René Zeidler created NIFI-10666:
-----------------------------------
Summary: PrometheusReportingTask does not use UTF-8 encoding on
/metrics/ endpoint
Key: NIFI-10666
URL: https://issues.apache.org/jira/browse/NIFI-10666
Project: Apache NiFi
Issue Type: Bug
Components: Extensions
Affects Versions: 1.16.3
Reporter: René Zeidler
Attachments: missing-header.png
We have created a default PrometheusReportingTask for our NiFi instance and
tried to consume the metrics with Prometheus. However, Prometheus threw the
following error:
{code:java}
ts=2022-10-19T12:25:18.110Z caller=scrape.go:1332 level=debug component="scrape
manager" scrape_pool=nifi-cluster target=http://***nifi***:9092/metrics
msg="Append failed" err="invalid UTF-8 label value" {code}
Upon further inspection, we noticed that the /metrics/ endpoint exposed by the
reporting task does not use UTF-8 encoding, which is required by Prometheus.
Our flow uses non-ASCII characters (in our case German umlauts like "ü"). As a
workaround, removing those characters fixes the Prometheus error, but this is
not practical for a large flow in German language.
Opening the /metrics/ endpoint in a browser confirms that the encoding used is
not UTF-8:
{code:java}
> document.characterSet
'windows-1252' {code}
----
The responsible code might be here:
[https://github.com/apache/nifi/blob/2be5c26f287469f4f19f0fa759d6c1b56dc0e348/nifi-nar-bundles/nifi-prometheus-bundle/nifi-prometheus-reporting-task/src/main/java/org/apache/nifi/reporting/prometheus/PrometheusServer.java#L67]
The PrometheusServer used by the reporting task uses an OutputStreamWriter with
the default encoding, instead of explicitly using UTF-8. The Content-Type
header set in that function also does not get passed along (see screenshot)
----
Our cluster is still on 1.16.3, but once we update to 1.18.0 we can update this
report to confirm that the issue still persists. Considering that there were no
code changes in the Prometheus reporting task, it is likely that the issue
still persists in the current version.
If someone can confirm this before we update, feel free to update the issue :)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
