Redis: Connection Leak if Connect or ConnectAsync called from multiple threads simultaneously

[JonCole]

We have a customer who, under load testing, noticed connections to Redis spiking significantly. After digging into the dump of the process, we could see that many ConnectionMultiplexer objects exist in the heap. We also noticed that there were many threads calling Connect at the same time.

If multiple threads make it past the if (_connection == null) check, then there will be a connection leak for each additional thread that makes it through that check.

[Tratcher]

Relevant code: https://github.com/aspnet/Caching/blob/1.0.0/src/Microsoft.Extensions.Caching.Redis/RedisCache.cs#L146-L162

@JonCole what version of the packages are they using? Do we need to backport this fix to 1.1.x or 1.0.x?

/cc: @muratg

[JonCole]

@Tratcher - looks like they are on 1.1.0.21115.

[muratg]

Putting this in 1.1.1 milestone so that it gets tracked with the rest of 1.1.1 candidates. We should also look into whether we want this in 1.0.4 as well.

[JunTaoLuo]

Fixed in 3522082. Will need to port this to 1.1.1 branch when available for release.

[runxc1]

Would this only happen on startup? I am noticing about once a day or so on a site with 1 million hits a day that Redis will run out of connections but just for 5-10 minutes.

I am using the distributedCache inside a singleton service that uses both the memory and distributed cache for short/medium term caching.

[muratg]

(1.0.4 tracking bug is here: #257)