How do leases work with Memcached and McRouter? · Issue #175 · facebook/mcrouter
What we really need is data consistency in our cache layer. Since CAS ops are point-to-point, they won't work for us since we use replicated pools. As the white paper states, the lease feature addresses staleness and thundering herd. That's exactly what we need.
- Would it be possible to publish Facebook's memcached fork? The lease feature is basically incomplete without that. This would be the fastest path to resolution for us.
- Would it be possible to send a PR to @dormando with your custom memcached fork? That way it can make it into open-source memcached and the community can benefit. Is there any reason why Facebook didn't do a PR for this?
- If you are unable to publish the memcached custom fork, our fallback option would be to implement our own version of acquiring a token for optimistic locking. In essence, that defeats the purpose of leveraging leases through mcrouter. It would potentially incur an additional cache roundtrip.
Read full article from How do leases work with Memcached and McRouter? · Issue #175 · facebook/mcrouter
No comments:
Post a Comment