[
https://issues.apache.org/jira/browse/NIFI-14454?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Curtis Kennington updated NIFI-14454:
-------------------------------------
Description:
When using NiFi 1.x, we used Counters to provide custom metrics - these were
commonly used to identify failures in the flow (e.g. SFTP failed, etc.). These
were exposed in Prometheus format via PrometheusReportingTask.
When upgraded to NiFi 2.x, the PrometheusReportingTask was removed in favour of
the {{nifi-api/flow/metrics/prometheus}} API. However (potentially as this API
is flow orientated?) the counters have a value of 0:
{quote}nifi_processor_counters\{processor_name="Inc upload success
counter",counter_name="billing_upload_success",processor_id="bb721341-39ed-3c32-70be-1972da8770d7",instance="004d0f09-0196-1000-d983-0eb3588c281c",}
0.0
{quote}
In the counter API we can see that these counters have positive values set:
{quote} \{
"id": "12fa996f-b0fb-3df0-afee-0fa598768c25",
"context": "All UpdateCounter's",
"name": "billing_upload_success",
"valueCount": 100,
"value": "100"
\},
{quote}
Is this a defect, or are we implementing our metrics incorrectly?
How can we expose the counters as prometheus metrics, so we can raise our own
alarms for operators to take human manual actions when something goes wrong?
was:
When using NiFi 1.x, we used Counters to provide custom metrics - these were
commonly used to identify failures in the flow (e.g. SFTP failed, etc.). These
were exposed in Prometheus format via PrometheusReportingTask.
When upgraded to NiFi 2.x, the PrometheusReportingTask was removed in favour of
the {{nifi-api/flow/metrics/prometheus}} API. However (potentially as this API
is flow orientated?) the counters have a value of 0:
{quote}nifi_processor_counters\{processor_name="Inc upload success
counter",counter_name="billing_upload_success",processor_id="bb721341-39ed-3c32-70be-1972da8770d7",instance="004d0f09-0196-1000-d983-0eb3588c281c",}
0.0
{quote}
In the counter API we can see that these counters have positive values set:
{quote} {
"id": "12fa996f-b0fb-3df0-afee-0fa598768c25",
"context": "All UpdateCounter's",
"name": "billing_upload_success",
"valueCount": 100,
"value": "100"
},
{quote}
Is this a defect, or are we implementing our metrics incorrectly?
How can we expose the counters as prometheus metrics, so we can raise our own
alarms for operators to take human manual actions when something goes wrong?
> Counter values are recorded as 0 in Prometheus Metrics
> ------------------------------------------------------
>
> Key: NIFI-14454
> URL: https://issues.apache.org/jira/browse/NIFI-14454
> Project: Apache NiFi
> Issue Type: Bug
> Components: NiFi API
> Affects Versions: 2.3.0
> Reporter: Curtis Kennington
> Priority: Minor
>
> When using NiFi 1.x, we used Counters to provide custom metrics - these were
> commonly used to identify failures in the flow (e.g. SFTP failed, etc.).
> These were exposed in Prometheus format via PrometheusReportingTask.
> When upgraded to NiFi 2.x, the PrometheusReportingTask was removed in favour
> of the {{nifi-api/flow/metrics/prometheus}} API. However (potentially as this
> API is flow orientated?) the counters have a value of 0:
> {quote}nifi_processor_counters\{processor_name="Inc upload success
> counter",counter_name="billing_upload_success",processor_id="bb721341-39ed-3c32-70be-1972da8770d7",instance="004d0f09-0196-1000-d983-0eb3588c281c",}
> 0.0
> {quote}
> In the counter API we can see that these counters have positive values set:
> {quote} \{
> "id": "12fa996f-b0fb-3df0-afee-0fa598768c25",
> "context": "All UpdateCounter's",
> "name": "billing_upload_success",
> "valueCount": 100,
> "value": "100"
> \},
> {quote}
> Is this a defect, or are we implementing our metrics incorrectly?
> How can we expose the counters as prometheus metrics, so we can raise our own
> alarms for operators to take human manual actions when something goes wrong?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
