock and Block interfaces implementation, custom Commit and CommitAck, adjust testing logic. Signed-off-by: Anna Shaleva <[email protected]>
-
dbft: don't react on transactions if PreCommit or Commit sent
Technically, this is implied by d.ResponseSent(), but we still have BlockSent() check as well, so these don't hurt. Signed-off-by: Roman Khimov <[email protected]>
-
dbft: remove redundant index check
It's done at the start of OnReceive() and it's even more strict there. Signed-off-by: Roman Khimov <[email protected]>
-
dbft: raise PrepareRequest sender mismatch log to Info
It's a security event, this should never happen and we better know if it does. Signed-off-by: Roman Khimov <[email protected]>
-
dbft: drop "empty PrepareRequest" log
It's useless, we log tx number down below anyway. Signed-off-by: Roman Khimov <[email protected]>
-
dbft: fix commit checking with AMEV enabled
Commit payloads can't be checked with PrepareRequest data only in AMEV case, we don't yet have a complete header. So check them after we have enough PreCommit payloads. Signed-off-by: Roman Khimov <[email protected]>
-
dbft: reduce cyclomatic complexity of on[Pre]Commit()
Be a bit more optimistic about the message (similar to how onPrepareResponse() works). Signed-off-by: Roman Khimov <[email protected]>
-
context: reset preBlockProcessed properly
Signed-off-by: Roman Khimov <[email protected]>
-
context: define preBlockProcessed, make it fire only once
Supposedly AMEV code will do its magic here and it doesn't need to do it twice. Signed-off-by: Roman Khimov <[email protected]>
-
For a single height/view there is a single proper hash and it's not a final block hash, likely logging it won't help much. Can be done in future if needed, but let's have view number for now. Signed-off-by: Roman Khimov <[email protected]>
-
recovery: add PreCommit support
Signed-off-by: Roman Khimov <[email protected]>
-
dbft: extend single node test with AMEV
Signed-off-by: Roman Khimov <[email protected]>
-
dbft: add caching support for precommits
Signed-off-by: Roman Khimov <[email protected]>
-
dbft: make future message cache work properly, add a test
start() was designed to be called at every view change, but looks like it _never_ worked this way. Which means two things: * on every view change Primary doesn't send PrepareRequest during initialization (which is mostly OK, OnTimeout() will be triggered immediately with 0 timeout) * our future message caching system has never really worked since start() is the only place where messages can be picked up from it Just drop start(), make caches work and add a test for them. Signed-off-by: Roman Khimov <[email protected]>