Once again, iothreads making trouble. When enabled, snapshots including RAM
deadlock QEMU, because our async-snapshot implementation (which recently moved
back to using coroutines) tries to access and modify the state of disks running
in seperate iothreads from the main one.
Patch 1/4 fixes the issue and can be applied standalone, patches 2 and 3 improve
snapshot performance for iothread-disks and patch 4 adds some useful debug
prints for testing the aforementioned performance patches. See individual patch
notes for more.
For easier reviewing I sent the patches for the QEMU source itself, if necessary
I can also apply them and then send pve-qemu patches including them as .patch
files.
Kevin Wolf (1):
util/async: Add aio_co_reschedule_self()
Stefan Reiter (3):
savevm-async: move more code to cleanup and rename to finalize
savevm-async: flush IOThread-drives async before entering blocking
part
savevm-async: add debug timing prints
include/block/aio.h | 10 ++++
savevm-async.c | 124 +++++++++++++++++++++++++++++++-------------
util/async.c | 30 +++++++++++
3 files changed, 129 insertions(+), 35 deletions(-)
--
2.20.1
_______________________________________________
pve-devel mailing list
pve-devel@pve.proxmox.com
https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
