#48 socket leak when connecting to non-existent exchange.

[psilva]

Reported by Jun on Monday... He noticed that sr_shovel, when connecting to a broker and attempting to bind to a non-existent exchange, will continously retry, likely not closing the previous attempts. this causes socket descriptors on the broker to be consumed eventually leading to system-wide exhaustion on the broker.

(incident 2016/08/15 02h00 Z... )

This is likely a problem in sr_consumer or sr_amqp that would affect all components.

[Murray Rennie]

When sr_shovel attempts to connect to a non-existant exchange and fails, it then defaults to the default exchange. This is the "(AMQP Default)" exchange, and this is what the queue is attached to.

What I found when I tried to conenct to a non-existant exchange was that it just sat there after failing to connect probably because it had then connected to the default exchange.

[Murray Rennie]

I tested shovelling to a nonexistant exchange. No increase in sockets.

[Murray Rennie]

While I had a shovel shovelling from one broker to another I deleted the receining broker. No difference

[Murray Rennie]

Dried killing an exchange which had one shovel feeding it and another feeding off of it. No new sockets.

[Murray Rennie]

I set up a shovel that reads from a non-existant exchange and writes to a non-existent exchange. The log show it retrying to connect to the non-existant exchange once every 5 seconds. But the socket descriptors are not increasing.

[psilva]

chat with Jun to see if on_poll scripts are causing the issue (try using his...)

[Murray Rennie]

sr_shovel doesn't use on_poll scripts

[psilva]

now: https://github.com/MetPX/sarracenia/issues/53