cheeeee commented on PR #23928:
URL: https://github.com/apache/kafka/pull/23928#issuecomment-6366724048
Test notes, continued: OpenBSD 8.0 (arm64, QEMU on macOS) and Windows
10 LTSC (x86-64, QEMU/KVM), both with JDK 17, on 73fc46a880 and on
trunk d18f93f985. Each VM runs one broker and controller, with t1-0 on
a data disk behind QEMU's blkdebug, which fails reads of the disk
sector holding position 4 MiB of the segment with EIO: every read, or
only the first one. A consumer reads the partition from the beginning.
* TLS, every read fails: on both, the network thread's read fails at
position 4194304. Trunk re-reads the region on the scheduler (3.0 MB
on OpenBSD, 3.6 MB on Windows), the re-read fails too, and the
directory goes offline; this change offlines it without a re-read.
* TLS, only the first read fails: on OpenBSD trunk re-reads the region
successfully and keeps the directory online, while this change
offlines it. On Windows the disk driver retries the failed read
(System event 153) and the retry succeeds, so the JVM never sees an
error, and both keep the directory online.
* PLAINTEXT, every read fails: the JDK writes the file to the socket
from a memory mapping on both, and the write fails with EFAULT ("Bad
address") on OpenBSD and WSAEINVAL ("An invalid argument was
supplied") on Windows rather than with an I/O error. Trunk and this
change both re-read the region, the re-read fails, and the directory
goes offline.
No InternalError was thrown in any run.
--
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]
