Chaos testing phase 3 - #3761
Conversation
…refusal correctly
…cross builds" This reverts commit 0ed1fb6.
| const VERSION_ON_THE_NEW_BUILD: u32 = 2; | ||
| const VERSION_AFTER_ROLLBACK: u32 = 1; | ||
|
|
||
| pub async fn run( |
There was a problem hiding this comment.
There seem to be some confusion in this scenario between component revision and component version. These are not guaranteed to be the same thing with rollbacks
| /// account of nothing. | ||
| const WAKE_PROOF_SAMPLE: usize = 5; | ||
|
|
||
| pub async fn run( |
There was a problem hiding this comment.
To be checked:
stood_down is a global count. Once any waiter stalls, every unreadable waiter missing a confirmed wakeup is classified as NeverWoke, even if that particular waiter never stood down. One genuine stall plus an unrelated read timeout can therefore produce false violations and fail the scenario.
Fix: retain stalled waiter identities and require both unreadable.contains(agent) and stalled_waiters.contains(agent).
| /// account of nothing. | ||
| const FIRE_PROOF_SAMPLE: usize = 5; | ||
|
|
||
| pub async fn run( |
There was a problem hiding this comment.
To check:
needs_attention() checks only total == 0. If pending actions exist exclusively on surviving executors, on_killed_executor == 0, but the result is classified as normal context even though the kill exercised no lease recovery on the killed executor. S11 correctly handles the analogous condition.
Fix: use self.total == 0 || self.on_killed_executor == 0, add the corresponding warning in describe(), and cover it with a test.
vigoo
left a comment
vigoo
left a comment
There was a problem hiding this comment.
Added three findings, other than that looks good
2 participants
No description provided.