fix reporting of dead upstream statsd metrics

[matthew-lucidchart]

we have statsd servers sending stats to other statsd processes. when the upstream servers die, we don't want to persist the metrics.

[mrtazz]

What does this fix exactly? It seems like it's changing the behaviour from setting it to 0 for the internal stats, to setting it to zero if the metrics string is empty?

[coykitten]

Is this something that can be fixed by setting deleteIdleStats true?
On Feb 6, 2015 1:47 PM, "Daniel Schauenberg" notifications@github.com
wrote:

What does this fix exactly? It seems like it's changing the behaviour from
setting it to 0 for the internal stats, to setting it to zero if the
metrics string is empty?

—
Reply to this email directly or view it on GitHub
#442 (comment).

[matthew-lucidchart]

Each server in my environment has a statsd process running with a repeater on it. They all feed to a collector statsd that aggregates all stats together and sends them to graphite. The problem is when the collector saves metrics of servers that have gone away. It keeps reporting them.... forever.... This fix makes sure that each instance of statsd only saves its own set of metrics as time goes on.

It cannot be fixed by deleteIdleStats.

[mrtazz]

Ah ok that makes sense. But still the branch should be checking for the full configured metrics string for the local instance and not the empty string, right?

[matthew-lucidchart]

The diff doesn't show it well, but that's exactly what's happening. If you open the full file, and track those two empty variables, you'll see what I'm talking about.

When the config is parsed, those two variables are set, but on global variables, not local to the config update function. My setting them to empty string is just to specify global scope. Then, I access them later, as found in the diff.