io: reduce unknown error code log level

[mattklein123]

2019-07-29T17:11:16.00568 [2019-07-29 17:11:16.005][26039][error][misc] [external/envoy/source/common/network/io_socket_error_impl.cc:29] Unknown error code 111 details Connection refused

I saw the previous log line on a production box and I think this should
be a debug log condition.

Risk Level: Low
Testing: Existing tests
Docs Changes: N/A
Release Notes: N/A

[danzh2010]

I'm fine with either logging levels, but shouldn't ECONNREFUSED be an error you want to propagate to the application? If so, instead of decrease the log level, we should add such error code to the enum.

[mattklein123]

I don't actually know where this is coming from, I just thought it doesn't make sense to have this be an error log since it might heavily spam the logs?

[danzh2010]

I don't actually know where this is coming from, I just thought it doesn't make sense to have this be an error log since it might heavily spam the logs?

This IO error is not affected by input. So it shouldn't spam the logs if upper level handles it appropriately, i.e. no un-limited number of retries.

[mattklein123]

This IO error is not affected by input. So it shouldn't spam the logs if upper level handles it appropriately, i.e. no un-limited number of retries.

Ultimately I agree, but I still think it's safer to reduce the log level here and we can deal with unhandled cases as they come up?

[danzh2010]

This IO error is not affected by input. So it shouldn't spam the logs if upper level handles it appropriately, i.e. no un-limited number of retries.

Ultimately I agree, but I still think it's safer to reduce the log level here and we can deal with unhandled cases as they come up?

SGTM

[alyssawilk]

Optional: I'd lean towards just adding econnrefused to the known errors, and continuing to log things we don't expect at error level, but this is a viable fix too.

[mattklein123]

I'm going to land this and we can look at what/where is dealing with conn refused and also that. I just think this is too high of a risk of log spam by accident.