[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH 1/3] qemu: handle new 'setup' migration state



On 07/26/2013 02:32 PM, Eric Blake wrote:
On 07/26/2013 11:47 AM, mrhines linux vnet ibm com wrote:
From: "Michael R. Hines" <mrhines us ibm com>

Previously, QEMU's 'setup' state was no a formal state in their
state machine, but it is now. This state is used by RDMA to optionally
perform memory pinning. This state is now exposed over the monitor
and also measured in the migration info status.

This patch consumes both the new setup state as well as the timestamp
of the total time spent in that state as reported by QEMU.

RDMA migrations perform an optional 'pin-all' operation du

Signed-off-by: Michael R. Hines <mrhines us ibm com>
---
  include/libvirt/libvirt.h.in |    2 ++
  src/qemu/qemu_migration.c    |    6 ++++++
  src/qemu/qemu_monitor.c      |    2 +-
  src/qemu/qemu_monitor.h      |   11 +++++++++++
  src/qemu/qemu_monitor_json.c |   18 ++++++++++++++++++
  5 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/include/libvirt/libvirt.h.in b/include/libvirt/libvirt.h.in
index c0eb25b..31fb37e 100644
--- a/include/libvirt/libvirt.h.in
+++ b/include/libvirt/libvirt.h.in
@@ -4048,6 +4048,8 @@ struct _virDomainJobInfo {
      /* Time is measured in mill-seconds */
      unsigned long long timeElapsed;    /* Always set */
      unsigned long long timeRemaining;  /* Only for VIR_DOMAIN_JOB_BOUNDED */
+    unsigned long long setupTime;      /* length of the SETUP phase */
+    double mbps;                       /* Migration throughput in Mbps */
This series adds a new feature, and we're already in freeze for 1.1.1,
so it would have to be post-freeze.

No problem - we're not in a hurry to get libvirt RDMA support
merged - just wanted to make sure we get these reviews through
so all the code plays nicely.


NACK; this is a change to ABI.  We cannot change the existing struct
size without invalidating apps compiled against an earlier version of
the library (ie. virDomainGetJobInfo is cast in stone).  Rather, we
should be adding new VIR_DOMAIN_JOB_* macros for use in the
virTypedParameters of virDomainGetJobStats().

Acknowledged.

It does point out a bug that can be fixed during freeze: we should fix
our documentation so that virDomainGetJobInfo() refers to the better
virDomainGetJobStats().

Somewhat related: I mentioned in another thread that neither
virDomainGetJob{Info,Stats} nor virDomainGetBlockJobInfo play nicely
with the idea of being able to have multiple jobs running in parallel,
which is necessary if we want to be able to support fleecing of
point-in-time snapshots from multiple clients.  Ultimately, that means I
think we need to enhance the job mechanism so that new jobs return a
handle, then we add new API that can query and abort jobs based on the
handle instead of being limited to only the current job; I guess for
back-compat, it would also help to have an API that sets the current job
out of a set of jobs so that the old APIs are still usable.  If we do an
upgrade of the job API, we would make sure that the new query method
sticks to virTypedParameters in the same was as virDomainGetJobStats().







[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]