Thanks for the comments, I'm going to send soon a new patch with the suggestions
applied.
About how the patch works:
The first attempt to stop the cascade changes was something like:
@@ -471,7 +512,7 @@ assess_weak_refs(struct ovsdb_txn *txn, struct
ovsdb_txn_row *txn_row)
struct ovsdb_table *table;
struct shash_node *node;
- if (txn_row->old) {
+ if (txn_row->old && !txn_row->new) {
The rationale was: the weak references are linked by UUID, and the UUID never
changes, unless
if the current row is being deleted. So, we in fact didn't need to change the
rows that weak-referenced
the current row if the row wasn't deleted.
It didn't works: we discover that the modified rows were lacking incoming weak
references in
the dst_refs list, so that gives us the idea of "moving" the references in
old->dst_refs to new->dst_refs.
In the original code (dst_refs is created from scratch):
old->dst_refs = all the rows that weak referenced old
new->dst_refs = all the rows that weak referenced old and are still weak
referencing new + rows in the transaction that weak referenced new
In the patch (dst_refs incrementally built):
Old->dst_refs = all the rows that weak referenced old
(Ideally, but expansive to calculate:)
New->dst_refs = old->dst_refs - "weak references removed within this TXN" +
"weak references created within this TXN"
(What was implemented:)
New->dst_refs = old->dst_refs - "weak references in old rows in TXN" + "weak
references in new rows in TXN"
The resulting sets should be equal in both cases.
There we do some more optimizations:
- If we know that the transactions must be successful at some point then,
instead of cloning dst_refs we could just move the elements between
the lists.
- At that point we lost the rollback feature, but we aren't going to need
it anyway (note that we didn't really touch the src_refs part).
- The references in dst_refs must point to new instead than old.
Previously we iterated over all the weak references in dst_refs
to change that pointer, but using an UUID is easier, and prevents
that iteration completely.
Regards,
Esteban
> -----Original Message-----
> From: Ben Pfaff [mailto:b...@ovn.org]
> Sent: sábado, 2 de julio de 2016 14:06
> To: Rodriguez Betancourt, Esteban <esteb...@hpe.com>
> Cc: dev@openvswitch.org
> Subject: Re: [ovs-dev] [PATCH] ovsdb: Weak references performance fix
>
> On Mon, Jun 27, 2016 at 08:07:02PM +0000, Rodriguez Betancourt, Esteban
> wrote:
> > Prevents the cloning of rows with outgoing or incoming weak references
> > when those rows aren't being modified.
> >
> > It improves the OVSDB Server performance when many rows with weak
> > references are involved in a transaction.
> >
> > Signed-off-by: Esteban Rodriguez Betancourt <esteb...@hpe.com>
>
> Great idea. Thanks for working on this!
>
> The new ovs_list_transplant() function takes two "const" pointers to lists but
> it modifies both lists. I don't think it makes sense for them to be const.
>
> The new ovs_list_transplant() function has a name that doesn't give much of
> a idea of what it does. I think that really it appends everything in 'src'
> to 'dst',
> although those are not great names since both lists are modified. Maybe
> ovs_list_push_back_all() would be a better name.
>
> I think that the new ovs_list_transplant() function can be implemented in
> terms of ovs_list_splice(), as:
> ovs_list_splice(dst, &src->next, src); I see a couple of existing
> uses of
> ovs_list_splice() to do that, so it's probably best to convert them to use the
> new function for consistency.
>
> Since add_weak_ref() doesn't use its first argument now, please delete the
> parameter entirely instead of marking it OVS_UNUSED.
>
> In add_weak_ref(), instead of using memcpy() to copy a struct uuid, please
> use an ordinary assignment with =.
>
> I don't think that this loop in ovsdb_txn_update_weak_refs() needs to be
> the _SAFE variant: it appears to me that it iterates on ->src_refs and
> ->src_node but only modifies ->dst_refs and ->dst_node:
> > + LIST_FOR_EACH_SAFE (weak, next, src_node, &txn_row->new-
> >src_refs) {
> > + /* dst_row MUST exist */
> > + dst_row = CONST_CAST(struct ovsdb_row *,
> > + ovsdb_table_get_row(weak->dst_table, &weak->dst));
> > + ovs_list_insert(&dst_row->dst_refs, &weak->dst_node);
> > + }
>
> It's taking me some thought to convince myself that this new version is as
> correct as the previous version. Do you have any arguments or explanations
> to help me out?
>
> Thanks,
>
> Ben.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
