ramesh-muthusamy opened a new pull request #10910:
URL: https://github.com/apache/kafka/pull/10910
Issue:
We noticed that the Error metrics reported in Kafka Connect worker continues
to stay even after the task is re distributed to another worker. As a result
you would notice over a period of time the task_error_metrics of a worker
would contain the errors of all the tasks that it had ever come across.
This is an anti pattern to what other metrics are reported by Kafka Connect
worker. The Kafka Connect worker should only report the error metrics of the
present task and leave the persistence of the previous tasks to the metrics
storage system that is consuming these metrics.
In the below example we notice that there is only 2 active tasks that are
running , but we have more than 20+ tasks error metrics that are available.
Task counter mbean:
{"request":\{"mbean":"kafka.connect:type=connect-worker-metrics","type":"read"}
,"value":{"connector-startup-failure-percentage":0.0,"task-startup-attempts-total":90.0,"connector-startup-success-total":1.0,"connector-startup-failure-total":0.0,"task-startup-success-percentage":0.0,"connector-startup-attempts-total":1.0,"connector-count":0.0,"connector-startup-success-percentage":0.0,"task-startup-success-total":90.0,"task-startup-failure-percentage":0.0,"task-count":2.0,"task-startup-failure-total":0.0},"timestamp":1623852927,"status":200}
Task Error metrics mbean:
{"request":\{"mbean":"kafka.connect:connector=*,task=*,type=task-error-metrics","type":"read"}
,"value":{"kafka.connect:connector=***********,task=35,type=task-error-metrics":
{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0}
,"kafka.connect:connector=**********,task=38,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=14,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=5,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=0,type=task-error-metrics":{"last-error-timestamp":
0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=29,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=37,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=28,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-fa
ilures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=25,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=91,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=31,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,
"total-retries":0.0},"kafka.connect:connector=*********,task=7,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=74,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=2,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=26,type=task-error-metrics":{"last-
error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=30,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=*********,task=53,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadletterqueue-produce-failures":0.0,"total-record-failures":0.0,"total-records-skipped":0.0,"total-record-errors":0.0,"total-retries":0.0},"kafka.connect:connector=**********,task=16,type=task-error-metrics":{"last-error-timestamp":0,"total-errors-logged":0.0,"deadletterqueue-produce-requests":0.0,"deadlett
erqueue-produce-failures":0.0,"total-record-failures":0.0, .....
Solution:
As part of the bug fix to KAFKA-12965 introducing code changes to gracefully
cleanup the error handling metrics associated with a task. This is required to
avoid duplicate metrics of task being reported from a worker that had the same
task in the past.
UT - Not yet covered, in progress
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
