[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: LAST_CHECKPOINT and save/restore/migrate compatibility (was Re: [Xen-devel] [PATCH 3 of 4] libxc: provide notification of final checkpoint to restore end)
On Fri, Apr 1, 2011 at 12:58 PM, Ian Campbell <Ian.Campbell@xxxxxxxxxxxxx> wrote: > On Fri, 2011-04-01 at 12:02 +0100, George Dunlap wrote: >> On Mon, Sep 6, 2010 at 11:03 AM, Ian Campbell <ian.campbell@xxxxxxxxxx> >> wrote: >> > # HG changeset patch >> > # User Ian Campbell <ian.campbell@xxxxxxxxxx> >> > # Date 1283766891 -3600 >> > # Node ID bdf8ce09160d715451e1204babe5f80886ea6183 >> > # Parent 5f96f36feebdb87eaadbbcab0399f32eda86f735 >> > libxc: provide notification of final checkpoint to restore end >> > >> > When the restore code sees this notification it will restore the >> > currently in-progress checkpoint when it completes. >> > >> > This allows the restore end to finish up without waiting for a >> > spurious timeout on the receive fd and thereby avoids unnecessary >> > error logging in the case of a successful migration or restore. >> > >> > In the normal migration or restore case the first checkpoint is always >> > the last. For a rolling checkpoint (such as Remus) the notification is >> > currently unused but could be used in the future for example to >> > provide a controlled failover for reasons other than error >> >> Unfortunatley, this introduces a backwards-compatibility problem, >> since older versions of the tools never send LAST_CHECKPOINT. > > Out of interest what actually happens? (I think you mentioned something > about failing when restoring from a file?) Steps to reproduce: * Install last release of XenServer (based on Xen 3.4) * Install a VM, suspend it. * Upgrade to development version of XenServer (based on 4.1) * Attempt to resume VM; resume fails. libxc output is below: Apr 1 14:27:29 localhost xenguest: xc: detail: xc_domain_restore start: p2m_size = fefff Apr 1 14:27:33 localhost xenguest: xc: error: Max batch size exceeded (1970103633). Giving up.: Internal error Apr 1 14:27:33 localhost xenguest: xc: error: error when buffering batch, finishing (90 = Message too long): Internal error Apr 1 14:27:33 localhost xenguest: xc: detail: Restore exit with rc=0 I had added debugging to print out the "batch" values read, and confirmed that this is *after* the page transfer and everything else was done, and after the "0" entry indicating the end of a checkpoint. So it's reading through one full checkpoint, and then reading some garbage value and failing. Initializing "ctx->last_checkpoint" to 1 makes the problem go away. Ah -- hmm. You know what I bet -- xc_domain_save() doesn't write the qemu checkpoint, but xc_domain_restore() reads it (again because of REMUS). Libxl handles this properly, but making xapi write in such a way that xc_domain_restore() could read it properly turned out to be non-trivial; making xc_domain_restore() not read the qemu file was the simplest solution. But of course, that means that there's still data in the file after you've finished a "checkpoint", so the next read succeeds in reading in the qemu save file. Not surprising that it doesn't look right. :-) > My intention at the time was that in the absence of a LAST_CHECKPOINT > things would continue to behave as before e.g. you would see the > spurious timeout on the receive fd but still restore correctly. > > Perhaps the associated changes to the O_NONBLOCK'ness of the fd at > various stages is what broken this for the non-Remus migrations? > >> Would it make sense to invert the logic on this -- i.e., default to >> only one checkpoint, and send "MORE_CHECKPOINTS" if it expects this >> *not* to be the last (either by expecting the checkpoint() function to >> do it, or by sending it from xc_domain_save)? > > I'd quite like to understand what I broke and investigate the > trivialness (or otherwise) of a fix, but this change would probably > independently make sense too. > >> Now that 4.1 is out, we have to consider version compatibilities. But >> we should be able to make it so that there would only be if: >> 1) You're running REMUS > > Is it the case that Remus only cares about checkpointing between like > versions of the toolstack? > >> 2) One of your toolstacks is 4.1.0, and one is not. >> >> That seems like it shouldn't be too bad. >> >> Thoughts? >> >> -George >> >> > >> > Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx> >> > Acked-by: Brendan Cully <brendan@xxxxxxxxx> >> > >> > diff -r 5f96f36feebd -r bdf8ce09160d tools/libxc/xc_domain_restore.c >> > --- a/tools/libxc/xc_domain_restore.c Mon Sep 06 10:54:51 2010 +0100 >> > +++ b/tools/libxc/xc_domain_restore.c Mon Sep 06 10:54:51 2010 +0100 >> > @@ -42,6 +42,7 @@ struct restore_ctx { >> > xen_pfn_t *p2m; /* A table mapping each PFN to its new MFN. */ >> > xen_pfn_t *p2m_batch; /* A table of P2M mappings in the current >> > region. */ >> > int completed; /* Set when a consistent image is available */ >> > + int last_checkpoint; /* Set when we should commit to the current >> > checkpoint when it completes. */ >> > struct domain_info_context dinfo; >> > }; >> > >> > @@ -765,6 +766,11 @@ static int pagebuf_get_one(xc_interface >> > // DPRINTF("console pfn location: %llx\n", buf->console_pfn); >> > return pagebuf_get_one(xch, ctx, buf, fd, dom); >> > >> > + case XC_SAVE_ID_LAST_CHECKPOINT: >> > + ctx->last_checkpoint = 1; >> > + // DPRINTF("last checkpoint indication received"); >> > + return pagebuf_get_one(xch, ctx, buf, fd, dom); >> > + >> > default: >> > if ( (count > MAX_BATCH_SIZE) || (count < 0) ) { >> > ERROR("Max batch size exceeded (%d). Giving up.", count); >> > @@ -1296,10 +1302,23 @@ int xc_domain_restore(xc_interface *xch, >> > goto out; >> > } >> > ctx->completed = 1; >> > - /* shift into nonblocking mode for the remainder */ >> > - if ( (flags = fcntl(io_fd, F_GETFL,0)) < 0 ) >> > - flags = 0; >> > - fcntl(io_fd, F_SETFL, flags | O_NONBLOCK); >> > + >> > + /* >> > + * If more checkpoints are expected then shift into >> > + * nonblocking mode for the remainder. >> > + */ >> > + if ( !ctx->last_checkpoint ) >> > + { >> > + if ( (flags = fcntl(io_fd, F_GETFL,0)) < 0 ) >> > + flags = 0; >> > + fcntl(io_fd, F_SETFL, flags | O_NONBLOCK); >> > + } >> > + } >> > + >> > + if ( ctx->last_checkpoint ) >> > + { >> > + // DPRINTF("Last checkpoint, finishing\n"); >> > + goto finish; >> > } >> > >> > // DPRINTF("Buffered checkpoint\n"); >> > diff -r 5f96f36feebd -r bdf8ce09160d tools/libxc/xc_domain_save.c >> > --- a/tools/libxc/xc_domain_save.c Mon Sep 06 10:54:51 2010 +0100 >> > +++ b/tools/libxc/xc_domain_save.c Mon Sep 06 10:54:51 2010 +0100 >> > @@ -1616,6 +1616,20 @@ int xc_domain_save(xc_interface *xch, in >> > } >> > } >> > >> > + if ( !callbacks->checkpoint ) >> > + { >> > + /* >> > + * If this is not a checkpointed save then this must be the first >> > and >> > + * last checkpoint. >> > + */ >> > + i = XC_SAVE_ID_LAST_CHECKPOINT; >> > + if ( wrexact(io_fd, &i, sizeof(int)) ) >> > + { >> > + PERROR("Error when writing last checkpoint chunk"); >> > + goto out; >> > + } >> > + } >> > + >> > /* Zero terminate */ >> > i = 0; >> > if ( wrexact(io_fd, &i, sizeof(int)) ) >> > diff -r 5f96f36feebd -r bdf8ce09160d tools/libxc/xg_save_restore.h >> > --- a/tools/libxc/xg_save_restore.h Mon Sep 06 10:54:51 2010 +0100 >> > +++ b/tools/libxc/xg_save_restore.h Mon Sep 06 10:54:51 2010 +0100 >> > @@ -131,6 +131,7 @@ >> > #define XC_SAVE_ID_TMEM_EXTRA -6 >> > #define XC_SAVE_ID_TSC_INFO -7 >> > #define XC_SAVE_ID_HVM_CONSOLE_PFN -8 /* (HVM-only) */ >> > +#define XC_SAVE_ID_LAST_CHECKPOINT -9 /* Commit to restoring after >> > completion of current iteration. */ >> > >> > /* >> > ** We process save/restore/migrate in batches of pages; the below >> > >> > _______________________________________________ >> > Xen-devel mailing list >> > Xen-devel@xxxxxxxxxxxxxxxxxxx >> > http://lists.xensource.com/xen-devel >> > > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@xxxxxxxxxxxxxxxxxxx > http://lists.xensource.com/xen-devel > _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |