Kapacitor input plugin #2031
Kapacitor input plugin #2031
Conversation
|
/cc @danielnelson |
rossmcdonald
force-pushed
the
ross-kapacitor-input
branch
from
e997cc0 to
71d6acf
Compare
There was a problem hiding this comment.
Could also be good to add support for the normal TLS overrides, isn't a must for now though. https://github.com/influxdata/telegraf/blob/master/plugins/outputs/influxdb/influxdb.go#L31
| func TestErrorHandling404(t *testing.T) { | ||
| badServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { | ||
| if r.URL.Path == "/endpoint" { | ||
| _, _ = w.Write([]byte(basicJSON)) |
There was a problem hiding this comment.
Why define all this JSON since we take the else branch?
| // There were errors, so join them all together as one big error. | ||
| errorStrings := make([]string, 0, len(errorChannel)) | ||
| for err := range errorChannel { | ||
| errorStrings = append(errorStrings, err.Error()) |
There was a problem hiding this comment.
We should be able to use acc.AddError(
|
We need to add stats for: |
|
@danielnelson can you look at added these other two metrics? |
danielnelson
force-pushed
the
ross-kapacitor-input
branch
2 times, most recently
from
9a864c4 to
da5d8e1
Compare
|
@nathanielc Do you know where I can get the stats @nhaugo requested? Also, can you check the metric renaming I did to be consistent with Telegraf's selfstats? There is one difference due to a bug in the internal plugin #2671 |
|
Just throwing in my own $0.02 because we've been bitten by this issue several times in the past with other plugins, but why not just pass through the fields directly, without unmarshaling onto a struct and then marshaling back out again? Unmarshaling onto a struct means that if kapacitor changes its metrics in any way, such as adding or removing, telegraf has to be updated. And if you want to use a different version of kapacitor than the one telegraf is coded for, then you're out of luck. |
|
@danielnelson Those stats will be added in v1.3 of Kapacitor, so to @phemmer's point we either need to change the plugin to be a pass through or update this plugin when they release. Also I am willing to make the name change in Kapacitor to make it consistent if that helps. |
There was a problem hiding this comment.
@danielnelson I am realizing I haven't ever reviewed this code, I took a quick look and added some comments. Sorry for the late review. Let me know how I can help move things forward.
| for _, obj := range *s.Kapacitor { | ||
|
|
||
| // Strip out high-cardinality or duplicative tags | ||
| excludeTags := []string{"host", "cluster_id", "server_id"} |
There was a problem hiding this comment.
This struct me as odd, may this behavior can be triggered via config instead of assuming that this information is unnecessary.
There was a problem hiding this comment.
What about the comment, is it correct? We should, of course, make it hard to produce high-cardinality.
plugins/inputs/kapacitor/README.md
Outdated
| - num_tasks, integer | ||
| - kapacitor_edges | ||
| - collected, float | ||
| - emitted, float |
There was a problem hiding this comment.
Both of these stats are integers
plugins/inputs/kapacitor/README.md
Outdated
| - collected, float | ||
| - emitted, float | ||
| - kapacitor_ingress | ||
| - points_received, float |
There was a problem hiding this comment.
This should be an integer.
plugins/inputs/kapacitor/README.md
Outdated
| - query_errors, float | ||
| - tags_defaulted, float | ||
| - warns_triggered, float | ||
| - write_errors, float |
There was a problem hiding this comment.
I believe everything in the kapacitor_nodes section is an integer.
|
I'm reluctant to do passthrough in general because that would mean new fields just appear that are undocumented, and if any additional processing needs to be done on them we might get stuck by the need to preserve backwards compatibility. I see some problems with using the current structs though, if the field does not exist then sending a zero value is not the right action. This might warrant doing an ad-hoc parsing, though I feel like unmarshal to struct is more easy to understand. I could turn all fields to pointers and add them only when not nil or maybe there is a better pattern that I can use? @nathanielc On the topic of rename on the Kapacitor side, either way works for me but I don't think it is a requirement for Telegraf. I see you have stuck with the exact naming and capitalization that go uses, even though it is in contradiction with the rest of the endpoint... |
|
I don't think it should be up to telegraf to document them. It should be up to the system which creates the metrics. If telegraf starts writing its own documentation on them, it might get it wrong. If we're just copy/pasting the upstream doc, why not just link to the upstream docs instead? |
|
I think telegraf just needs to document that they exist and may be sent, not the meaning of them. Otherwise you either need to run --test or read the source to know what the plugin will do. |
danielnelson
force-pushed
the
ross-kapacitor-input
branch
from
da5d8e1 to
bdd3d96
Compare
Required for all PRs: