I was able to reproduce the issue for a VM without IO thread busy
with IO on a heavily rate-limited storage.
The fix has two independent parts:
1) increase the QMP timeout:
The 'quit' QMP command itself is fast, since it only records the
shutdown request and notifies the main loop, but getting the response
can take a while. The reason is that qmp_dispatch() yields and must be
woken after executing the command and at that stage, QEMU is already
busy with teardown too.
2) skip VM start after successfully saving state:
When creating a snapshot of the VM state for hibernation, the VM must
not be started again if the snapshot operation was completed
successfully. The VM should remain stopped, so that the QMP 'quit' is
issued without further activity from the VM.
The latter addresses a more fundamental issue, but it still makes
sense to bump the timeout for 'quit' regardless, because the
explanation in 1) still applies.
I also added the rebase to QEMU 11.0.1 as part of this series, since
using the new parameter for 2) needs to be somehow version-guarded.
There would be other ways like extending QMP 'query-proxmox-support',
but since the opportunity with 11.0.1 is there, I felt like it is
easier in this case.
qemu:
Fiona Ebner (3):
regenerate patches to restore incremental numbering
async snapshot: allow skipping VM start after successful completion
update submodule and patches to QEMU 11.0.1
...d-support-for-sync-bitmap-mode-never.patch | 6 +-
...-support-for-conditional-and-always-.patch | 2 +-
...-to-bdrv_dirty_bitmap_merge_internal.patch | 2 +-
.../0006-mirror-move-some-checks-to-qmp.patch | 2 +-
...-to-bounce-buffer-if-BLKZEROOUT-is-.patch} | 0
...ial-deadlock-when-draining-during-tr.patch | 100 ------------------
...oid-idle-event-loop-being-accounted.patch} | 0
...-fix-regression-when-mmap-ing-expor.patch} | 0
...-fix-regression-with-block-device-e.patch} | 0
...fix-decoding-of-MOVBE-and-CRC32-in-1.patch | 84 ---------------
...-accidentally-autofree-existing-virg.patch | 59 -----------
...store-IRQ-polling-for-non-kernel-irq.patch | 47 --------
...strList-leak-in-x86_cpu_get_unavaila.patch | 36 -------
...-missing-PF_INSTR-in-SIGSEGV-context.patch | 43 --------
...e_save_state_v-fix-double-error_setg.patch | 46 --------
...async-for-background-state-snapshots.patch | 49 ++++++---
...add-optional-buffer-size-to-QEMUFile.patch | 6 +-
...add-the-zeroinit-block-driver-filter.patch | 4 +-
...le-posix-make-locking-optiono-on-cre.patch | 6 +-
...VE-Backup-add-vma-backup-format-code.patch | 6 +-
...ckup-Proxmox-backup-patches-for-QEMU.patch | 8 +-
...estore-new-command-to-restore-from-p.patch | 4 +-
...k-driver-to-map-backup-archives-into.patch | 8 +-
...ct-stderr-to-journal-when-daemonized.patch | 6 +-
...igrate-dirty-bitmap-state-via-savevm.patch | 4 +-
.../0037-block-add-alloc-track-driver.patch | 4 +-
.../0038-PVE-backup-add-fleecing-option.patch | 2 +-
...ment-backup-access-setup-and-teardow.patch | 2 +-
...se-migration-blocker-check-for-snaps.patch | 6 +-
debian/patches/series | 15 +--
qemu | 2 +-
31 files changed, 78 insertions(+), 481 deletions(-)
rename
debian/patches/extra/{0003-block-io-fallback-to-bounce-buffer-if-BLKZEROOUT-is-.patch
=> 0002-block-io-fallback-to-bounce-buffer-if-BLKZEROOUT-is-.patch} (100%)
delete mode 100644
debian/patches/extra/0002-ide-avoid-potential-deadlock-when-draining-during-tr.patch
rename
debian/patches/extra/{0004-fdmon-io_uring-avoid-idle-event-loop-being-accounted.patch
=> 0003-fdmon-io_uring-avoid-idle-event-loop-being-accounted.patch} (100%)
rename
debian/patches/extra/{0012-block-export-fuse-fix-regression-when-mmap-ing-expor.patch
=> 0004-block-export-fuse-fix-regression-when-mmap-ing-expor.patch} (100%)
rename
debian/patches/extra/{0013-block-export-fuse-fix-regression-with-block-device-e.patch
=> 0005-block-export-fuse-fix-regression-with-block-device-e.patch} (100%)
delete mode 100644
debian/patches/extra/0005-target-i386-tcg-fix-decoding-of-MOVBE-and-CRC32-in-1.patch
delete mode 100644
debian/patches/extra/0006-hw-display-don-t-accidentally-autofree-existing-virg.patch
delete mode 100644
debian/patches/extra/0007-hw-i386-vapic-restore-IRQ-polling-for-non-kernel-irq.patch
delete mode 100644
debian/patches/extra/0009-target-i386-fix-strList-leak-in-x86_cpu_get_unavaila.patch
delete mode 100644
debian/patches/extra/0010-target-i386-fix-missing-PF_INSTR-in-SIGSEGV-context.patch
delete mode 100644
debian/patches/extra/0011-migration-vmstate_save_state_v-fix-double-error_setg.patch
qemu-server:
Fiona Ebner (3):
fix #6424: increase timeout for QMP 'quit' to 60s to avoid issue with
bulk suspend
run state: use v5.36 and signatures in module
suspend: skip VM start after successfully saving state
src/PVE/QMPClient.pm | 6 ++++++
src/PVE/QemuServer/RunState.pm | 32 ++++++++++++++++----------------
2 files changed, 22 insertions(+), 16 deletions(-)
Summary over all repositories:
33 files changed, 100 insertions(+), 497 deletions(-)
--
Generated by git-murpp 0.5.0
