Default the failure limit to 1 as zero is an invalid value

[rossriley]

Fixes: bolt/bolt#7357

The default of zero throws an invalid argument error despite the Memcached docs indicating that this should be a valid value.

[GwendolenLynch]

Are you sure? From http://php.net/manual/en/memcached.constants.php

Memcached::OPT_SERVER_FAILURE_LIMIT
Specifies the failure limit for server connection attempts. The server will be removed 
after this many continuous connection failures.

Type: integer, default: 0.

[GwendolenLynch]

Also, I'll grab the CI failures in the morning if that is OK … obviously not related to this.

[rossriley]

Yes, sorry should have mentioned, there's a bug in the php docs.

To verify:

echo '<?php $m = new Memcached; $m->setOption(21,0);' | php

And you'll get:

PHP Warning: Memcached::setOption(): error setting memcached option: INVALID ARGUMENTS in - on line 1

Change the zero to 1 or more and all goes through ok.

[codecov-io]

Codecov Report

Merging #3 into 1.1 will not change coverage.
The diff coverage is 100%.

Impacted Files	Coverage Δ	Complexity Δ
src/Handler/Factory/MemcachedFactory.php	93.37% <100%> (ø)	52 <0> (ø)	⬇️

[GwendolenLynch]

OK, this took a bit to track down what is going on upstream.

What I see happening is that libmemcached requires MEMCACHED_BEHAVIOR_SERVER_FAILURE_LIMIT to be a non-zero integer … easy enough, right?!

But in the current stable, we have

MEMC_SESSION_INI_ENTRY("server_failure_limit", "0", OnUpdateLongGEZero, server_failure_limit)

Intreagued yet? I was 😄

So then looking at the globals initialization function it seems they default to 1.

Soooooo … my question now becomes, is 5 too high a value for the default?

[bobdenotter]

I think 1 could work just as well, as a default. Main improvement is no breakage, and if people need a different value, they can just set it.