Support Alibaba pseudo-cluster configurations

[mgravell]

1: don't treat trivial clusters as clusters - Alibaba uses this config 2: report synchronous failures immidiately and accurately

fix #2642

For context, Alibaba reports from cluster nodes as 2 nodes:

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa {proxy endpoint} myself,master - 0 0 0 connected 0-16383
bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb {proxy endpoint} slave aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 0 0 0 connected

which confuses things somewhat; this isn't a valid/normal cluster configuration, so: ignore it!

Other relevant metadata:

from config:

databases=256

from info:

# Cluster
cluster_enabled:1
databases:256
nodecount:2

[NickCraver]

How does this affect single node clusters? For example, even in Azure hosting people start with a single cluster node in a lot of cases which makes scaling to n easier later...but they don't support SELECT because they are cluster. I worry about the assumption of 2+ here being needed.

[mgravell]

Untested and it is a good question, but: if the worst side effect here is that we get a server-side "nope" rather than a client-side "nope" in genuine single-server clusters: I'm comfortable with that...?

[NickCraver]

I used SELECT as an example, but I think this had many more side-effects such as not detecting when the cluster changes and such. If for example we were viewing this in Opserver…it'd be wrong. Any code basing off cluster or not on top of us has the same downstream fun. Lots of potential there.

I'd vote we should not make such an assumption based on node count - likely better for the server-side to not represent itself as a cluster here.

[mgravell]

@NickCraver thanks for keeping me honest; refactored to use the "databases" entry as reported from config; this data is also available under info cluster, but I don't propose to add an extra info call unless this doesn't work reliably

[yangbodong22011]

@mgravell Thank you for your quick fix. We have verified the 2.7.20 version of MyGet and it works very well. When do you expect to release it to NuGet? After that, we will push users to upgrade.

[mgravell]

Thanks for confirming. I'll discuss release in our weekly sync tomorrow (20th) - assuming there's no known blockers, I expect we'll release to NuGet either end-of-day 20th, or very shortly after