wildcard_alert_patterns

[earthgecko]

Added the ability to match wildcard metric patterns in the analyzer.py alert context.

Currently the alert matching is a one to one relationship - metric -> alert setting.

Therefore it is not possible to map wildcarded metrics to a single ALERT and strategy settings e.g.

metrics namespaces:
stats.publishers.alice.rpm
stats.publishers.bob.rpm

Currently we would have to define an individual ALERT for each metric

            ('stats.publishers.alice.rpm', 'smtp', 300),
            ('stats.publishers.bob.rpm', 'smtp', 300),

and an individual strategy for each whether in the SMTP_OPTS or otherwise

        'stats.publishers.alice.rpm': '[email protected]',
        'stats.publishers.bob.rpm': '[email protected]',

This change to using python re and not if alert[0] in metric[1] allows for both the current functionality, but it extends the alerting to allow wildcarded namespaces to one alert setting

            ('stats.publishers.*.rpm', 'smtp', 300),

        'stats.publishers.*.rpm': '[email protected]',

This allows much more granular and configurable alerting

            ('stats.publishers.*.rpm', 'smtp', 300),

# this would probably require another pull request
        'stats.publishers.[a-m].rpm': '[email protected]',
        'stats.publishers.[n-z].rpm': '[email protected]',

This does not break any functionality and fits with the current cache_key = 'last_alert.%s.%s' % (alert[1], metric[1])

[earthgecko]

Further there is no discernible impact on CPU use / load average changing the in to the regular expression match, however the skyline analyzer run_time has increased by 8 seconds from 58 seconds to 66 seconds. I worried that it may be heavier on CPU but is heavier on run_time.

This is based on running against 11890 metrics.