Name: secor
Owner: Pinterest
Description: Secor is a service implementing Kafka log persistence
Created: 2014-04-15 22:26:44.0
Updated: 2018-01-17 09:14:51.0
Pushed: 2018-01-17 17:49:47.0
Size: 1063
Language: Java
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Secor is a service persisting Kafka logs to Amazon S3, Google Cloud Storage, Microsoft Azure Blob Storage and Openstack Swift.
clone [git-repo-url] secor
ecor
Edit src/main/config/*.properties
files to specify parameters describing the environment. Those files contain comments describing the meaning of individual parameters.
default this will install the "release" (Kafka 0.10 profile)
package
r ${SECOR_INSTALL_DIR} # directory to place Secor binaries in.
-zxvf target/secor-0.1-SNAPSHOT-bin.tar.gz -C ${SECOR_INSTALL_DIR}
use the Kafka 0.8 client you should use the kafka-0.8-dev profile
-Pkafka-0.8-dev package
{SECOR_INSTALL_DIR}
ripts/run_tests.sh
{SECOR_INSTALL_DIR}
-ea -Dsecor_group=secor_backup \
log4j.configuration=log4j.prod.properties \
config=secor.prod.backup.properties \
p secor-0.1-SNAPSHOT.jar:lib/* \
m.pinterest.secor.main.ConsumerMain
One of the convenience features of Secor is the ability to group messages and save them under common file prefixes. The partitioning is controlled by a message parser. Secor comes with the following parsers:
offset parser: parser that groups messages based on offset ranges. E.g., messages with offsets in range 0 to 999 will end up under `s3n://bucket/topic/offset=0/
, offsets 1000 to 2000 will go to ``
s3n://bucket/topic/offset=1000/```. To use this parser, start Secor with properties file secor.prod.backup.properties.
Thrift date parser: parser that extracts timestamps from thrift messages and groups the output based on the date (at a day granularity). To keep things simple, this parser assumes that the timestamp is carried in the first field (id 1) of the thrift message schema by default. The field id can be changed by setting `message.timestamp.id
as long as the field is at the top level of the thrift object (i.e. it is not in a nested structure). The timestamp may be expressed either in seconds or milliseconds, or nanoseconds since the epoch. The output goes to date-partitioned paths (e.g., ``
s3n://bucket/topic/dt=2014-05-01`,
s3n://bucket/topic/dt=2014-05-02``
). Date partitioning is particularly convenient if the output is to be consumed by ETL tools such as Hive. To use this parser, start Secor with properties file secor.prod.partition.properties. Note the `message.timestamp.name
` property has no effect on the thrift parsing, which is determined by the field id.
JSON timestamp parser: parser that extracts UNIX timestamps from JSON messages and groups the output based on the date, similar to the Thrift parser above. To use this parser, start Secor with properties file secor.prod.partition.properties and set secor.message.parser.class=com.pinterest.secor.parser.JsonMessageParser
. You may override the field used to extract the timestamp by setting the “message.timestamp.name” property.
Avro timestamp parser: parser that extracts UNIX timestamps from AVRO messages and groups the output based on the date, similar to the Thrift parser above. To use this parser, start Secor with properties file secor.prod.partition.properties and set secor.message.parser.class=com.pinterest.secor.parser.AvroMessageParser
. You may override the field used to extract the timestamp by setting the “message.timestamp.name” property.
JSON ISO 8601 date parser: Assumes your timestamp field uses ISO 8601. To use this parser, start Secor with properties file secor.prod.partition.properties and set secor.message.parser.class=com.pinterest.secor.parser.Iso8601MessageParser
. You may override the field used to extract the timestamp by setting the “message.timestamp.name” property.
MessagePack date parser: parser that extracts timestamps from MessagePack messages and groups the output based on the date, similar to the Thrift and JSON parser. To use this parser, set secor.message.parser.class=com.pinterest.secor.parser.MessagePackParser
. Like the Thrift parser, the timestamp may be expressed either in seconds or milliseconds, or nanoseconds since the epoch and respects the “message.timestamp.name” property.
Protocol Buffers date parser: parser that extracts timestamps from protobuf messages and groups the output based on the date, similar to the Thrift, JSON or MessagePack parser. To use this parser, set secor.message.parser.class=com.pinterest.secor.parser.ProtobufMessageParser
. Like the Thrift parser, the timestamp may be expressed either in seconds or milliseconds, or nanoseconds since the epoch and respects the “message.timestamp.name” property.
Output grouping with Flexible partitions: The default partitioning granularity for date, hours and minutes have prefix for convenient consumption for Hive
. If you require different naming of partition with(out) prefix and other date, hour or minute format update the following properties in secor.common.properties
partitioner.granularity.date.prefix=dt=
partitioner.granularity.hour.prefix=hr=
partitioner.granularity.minute.prefix=min=
partitioner.granularity.date.format=yyyy-MM-dd
partitioner.granularity.hour.format=HH
partitioner.granularity.minute.format=mm
If none of the parsers available out-of-the-box is suitable for your use case, note that it is very easy to implement a custom parser. All you have to do is to extend MessageParser and tell Secor to use your parser by setting `secor.message.parser.class
` in the properties file.
Currently secor supports the following output formats
Sequence Files: Flat file containing binary key value pairs. To use this format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.SequenceFileReaderWriterFactory
option.
Delimited Text Files: A new line delimited raw text file. To use this format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.DelimitedTextFileReaderWriterFactory
option.
ORC Files: Optimized row columnar format. To use this format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.JsonORCFileReaderWriterFactory
option. Additionally, ORC schema must be specified per topic like this secor.orc.message.schema.<topic>=<orc schema>
. If all Kafka topics receive same format data then this option can be used secor.orc.message.schema.*=<orc schema>
. User can implement custom ORC schema provider by implementing ORCScehmaProvider interface and the new provider class should be specified using option secor.orc.schema.provider=<orc schema provider class name>
. By default this property is DefaultORCSchemaProvider.
Parquet Files (for Protobuf messages): Columnar storage format. To use this output format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.ProtobufParquetFileReaderWriterFactory
option. In addition, Protobuf message class per Kafka topic must be defined using option secor.protobuf.message.class.<topic>=<protobuf class name>
. If all Kafka topics transfer the same protobuf message type, set secor.protobuf.message.class.*=<protobuf class name>
.
Parquet Files (for JSON messages): Columnar storage format. In addition to setting all options necessary to write Protobuf to Parquet (see above), the JSON topics must be explicitly defined using the option secor.topic.message.format.<topic>=JSON
or secor.topic.message.format.*=JSON
if all Kafka topics use JSON. The protobuf classes defined per topic will be used as intermediaries between the JSON messages and Parquet files.
Parquet Files (for Thrift messages): Columnar storage format. To use this output format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.ThriftParquetFileReaderWriterFactory
option. In addition, thrift message class per Kafka topic must be defined using option secor.thrift.message.class.<topic>=<thrift class name>
. If all Kafka topics transfer the same thrift message type, set secor.thrift.message.class.*=<thrift class name>
. It is asumed all messages use the same thrift protocol. Thrift protocol is set in secor.thrift.protocol.class
.
Parquet Files (for Avro messages): Columnar storage format. To use this output format, set secor.file.reader.writer.factory=com.pinterest.secor.io.impl.AvroParquetFileReaderWriterFactory
option. The schema.registry.url
option must be set.
Gzip upload format: To enable compression on uploaded files to the cloud, in secor.common.properties
set secor.compression.codec
to a valid compression codec implementing org.apache.hadoop.io.compress.CompressionCodec
interface, such as org.apache.hadoop.io.compress.GzipCodec
.
Secor comes with a number of tools implementing interactions with the environment.
Log file printer displays the content of a log file.
-ea -Dlog4j.configuration=log4j.prod.properties -Dconfig=secor.prod.backup.properties -cp "secor-0.1-SNAPSHOT.jar:lib/*" com.pinterest.secor.main.LogFilePrinterMain -f s3n://bucket/path
Log file verifier checks the consistency of log files.
-ea -Dlog4j.configuration=log4j.prod.properties -Dconfig=secor.prod.backup.properties -cp "secor-0.1-SNAPSHOT.jar:lib/*" com.pinterest.secor.main.LogFileVerifierMain -t topic -q
Topic finalizer writes _SUCCESS files to date partitions that very likely won't be receiving any new messages and (optionally) adds the corresponding dates to Hive through Qubole API.
-ea -Dlog4j.configuration=log4j.prod.properties -Dconfig=secor.prod.backup.properties -cp "secor-0.1-SNAPSHOT.jar:lib/*" com.pinterest.secor.main.PartitionFinalizerMain
Progress monitor exports offset consumption lags per topic partition to OpenTSDB / statsD. Lags track how far Secor is behind the producers.
-ea -Dlog4j.configuration=log4j.prod.properties -Dconfig=secor.prod.backup.properties -cp "secor-0.1-SNAPSHOT.jar:lib/*" com.pinterest.secor.main.ProgressMonitorMain
Set monitoring.interval.seconds
to a value larger than 0 to run in a loop, exporting stats every monitoring.interval.seconds
seconds.
Design details are available in DESIGN.md.
Secor is distributed under Apache License, Version 2.0.
If you have any questions or comments, you can reach us at secor-users@googlegroups.com