Cache and expire metrics for prometheus output #2016
Conversation
| type PrometheusClient struct { | ||
| Listen string | ||
| Listen string | ||
| ExpirationInterval int64 `toml:"expiration_interval"` |
There was a problem hiding this comment.
change this type to internal.Duration, that way it can reprepsented like expiration_interval = "10s"
There was a problem hiding this comment.
you'll need to import github.com/influxdata/telegraf/internal
| @@ -171,7 +174,12 @@ func (p *PrometheusClient) Write(metrics []telegraf.Metric) error { | |||
| "key: %s, labels: %v,\nerr: %s\n", | |||
| mname, l, err.Error()) | |||
| } | |||
| p.metrics[desc.String()] = metric | |||
|
|
|||
| metricWithTime := &MetricWithExpiration{ | |||
There was a problem hiding this comment.
don't declare a tmp var for metricWithTime, just do p.metrics[desc.String()] = &MetricWithExpiration{ ....
|
|
||
| sync.Mutex | ||
| } | ||
|
|
||
| var sampleConfig = ` | ||
| ## Address to listen on | ||
| # listen = ":9126" | ||
|
|
||
| ## Interval to expire metrics and not deliver to prometheus, 0 == no expiration | ||
| # expiration_interval = 0 |
There was a problem hiding this comment.
I would say this should default to "60s", since the current behavior of the plugin is to expire metrics.
There was a problem hiding this comment.
you'll also need to set the value in the struct that gets returned in the init() function
There was a problem hiding this comment.
I think this is what you're looking for here, let me know if I did it wrong
|
@sparrc I updated the changelog on this; is there anything else you'd like me to do? |
|
looks good but the changelog update is in the wrong section, you should rebase and put it into release 1.2 |
|
Running 1.2.1 and still seeing same behavior (as detailed in #1775). |
Required for all PRs:
This implements the solution we discussed in #1775