You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Seen in dogfood. sis_v2p_ensure_undo and sis_ensure_registered_undo both try to get an instance ID from a sagactx.lookup that refers to a nonexistent node. They should get the ID from the saga params instead like all the other nodes.
The fix itself is very simple; a more interesting problem is why the action unwind test for this saga didn't catch the problem. Investigating that now.
The text was updated successfully, but these errors were encountered:
The fix itself is very simple; a more interesting problem is why the action unwind test for this saga didn't catch the problem. Investigating that now.
You might think that, as the author of #3309, I would have been able to avoid recreating a similar problem in the start saga. If so, you (and I) thought incorrectly: the bug here is that the start saga unwind test is failing ever to get past the first node and so is not actually exercising the undo steps for later nodes. This points to a bug in the "set instance to starting" logic or its undo step that I'm now investigating.
This points to a bug in the "set instance to starting" logic or its undo step that I'm now investigating.
The problem is actually a test issue: prior iterations of the saga are changing the parameters that need to be passed to subsequent iterations, but the test isn't regenerating those parameters. (IOW, new test iterations aren't starting from a clean slate the way a new API call to the start endpoint would.) After fixing this issue I can reproduce the actual bug and am now testing the fix.
The instance start saga has two undo actions that try to read an
instance ID from a nonexistent prior action (I incorrectly copied them
from the instance create saga). The saga unwind test missed this because
it erroneously checked only that its saga invocations were failing and
not that they were failing at the correct step.
Fix the test so that it can reproduce this issue, then make the undo
nodes read the saga IDs from the correct place.
Tested: cargo test, both with the test fix and no code fix (fails as
expected) and with both fixes (passes as expected).
Fixes#3894.
Seen in dogfood.
sis_v2p_ensure_undo
andsis_ensure_registered_undo
both try to get an instance ID from asagactx.lookup
that refers to a nonexistent node. They should get the ID from the saga params instead like all the other nodes.The fix itself is very simple; a more interesting problem is why the action unwind test for this saga didn't catch the problem. Investigating that now.
The text was updated successfully, but these errors were encountered: