xmg333 commented on PR #18463:
URL:
https://github.com/apache/dolphinscheduler/pull/18463#issuecomment-6109135901
All three points are addressed in 176655c350:
- **Premature EOF**: an EOF/empty chunk that ends before the pinned length
now throws instead of
returning — the response is aborted, nothing short is published. When the
pin is unknown
(`observedLength == 0`, i.e. the pre-PR worker during a rolling upgrade)
there is nothing to
compare, so the EOF-only termination stays; that's spelled out in the
javadoc.
- **Shrink below the pin**: a later chunk reporting a length below the pin
is rejected before its
bytes are written (in-place truncation, or rotation to a shorter file).
- **Coverage**: the previous rotation test only shrank the file below the
delivered offset. The new
unit tests cover shrinking above the delivered offset, shrinking to
exactly the offset, and
premature EOF with no `observedLength`. The integration test now cuts a
real 9 MiB log to 8.5 MiB
while the download is in flight over the real Netty RPC — the new length
is above the delivered
offset, so the worker's own truncation check stays silent and the API has
to fail by itself.
Both rejection paths carry the task instance id and host, so the aborted
download is attributable in
the API log.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
