Add the :nearest_slave role for Sentinel mode #588
Conversation
|
The 1.8.7 failures appear to be present in master, as well, running them locally with |
|
It looks like the issue is the difference between test_unit 3.1.5 and 3.1.7. Locking test_unit to 3.1.5 causes the suite to pass. |
|
Would you mind submitting the version pinning separately? Then we can already merge that regardless of whether this new feature is accepted. |
|
Certainly! You'll find it at #593 |
cheald
force-pushed
the
nearest-slave
branch
2 times, most recently
from
998fb73
to
14b86ef
Compare
byroot
force-pushed
the
master
branch
6 times, most recently
from
efe158b
to
dd4b6fe
Compare
|
@cheald Can you update your PR please? One of our redis replica is in US Eastern, and one of our Rails instance in US Western region... The delay between two instances are 60ms~, but ideally our Rails application should only connect to its region's redis |
This will cause the client to measure roundtrip latency to each slave and select the slave with the lowest latency. The intent for this is to enable sentinel-managed clusters of servers for which eventually-consistent reads are acceptable, but to maintain minimum latencies between any individual client-slave pair. The case I did this for is is shared web application caching across multiple datacenters, where you would not want Redis to connect to a slave in another datacenter, but you would want all datacenters to share a cache. Remove trailing comma from client creation options; should fix 1.8 builds If we can't get the role, use a translated role Ensure that ping test clients are always disconnected after use. Don't assume that a good slave was found.
|
Looks like a simple merge. I've rebased it to master. I don't currently have a production use case for this, but let me know if it works for you. |
|
Rubocop has a few complaints. The code looks sane to me, but @supercaracal if you have time and are willing to give this a second look, it would be much appreciated. |
|
I'll see what I can do to make Rubocop happy. |
There was a problem hiding this comment.
CI is failed but these test cases are not related the PR. We can ignore them. #930
This will cause the Sentinel client, upon connection, to measure roundtrip latency to each slave and select the slave with the lowest latency. The intent for this is to enable sentinel-managed clusters of servers for which eventually-consistent reads are acceptable, but to maintain minimum latencies between any individual client-slave pair.
The case I did this for is is shared web application caching across multiple datacenters, where you would not want Redis to read from a slave in another datacenter, but you want to be able to share the cache across the datacenter.
(If this would be more appropriate as a separate gem or something, I'm happy to do that, but there didn't seem to be enough exposed API to accomplish it without a PR.)