Need up optimizing php-driver extension.
Philip Z
Our setup:
- PH 7.1.x
- Apache 2.2.x with mod_php
- Cassandra 2.1.x (4 in the cluster)
Historically, we've used YACassandra PDO extension. We're migrating over to datastax/php-driver and having _a lot_ of issues with it.
In short: when we put our server into our production cluster, Apache completely halts the system, due to exhaustion of all available TCP sockets. From the best we can tell, this is from connections to our Cassandra cluster: 20-30 workers, 4 in the cluster, that's 80 - 120 active TCP connections, just from 20-30. Can't imagine 100+
We run a service that gets hundreds, thousands of hits per minute. We chose Cassandra for our backend because we need those fast "write" and "leave" connections. Milliseconds here.
This worked great with YACassandra PDO - quick write and out.
Now, datastax/php-driver is taking things over and adding "features" which are really causing us some major headaches. We don't need persistent connections. We don't see sessions. We don't need schema. We don't want to build a connection pool. We just need to throw a quick "insert" and leave.
When we disable persistent connection, and schemas, connection start up time dramatically increases... How can we get back to the old way of *fast* write and leave?
Philip