On 30/10/2023 16:09, Peter Xu wrote: > On Mon, Oct 30, 2023 at 11:13:55AM -0400, Peter Xu wrote: >>> Perhaps it is easy to wrap the checkpoint tracepoint in its own function to >>> allow extension of something else e.g. add the timestamp (or any other data >>> into >>> the checkpoints) or do something in a particular checkpoint of the >>> migration. >>> The timing function from qemu would give qemu own idea of time (directly >>> correlable with the downtime metric that applications consume) which would >>> be >>> more consistent? though I am at too minds on this whether to rely on tracing >>> stamps or align with what's provided to applications. >> >> Yes it should be more consistent. Let me add them into the checkpoints. > > I just noticed dst qemu doesn't have downtime_start to offset.. Only > offset the src downtime checkpoints will make it less consistent. I'll > leave this one for later. Posting a new version soon. > Argh, yes.
Somehow was thinking in source only -- let's leave for later then.
