mbox series

[v6,0/5] migration: do not exit on incoming failure

Message ID 20240430085646.2359711-1-vsementsov@yandex-team.ru
Headers show
Series migration: do not exit on incoming failure | expand

Message

Vladimir Sementsov-Ogievskiy April 30, 2024, 8:56 a.m. UTC 
Hi all!

The series brings an option to not immediately exit on incoming
migration failure, giving a possibility to orchestrator to get the error
through QAPI and shutdown QEMU by "quit".

v6:
01,02: add r-b by Peter
03: only fix potential use-after-free
04: rework error handling, drop r-b


v5:
- add "migration: process_incoming_migration_co(): fix reporting s->error"

v4:
- add r-b and a-b by Fabiano and Markus
- improve wording in 04 as Markus suggested

v3:
- don't refactor the whole code around setting migration error, it seems
  too much and necessary for the new feature itself
- add constant
- change behavior for HMP command
- split some things to separate patches
- and more, by Peter's suggestions


New behavior can be demonstrated like this:

bash:

(
cat <<EOF
{'execute': 'qmp_capabilities'}
{'execute': 'migrate-set-capabilities', 'arguments': {'capabilities': [{'capability': 'events', 'state': true}]}}
{'execute': 'migrate-incoming', 'arguments': {'uri': 'exec:echo x', 'exit-on-error': false}}
EOF
sleep 1
cat <<EOF
{'execute': 'query-migrate'}
{'execute': 'quit'}
EOF
) | ./build/qemu-system-x86_64 -incoming 'defer' -qmp stdio -nographic -nodefaults

output:

{"QMP": {"version": {"qemu": {"micro": 50, "minor": 0, "major": 9}, "package": "v9.0.0-149-gb6295ad58c"}, "capabilities": ["oob"]}}
{"return": {}}
{"return": {}}
{"timestamp": {"seconds": 1714068847, "microseconds": 263907}, "event": "MIGRATION", "data": {"status": "setup"}}
{"return": {}}
{"timestamp": {"seconds": 1714068847, "microseconds": 266696}, "event": "MIGRATION", "data": {"status": "active"}}
qemu-system-x86_64: Not a migration stream
{"timestamp": {"seconds": 1714068847, "microseconds": 266766}, "event": "MIGRATION", "data": {"status": "failed"}}
{"return": {"status": "failed", "error-desc": "load of migration failed: Invalid argument"}}
{"timestamp": {"seconds": 1714068848, "microseconds": 237187}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
{"return": {}}

Vladimir Sementsov-Ogievskiy (5):
  migration: move trace-point from migrate_fd_error to migrate_set_error
  migration: process_incoming_migration_co(): complete cleanup on
    failure
  migration: process_incoming_migration_co(): fix reporting s->error
  migration: process_incoming_migration_co(): rework error reporting
  qapi: introduce exit-on-error parameter for migrate-incoming

 migration/migration-hmp-cmds.c |  2 +-
 migration/migration.c          | 54 ++++++++++++++++++++++++----------
 migration/migration.h          |  3 ++
 migration/trace-events         |  2 +-
 qapi/migration.json            |  7 ++++-
 system/vl.c                    |  3 +-
 6 files changed, 52 insertions(+), 19 deletions(-)

Comments

Peter Xu May 1, 2024, 4 p.m. UTC | #1 
On Tue, Apr 30, 2024 at 11:56:41AM +0300, Vladimir Sementsov-Ogievskiy wrote:
> Hi all!
> 
> The series brings an option to not immediately exit on incoming
> migration failure, giving a possibility to orchestrator to get the error
> through QAPI and shutdown QEMU by "quit".

There seem to have one comment overlooked in the last patch I left before
(on a probably useless error_desc pre-check), but doesn't look a huge
deal..

Reviewed-by: Peter Xu <peterx@redhat.com>

Thanks,