mirror of https://github.com/apache/nifi.git
NIFI-2465 - InferAvroSchema EL support based on incoming FlowFiles
This closes #863. Signed-off-by: Bryan Bende <bbende@apache.org>
This commit is contained in:
parent
26f5c496d1
commit
b0122c6a73
|
@ -95,7 +95,7 @@ public class InferAvroSchema
|
|||
.subject(subject)
|
||||
.input(input)
|
||||
.explanation("Only non-null single characters are supported")
|
||||
.valid(input.length() == 1 && input.charAt(0) != 0)
|
||||
.valid(input.length() == 1 && input.charAt(0) != 0 || context.isExpressionLanguagePresent(input))
|
||||
.build();
|
||||
}
|
||||
};
|
||||
|
@ -175,6 +175,7 @@ public class InferAvroSchema
|
|||
public static final PropertyDescriptor DELIMITER = new PropertyDescriptor.Builder()
|
||||
.name("CSV delimiter")
|
||||
.description("Delimiter character for CSV records")
|
||||
.expressionLanguageSupported(true)
|
||||
.addValidator(CHAR_VALIDATOR)
|
||||
.defaultValue(",")
|
||||
.build();
|
||||
|
@ -212,6 +213,7 @@ public class InferAvroSchema
|
|||
.description("Character encoding of CSV data.")
|
||||
.required(true)
|
||||
.defaultValue("UTF-8")
|
||||
.expressionLanguageSupported(true)
|
||||
.addValidator(StandardValidators.CHARACTER_SET_VALIDATOR)
|
||||
.build();
|
||||
|
||||
|
@ -391,14 +393,14 @@ public class InferAvroSchema
|
|||
}
|
||||
|
||||
//Prepares the CSVProperties for kite
|
||||
final CSVProperties props = new CSVProperties.Builder()
|
||||
.delimiter(context.getProperty(DELIMITER).getValue())
|
||||
.escape(context.getProperty(ESCAPE_STRING).evaluateAttributeExpressions().getValue())
|
||||
.quote(context.getProperty(QUOTE_STRING).evaluateAttributeExpressions().getValue())
|
||||
CSVProperties props = new CSVProperties.Builder()
|
||||
.charset(context.getProperty(CHARSET).evaluateAttributeExpressions(inputFlowFile).getValue())
|
||||
.delimiter(context.getProperty(DELIMITER).evaluateAttributeExpressions(inputFlowFile).getValue())
|
||||
.quote(context.getProperty(QUOTE_STRING).evaluateAttributeExpressions(inputFlowFile).getValue())
|
||||
.escape(context.getProperty(ESCAPE_STRING).evaluateAttributeExpressions(inputFlowFile).getValue())
|
||||
.linesToSkip(context.getProperty(HEADER_LINE_SKIP_COUNT).evaluateAttributeExpressions(inputFlowFile).asInteger())
|
||||
.header(header.get())
|
||||
.hasHeader(hasHeader.get())
|
||||
.linesToSkip(context.getProperty(HEADER_LINE_SKIP_COUNT).evaluateAttributeExpressions().asInteger())
|
||||
.charset(context.getProperty(CHARSET).getValue())
|
||||
.build();
|
||||
|
||||
final AtomicReference<String> avroSchema = new AtomicReference<>();
|
||||
|
@ -408,7 +410,7 @@ public class InferAvroSchema
|
|||
public void process(InputStream in) throws IOException {
|
||||
avroSchema.set(CSVUtil
|
||||
.inferSchema(
|
||||
context.getProperty(RECORD_NAME).evaluateAttributeExpressions().getValue(), in, props)
|
||||
context.getProperty(RECORD_NAME).evaluateAttributeExpressions(inputFlowFile).getValue(), in, props)
|
||||
.toString(context.getProperty(PRETTY_AVRO_OUTPUT).asBoolean()));
|
||||
}
|
||||
});
|
||||
|
@ -435,8 +437,8 @@ public class InferAvroSchema
|
|||
@Override
|
||||
public void process(InputStream in) throws IOException {
|
||||
Schema as = JsonUtil.inferSchema(
|
||||
in, context.getProperty(RECORD_NAME).evaluateAttributeExpressions().getValue(),
|
||||
context.getProperty(NUM_RECORDS_TO_ANALYZE).evaluateAttributeExpressions().asInteger());
|
||||
in, context.getProperty(RECORD_NAME).evaluateAttributeExpressions(inputFlowFile).getValue(),
|
||||
context.getProperty(NUM_RECORDS_TO_ANALYZE).evaluateAttributeExpressions(inputFlowFile).asInteger());
|
||||
avroSchema.set(as.toString(context.getProperty(PRETTY_AVRO_OUTPUT).asBoolean()));
|
||||
|
||||
}
|
||||
|
|
|
@ -106,9 +106,7 @@ public class TestInferAvroSchema {
|
|||
|
||||
MockFlowFile data = runner.getFlowFilesForRelationship(InferAvroSchema.REL_SUCCESS).get(0);
|
||||
String avroSchema = data.getAttribute(InferAvroSchema.AVRO_SCHEMA_ATTRIBUTE_NAME);
|
||||
String knownSchema = new String(unix2PlatformSpecificLineEndings(
|
||||
new File("src/test/resources/Shapes.json.avro")),
|
||||
StandardCharsets.UTF_8);
|
||||
String knownSchema = new String(unix2PlatformSpecificLineEndings(new File("src/test/resources/Shapes.json.avro")), StandardCharsets.UTF_8);
|
||||
Assert.assertEquals(avroSchema, knownSchema);
|
||||
|
||||
// Since that avro schema is written to an attribute this should be teh same as the original
|
||||
|
@ -223,9 +221,77 @@ public class TestInferAvroSchema {
|
|||
flowFile.assertContentEquals(new File("src/test/resources/Shapes_Header_TabDelimited.csv").toPath());
|
||||
flowFile.assertAttributeEquals(CoreAttributes.MIME_TYPE.key(), "text/csv");
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
public void specifyCSVparametersInExpressionLanguage() throws Exception {
|
||||
runner.setProperty(InferAvroSchema.DELIMITER, "${csv.delimiter}");
|
||||
runner.setProperty(InferAvroSchema.ESCAPE_STRING, "${csv.escape}");
|
||||
runner.setProperty(InferAvroSchema.QUOTE_STRING, "${csv.quote}");
|
||||
runner.setProperty(InferAvroSchema.CHARSET, "${csv.charset}");
|
||||
runner.setProperty(InferAvroSchema.GET_CSV_HEADER_DEFINITION_FROM_INPUT, "true");
|
||||
|
||||
runner.assertValid();
|
||||
|
||||
@SuppressWarnings("serial")
|
||||
Map<String, String> attributes = new HashMap<String, String>() {
|
||||
{
|
||||
put("csv.delimiter",",");
|
||||
put("csv.escape", "\\");
|
||||
put("csv.quote", "\"");
|
||||
put("csv.charset", "UTF-8");
|
||||
put(CoreAttributes.MIME_TYPE.key(), "text/csv");
|
||||
}
|
||||
};
|
||||
|
||||
runner.enqueue(new File("src/test/resources/Shapes_Header.csv").toPath(), attributes);
|
||||
|
||||
runner.run();
|
||||
runner.assertTransferCount(InferAvroSchema.REL_UNSUPPORTED_CONTENT, 0);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_FAILURE, 0);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_ORIGINAL, 1);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_SUCCESS, 1);
|
||||
|
||||
MockFlowFile flowFile = runner.getFlowFilesForRelationship(InferAvroSchema.REL_SUCCESS).get(0);
|
||||
flowFile.assertContentEquals(unix2PlatformSpecificLineEndings(new File("src/test/resources/Shapes_header.csv.avro")));
|
||||
flowFile.assertAttributeEquals(CoreAttributes.MIME_TYPE.key(), "application/avro-binary");
|
||||
|
||||
}
|
||||
|
||||
@Test
|
||||
public void specifyJsonParametersInExpressionLanguage() throws Exception {
|
||||
runner.assertValid();
|
||||
runner.setProperty(InferAvroSchema.INPUT_CONTENT_TYPE, InferAvroSchema.USE_MIME_TYPE);
|
||||
|
||||
// Purposely set to True to test that none of the JSON file is read which would cause issues.
|
||||
runner.setProperty(InferAvroSchema.GET_CSV_HEADER_DEFINITION_FROM_INPUT, "true");
|
||||
runner.setProperty(InferAvroSchema.SCHEMA_DESTINATION, InferAvroSchema.DESTINATION_ATTRIBUTE);
|
||||
runner.setProperty(InferAvroSchema.RECORD_NAME, "${record.name}");
|
||||
runner.setProperty(InferAvroSchema.NUM_RECORDS_TO_ANALYZE, "${records.analyze}");
|
||||
|
||||
Map<String, String> attributes = new HashMap<>();
|
||||
attributes.put(CoreAttributes.MIME_TYPE.key(), "application/json");
|
||||
attributes.put("record.name", "myrecord");
|
||||
attributes.put("records.analyze", "2");
|
||||
runner.enqueue(new File("src/test/resources/Shapes.json").toPath(), attributes);
|
||||
|
||||
runner.run();
|
||||
runner.assertTransferCount(InferAvroSchema.REL_UNSUPPORTED_CONTENT, 0);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_FAILURE, 0);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_ORIGINAL, 1);
|
||||
runner.assertTransferCount(InferAvroSchema.REL_SUCCESS, 1);
|
||||
|
||||
MockFlowFile data = runner.getFlowFilesForRelationship(InferAvroSchema.REL_SUCCESS).get(0);
|
||||
String avroSchema = data.getAttribute(InferAvroSchema.AVRO_SCHEMA_ATTRIBUTE_NAME);
|
||||
Assert.assertTrue(avroSchema.contains("\"name\" : \"myrecord\""));
|
||||
|
||||
// Since that avro schema is written to an attribute this should be teh same as the original
|
||||
data.assertAttributeEquals(CoreAttributes.MIME_TYPE.key(), "application/json");
|
||||
}
|
||||
|
||||
|
||||
static byte[] unix2PlatformSpecificLineEndings(final File file) throws IOException {
|
||||
try ( final BufferedInputStream in = new BufferedInputStream(new FileInputStream(file));
|
||||
final ByteArrayOutputStream out = new ByteArrayOutputStream()) {
|
||||
try (final BufferedInputStream in = new BufferedInputStream(new FileInputStream(file)); final ByteArrayOutputStream out = new ByteArrayOutputStream()) {
|
||||
byte eol[] = System.lineSeparator().getBytes(StandardCharsets.UTF_8);
|
||||
int justRead;
|
||||
while ((justRead = in.read()) != -1) {
|
||||
|
@ -238,4 +304,5 @@ public class TestInferAvroSchema {
|
|||
return out.toByteArray();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue