Add locking to FetchAuthTokenCache

[ericnorris]

Per #225, filing an additional ticket for this feature request. Thanks again!

Is your feature request related to a problem? Please describe.

The SysVCacheItemPool class points out that calling code should handle race conditions, but the FetchAuthTokenCache class, which is used by the google-cloud-php library when the authCache parameter is passed in (see here), does not do any locking.

This means that users that follow code examples such as the following (from here):

$authCache = new SysVCacheItemPool();
$sessionCache = new SysVCacheItemPool([
    // Use a different project identifier for ftok than the default
    'proj' => 'B'
]);

$spanner = new SpannerClient([
    'authCache' => $authCache
]);

...may run into race condition issues unknowingly.

Describe the solution you'd like

Either of the two:

• FetchAuthTokenCache adds a locking mechanism to safely use the passed inCacheItemPoolInterface
• I file an issue in the google-cloud-php repository to request it use a cache with locking built in.

[bshaffer]

I am trying to fix this for v2.0 but I can't seem to recreate the error. I can add some locking to ftok, but I'd like to confirm the fix. I've done the following:

> while [ true ]; do php -r "ftok('/tmp/foo', 'A');"; done &
> while [ true ]; do php -r "ftok('/tmp/foo', 'B');"; done

I was hoping this would eventually throw the error but it hasn't happened yet. I've also tried this:

use Google\Cache\SysVCacheItemPool;

while (true) {
    $rand = rand() % 256;
    $key = chr($rand);
    $var = 'varname' . $key;
    $$var = new Google\Cache\SysVCacheItemPool(['proj' => $key]);
}

[indreka]

This is probably reproducible with similar pcntl_fork test that you added in #637 @bshaffer
We stumbled upon SysVCacheItemPool intermittent issues in our project also, which triggered switch to file based cache (which had its own race condition issue).

[bshaffer]

Thank you @indreka ! This is a good idea. I've updated the tests to include our other caches as well, and will look into the shm_put_var(): Not enough shared memory left error!

[bshaffer]

This (and hopefully the "Not enough shared memory" error) should now be resolved by #640