Mark Payne created NIFI-5801:
--------------------------------
Summary: Evaluating Expression Language can in many cases be made
much more efficient
Key: NIFI-5801
URL: https://issues.apache.org/jira/browse/NIFI-5801
Project: Apache NiFi
Issue Type: Bug
Components: Core Framework
Reporter: Mark Payne
Assignee: Mark Payne
When a StandardPropertyValue is obtained and evaluateAttributeExpressions is
called, it builds the entire Evaluator Tree each time. This was done to ensure
that Evaluator.evaluate() is called only once. However, the requirement to call
this only once was introduced as a way to have anyMatchingAttribute,
anyAttribute, allMatchingAttributes, allAttributes, etc. methods work, and
these are rarely used. I.e., we introduced semantics that significantly slow
performance in order to provide functionality that is used maybe 1% of the
time. Instead, we should optimize for the 99% use case and incur a penalty, if
necessary, in the 1% use case instead. Profiling the ConsumeKafkaRecord
processor shows that 80% of the time in that method is evaluating Expression
Language for `${schema.name}` to determine which schema should be used. We can
likely make this evaluation just as quick as attributeMap.get("schema.name") by
pre-building the Evaluators and re-using them.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
