Graylog Performance Unstable

In our integration environment, there are two graylog nodes within a Kubernetes cluster. The process buffer is full (input buffer and output buffer are empty).
I have been trying to tune the performance by modifying the # of the processor for each buffer, besides increasing heap size. The # of out message did reach 3,000~4,000 per second for a while, then it dropped back to ~300 per second without any configuration change.

Question:
What may cause the process buffer full while others are empty?
Why the number of out messages changes while no configuration change?
What parameters should I try to config?

Thanks,
Tiger

There are indexer failures with message:
{“type”:“mapper_parsing_exception”,“reason”:“failed to parse field [metricValue] of type [float] in document with id ‘cbbff660-46bd-11ea-9de4-0a580ae96622’”,“caused_by”:{“type”:“illegal_argument_exception”,“reason”:"[float] supports only finite values, but got [NaN]"}}

Could this the cause? How to fix it?
Please near with me, a beginner of Graylog. It is really appreciate if you can point me to any step by step guidance.

I found the performance cause: greedy regex is used for two extractors.
Still need help on how to fix the “mapper_parsing_exception”

every time performance issue, like your regex. Give more CPU instead of set the number of processors.

Please read the error message.
{“type”:“mapper_parsing_exception”,“reason”:“failed to parse field [metricValue] of type [float] in document with id ‘cbbff660-46bd-11ea-9de4-0a580ae96622’”,“caused_by”:{“type”:“illegal_argument_exception”,“reason”:"[float] supports only finite values, but got [NaN]"}}

Maybe the wikipedia give some help for you.

There is a limit of available CPU :slight_smile: I am trying to tune the regex.
For the mapper_parsing_exception error, is it emitted by Graylog or Elasticsearch?

user error.
Check the log what you send to graylog.
//ES can’t handle it.
Please read again the error message. If you have question, please repeat it.