[openstack-dev] [Swift] 404 on geo-replicated clusters with write affinity
clay.gerrard at gmail.com
Wed May 17 23:22:21 UTC 2017
On Wed, May 17, 2017 at 3:47 PM, Bruno L <teolupus.ext at gmail.com> wrote:
> I see multiple bugs in launchpad that are related to this issue.
AFAIK, only one bug for this issue is still open, and has the most recent
thoughts added from the Forum
write a brief summary of the changes proposed,
I think the skinny of what's in the bug report is "make more backend DELETE
requests to handoffs".
Personally I was coming around to the idea that an explicit configurable
(i.e. similar to "request_node_count" for GET) would be easy to reason
about and give us a lot of flexibility (it would pair well per-policy
configuration WIP https://review.openstack.org/#/c/448240/). It's possible
this could be implicit using some heuristic over the sort order of
primaries in the ring - but I think it'd be whole 'nother thing, and could
be added later as an "auto" sort of value for the option (i.e. workers =
[N|auto], or "replicas + 2" sort of syntax).
Additionally, it's been pointed out various times that collecting
X-Backend-Timestamp from the responses would allow for further reasoning
over the collected responses in addition to just the status codes (similar
to WIP for GET https://review.openstack.org/#/c/371150/ ) - but I'm
starting to think that'd be an enhancement to the extra handoff DELETE
requests rather than an complete alternative solution.
I don't think anyone really likes the idea of blindly translating a
majority 404 response on DELETE to 204 and calling it a win - so
unfortunately the fix is non-trival. Glad to see you're interested in
getting your teeth into this one - let me know if there's anything I can do
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the OpenStack-dev