Errors in mailin don't abort the transaction
Bug #1375884 reported by
Chris Rossi
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
KARL4 |
Confirmed
|
Medium
|
Unassigned |
Bug Description
In testing mailin on staging I was able to corrupt the blog, a blog post was created and then exception was raise, and instead of rolling back the transaction, the transaction was committed, leading to a corrupt blog entry:
https:/
Changed in karl3: | |
importance: | Undecided → High |
status: | New → In Progress |
milestone: | none → m140 |
assignee: | nobody → Chris Rossi (chris-archimedeanco) |
importance: | High → Undecided |
Changed in karl3: | |
status: | In Progress → Confirmed |
Changed in karl3: | |
importance: | Undecided → Medium |
milestone: | m140 → m141 |
affects: | karl3 → karl4 |
Changed in karl4: | |
milestone: | m142 → none |
Changed in karl4: | |
milestone: | none → 002 |
Changed in karl4: | |
milestone: | 002 → 003 |
Changed in karl4: | |
milestone: | 003 → 004 |
Changed in karl4: | |
milestone: | 004 → 005 |
Changed in karl4: | |
milestone: | 005 → 999 |
assignee: | Chris Rossi (chris-archimedeanco) → nobody |
To post a comment you must log in.
Now that I look at this, this doesn't have to do with karlserveless. It's been latent in Karl for some time. Basically when we put messages in the quarantine, because they have an exception for some reason, we don't do anything to roll back the work that might have been done before we hit the exception. We don't want to abort the entire transaction, since that would defeat the point of moving messages to the quarantine. And, of course, moving messages to the quarantine is part of the transaction. It might make sense to start using savepoints, though, and roll back to a savepoint made before attempting to process a message, before moving the message to the quarantine. I've never been superclear how well savepoints work in ZODB. Will look into it.