peillis / memcachir Goto Github PK

Hello!

Telemetry is getting traction now that it's included in Phoenix and Ecto out-of-the-box. What's your stance on emitting telemetry events from this library? Would you welcome contributions?

It would allow users of this library to capture important metrics and send them to their monitoring solution. This way, we save everyone the time to instrument their own usage of this library.

Stephan

Hi:

thank for your effort to develop/maintain memcachir,

it's the best memcache client of elixir as I know,

and, is it possible to handle failed node to libring?
when a node failed, remove it from librings list,
and when the node back online, readd it to librings list.

thanks.

We recently upgraded to the latest memcachir version from one of our local forks, and immediately found a connection leak. I'm not entirely certain of the root cause, but we witnessed monotonic increases in connection counts on our elasticache cluster (eventually terminated when we hit process ulimits). We're currently running on elixir 1.7.2 in docker (with a distillery release).

My guess is the root cause might be due to some nuance in the supervision of the health check. Or perhaps the pool isn't tearing down the memcachex connection processes correctly so that they actually close their sockets.

Here's our fork for comparison: https://github.com/Frameio/memcachir

We noticed some odd behaviour in production: under certain circumstances, memcachir lost all nodes in our app. But they were running just fine.

Turned out, that if there is a connection timeout, memcachir thinks all nodes are gone:

cc @michaeljguarino