syncano / codebox Goto Github PK

For some reason stdin/out/err communication in Docker seems to be ~8 times slower than communicating with container through TCP.

If CPU Weight > 1, scale memory as well.
Reserve memory for it that weight and pass it in context to set it dynamically for first run.

Requires wrapper that sets v8 option to have stracktrace working.
Could standarize wrapper here if needed and maybe use /init, /run?

Wrapper (muxed) would simply call subprocess and then communicate with it through unix socket. Each new call -> new call to unix socket.

• disable stdout/stderr
• require http response to be returned?
• on timeout, do not accept any new connections and then kill container.
• add state to container (starting/ready/terminating)

when selecting worker

connect to node through tcp or unix socket

Container has name, instance, version
When replacing, propagate cleanup to same name, instance

Delete trigger - cleanup

Redis based LRU

GET (incr if autorefresh)

GET key
ZADD client_key XX CH [incr] score key
if result[0] == null && result[1] {
    ZREM client_key key
}

SET (incr if autorefresh)

SET key value PX age
ZADD client_key INCR score key
ZRANGE client_key 0 -(max_size+1)
ZREMRANGEBYRANK client_key 0 -(max_size+1) 

for key in result[-2] {
    DEL key
}

• runs node wrapper
• sets memory limit
• disables/enables muxing stdout/stderr