Re: [tarantool-patches] [PATCH v2 2/3] Syslog remote destination test

[Vladimir Davydov <vdavydov.dev@gmail.com>]

On Tue, Jul 17, 2018 at 05:15:24PM +0300, Olga Arkhangelskaia wrote:
> >>  test/app-tap/syslog_remote.test.lua | 37 +++++++++++++++++++++++++++++++++++++
> >>  1 file changed, 37 insertions(+)
> >>  create mode 100755 test/app-tap/syslog_remote.test.lua
> >
> >In review to v1 I asked you to fold this test in box-tap/cfg.
> >Any reason why this couldn't/shouldn't be done?
> Sorry, I did not notice that comment. However, why in  box-tap/cfg.test.lua and not separately?

Hmm, good question. For one thing, what's you're testing here is not a
part of the application server, but is rather relevant to the box API so
the test should live in box/ or box-tap/. Since box-tap/cfg.test.lua is
supposed to test box.cfg() and already has all the infrastructure for
starting a new tarantool instance (I mean, run_code), it feels like a
good choice.

Anyway, we don't usually create a new test file for each bug/feature -
that'd be an overkill. We try to group tests by subsystem. I suppose, in
your particular case it would be acceptable to introduce a new test
file, but it should test both unix and udp sockets then IMHO, which
means that you need the run_code() from box-tap/cfg.test.lua, so why not
just fold the test cases there - they aren't that big, are they.

> 
> >
> >> 
> >> diff --git a/test/app-tap/syslog_remote.test.lua b/test/app-tap/syslog_remote.test.lua
> >> new file mode 100755
> >> index 000000000..01a75fa64
> >> --- /dev/null
> >> +++ b/test/app-tap/syslog_remote.test.lua
> >> @@ -0,0 +1,37 @@
> >> +#!/usr/bin/env tarantool
> >> +
> >> +local tap = require('tap')
> >> +local socket = require('socket')
> >> +local os = require('os')
> >> +local test = tap.test("syslog_remote")
> >> +local log = require('log')
> >> +local errno = require('errno')
> >> +
> >> +local addr = '127.0.0.1'
> >> +local port = 1000 + math.random(32768)
> >
> >No, that's not what I meant. Using a random port like that won't save
> >your from conflicts. You should retry until 'bind' succeeds. 
> And how to choose number of attempts? Ten is enough?

I guess so. Even if you have half of the ports busy, after ten attempts
the probability of repeatedly choosing a busy port is less than 0.1%,
which is acceptable, I think.

Also, I asked you to test invalid syslog destination (non-existing unix
socket or host address). Please do so as to cover the error paths.