-
Notifications
You must be signed in to change notification settings - Fork 751
Configuration Properties Glossary
Configuration properties are key/value pairs that are set in text files. They include system properties that control how Gobblin will pull data, and control what source Gobblin will pull the data from. Configuration files end in some user-specified suffix (by default text files ending in .pull or .job are recognized as configs files, although this is configurable). Each file represents some unit of work that needs to be done in Gobblin. For example, there will typically be a separate configuration file for each table that needs to be pulled from a database.
The first section of this document contains all the required properties needed to run a basic Gobblin job. The rest of the document is dedicated to other properties that can be used to configure Gobbin jobs. The description of each configuration parameter will often refer to core Gobblin concepts and terms. If any of these terms are confusing, check out the Gobblin Archiecture page for a more detailed explanation of how Gobblin works. The GitHub repo also contains sample config files for specific sources. For example, there are sample config files to connect to MySQL databases and SFTP servers.
Gobblin also allows you to specify a global configuration file that contains common properties that are shared across all jobs. The Job Launcher Properties section has more information on how to specify a global properties file.
- Properties File Format
- Creating a Basic Properties File
- Job Launcher Properties
- Job Type Properties
- Task Execution Properties
- State Store Properties
- Metrics Properties
- Email Alert Properties
- Source Properties
- Converter Properties
- Quality Checker Properties
- Writer Properties
- Data Publisher Properties
- Generic Properties
Configuration properties files follow the Java Properties text file format. Further, file includes and variable expansion/interpolation as defined in Apache Commons Configuration are also supported.
Example:
- common.properties
writer.staging.dir=/path/to/staging/dir/
writer.output.dir=/path/to/output/dir/
- my-job.properties
include=common.properties
job.name=MyFirstJob
In order to create a basic configuration property there is a small set of required properties that need to be set. The following properties are required to run any Gobblin job:
-
job.name- Name of the job -
source.class- Fully qualified path to the Source class responsible for connecting to the data source -
writer.staging.dir- The directory each task will write staging data to -
writer.output.dir- The directory each task will commit data to -
data.publisher.final.dir- The final directory where all the data will be published -
state.store.dir- The directory where state-store files will be written
For more information on each property, check out the comprehensive list below.
If only these properties are set, then by default, Gobblin will run in Local mode, as opposed to running on Hadoop M/R. This means Gobblin will write Avro data to the local filesystem. In order to write to HDFS, set the writer.fs.uri property to the URI of the HDFS NameNode that data should be written to. Since the default version of Gobblin writes data in Avro format, the writer expects Avro records to be passed to it. Thus, any data pulled from an external source must be converted to Avro before it can be written out to the filesystem.
The source.class property is one of the most important properties in Gobblin. It specifies what Source class to use. The Source class is responsible for determining what work needs to be done during each run of the job, and specifies what Extractor to use in order to read over each sub-unit of data. Examples of Source classes are WikipediaSource and SimpleJsonSource, which can be found in the GitHub repository. For more information on Sources and Extractors, check out the Architecture page.
Typically, Gobblin jobs will be launched using the launch scripts in the bin folder. These scripts allow jobs to be launched on the local machine (e.g. SchedulerDaemon) or on Hadoop (e.g. CliMRJobLauncher). Check out the Job Launcher section below, to see the configuration difference between each launch mode. The [Deployment](Gobblin Deployment) page also has more information on the different ways a job can be launched.
Gobblin jobs can be launched and scheduled in a variety of ways. They can be scheduled via a Quartz scheduler or through Azkaban. Jobs can also be run without a scheduler via the Command Line. For more information on launching Gobblin jobs, check out the Deployment page.
These properties are common to both the Job Launcher and the Command Line.
The name of the job to run. This name must be unique within a single Gobblin instance.
None
Yes
A way to group logically similar jobs together.
None
No
A description of what the jobs does.
None
No
Directory where job locks are stored. Job locks are used by the scheduler to ensure two executions of a job do not run at the same time. If a job is scheduled to run, Gobblin will first check this directory to see if there is a lock file for the job. If there is one, it will not run the job, if there isn't one then it will run the job.
None
No
If set to true job locks are enabled, if set to false they are disabled
True
No
A boolean specifying whether the job will be only once, or multiple times. If set to true the job will only be run once even if a job.schedule is specified. If set to false and a job.schedule is specified then it will run according to the schedule. If set false and a job.schedule is not specified, it will run only once.
False
No
Whether the job is disabled or not. If set to true, then Gobblin will not run this job.
False
No
This class is used to schedule Gobblin jobs on Quartz. The job can be launched via the command line, and takes in the location of a global configuration file as a parameter. This configuration file should have the property jobconf.dir in order to specify the location of all the .job or .pull files. Another core difference, is that the global configuration file for the SchedulerDaemon must specify the following properties:
writer.staging.dirwriter.output.dirdata.publisher.final.dirstate.store.dir
They should not be set in individual job files, as they are system-level parameters. For more information on how to set the configuration parameters for jobs launched through the SchedulerDaemon, check out the Deployment page.
Cron-Based job schedule. This schedule only applies to jobs that run using Quartz.
None
No
When running in local mode, Gobblin will check this directory for any configuration files. Each configuration file should correspond to a separate Gobblin job, and each one should in a suffix specified by the jobconf.extensions parameter.
None
No
Comma-separated list of supported job configuration file extensions. When running in local mode, Gobblin will only pick up job files ending in these suffixes.
pull,job
No
Controls how often Gobblin checks the jobconf.dir for new configuration files, or for configuration file updates. The parameter is measured in milliseconds.
300000
No
There are no configuration parameters specific to CliMRJobLauncher. This class is used to launch Gobblin jobs on Hadoop from the command line, the jobs are not scheduled. Common properties are set using the --sysconfig option when launching jobs via the command line. For more information on how to set the configuration parameters for jobs launched through the command line, check out the Deployment page.
There are no configuration parameters specific to AzkabanJobLauncher. This class is used to schedule Gobblin jobs on Azkaban. Common properties can be set through Azkaban by creating a .properties file, check out the Azkaban Documentation for more information. For more information on how to set the configuration parameters for jobs scheduled through the Azkaban, check out the Deployment page.
Job launcher type; one of LOCAL, MAPREDUCE, YARN. LOCAL mode runs on a single machine (LocalJobLauncher), MAPREDUCE runs on a Hadoop cluster (MRJobLauncher), and YARN runs on a YARN cluster (not implemented yet).
LOCAL
No
There are no configuration parameters specific to LocalJobLauncher. The LocalJobLauncher will launch a Hadoop job on a single machine. If launcher.type is set to LOCAL then this class will be used to launch the job. Properties required by the MRJobLauncher class.
Comma-separated list of jars the Gobblin framework depends on. These jars will be added to the classpath of the job, and to the classpath of any containers the job launches.
None
No
Comma-separated list of jar files the job depends on. These jars will be added to the classpath of the job, and to the classpath of any containers the job launches.
None
No
Comma-separated list of local files the job depends on. These files will be available to any map tasks that get launched via the DistributedCache.
None
No
Comma-separated list of files on HDFS the job depends on. These files will be available to any map tasks that get launched via the DistributedCache.
None
No
Working directory for a Gobblin Hadoop MR job. Gobblin uses this to write intermediate data, such as the workunit state files that are used by each map task. This has to be a path on HDFS.
None
Yes
Maximum number of mappers to use in a Gobblin Hadoop MR job. If no explicit limit is set then a map task for each workunit will be launched. If the value of this properties is less than the number of workunits created, then each map task will run multiple tasks.
None
No
Whether to include task-level counters in the set of counters reported as Hadoop counters. Hadoop imposes a system-level limit (default to 120) on the number of counters, so a Gobblin MR job may easily go beyond that limit if the job has a large number of tasks and each task has a few counters. This property gives users an option to not include task-level counters to avoid going over that limit.
False
No
Properties that control how tasks and jobs get retried on failure.
Whether retries of failed work units across job runs are enabled or not.
True
No
Work unit retry policy, can be one of {always, never, onfull, onpartial}.
always
No
Maximum number of task retries. A task will be re-tried this many times before it is considered a failure.
5
No
Interval in seconds between task retries. The interval increases linearly with each retry. For example, if the first interval is 300 seconds, then the second one is 600 seconds, etc.
300
No
Maximum number of failures before an alert email is triggered.
1
These properties control how tasks get executed for a job. Gobblin uses thread pools in order to executes the tasks for a specific job. In local mode there is a single thread pool per job that executes all the tasks for a job. In MR mode there is a thread pool for each map task (or container), and all Gobblin tasks assigned to that mapper are executed in that thread pool.
Size of the thread pool used by task executor for task execution. Each task executor will spawn this many threads to execute any Tasks that is has been allocated.
10
No
Core size of the thread pool used by task tracker for task state tracking and reporting.
10
No
Maximum size of the thread pool used by task tracker for task state tracking and reporting.
10
No
Core size of the thread pool used by the task executor for task retries.
2
No
Maximum size of the thread pool used by the task executor for task retries.
2
No
Task status reporting interval in milliseconds.
30000
No
Root directory where job and task state files are stored. The state-store is used by Gobblin to track state between different executions of a job. All state-store files will be written to this directory.
None
Yes
File system URI for file-system-based state stores.
file:///
No
Whether metrics collecting and reporting are enabled or not.
True
No
Metrics reporting interval in milliseconds.
60000
No
The directory where metric files will be written to.
None
No
A boolean indicating whether or not metrics should be reported to a file.
True
No
A boolean indicating whether or not metrics should be exposed via JMX.
False
No
Whether alert emails are enabled or not. Email alerts are only sent out when jobs fail consecutively job.max.failures number of times.
False
No
Whether job completion notification emails are enabled or not. Notification emails are sent whenever the job completes, regardless of whether it failed or not.
False
No
Host name of the email server.
None
Yes, if email notifications or alerts are enabled.
SMTP port number.
None
Yes, if email notifications or alerts are enabled.
User name of the sender email account.
None
No
User password of the sender email account.
None
No
Sender email address.
None
Yes, if email notifications or alerts are enabled.
Comma-separated list of recipient email addresses.
None
Yes, if email notifications or alerts are enabled.
These properties are common properties that are used among different Source implementations. Depending on what source class is being used, these parameters may or may not be necessary. These parameters are not tied to a specific source, and thus can be used in new source classes.
Fully qualified name of the Source class. For example, com.linkedin.gobblin.example.wikipedia
None
Yes
Name of the source entity that needs to be pulled from the source. The parameter represents a logical grouping of data that needs to be pulled from the source. Often this logical grouping comes in the form a database table, a source topic, etc. In many situations, such as when using the QueryBasedExtractor, it will be the name of the table that needs to pulled from the source.
None
Required for QueryBasedExtractors, FileBasedExtractors.
Timezone of the data being pulled in by the extractor. Examples include "PST" or "UTC".
None
Required for QueryBasedExtractors
Maximum number of partitions to split this current run across. Only used by the QueryBasedSource and FileBasedSource.
20
No
True if you want to skip the first record of each data partition. Only used by the FileBasedExtractor.
False
No
Namespace for the extract data. The namespace will be included in the default file name of the outputted data.
None
No
The URL of the proxy to connect to when connecting to the source. This parameter is only used for SFTP and REST sources.
None
No
The port of the proxy to connect to when connecting to the source. This parameter is only used for SFTP and REST sources.
None
No
The username to authenticate with the source. This is parameter is only used for SFTP and JDBC sources.
None
No
The password to use when authenticating with the source. This is parameter is only used for JDBC sources.
None
No
The name of the host to connect to.
None
Required for SftpExtractor, MySQLExtractor, and SQLServerExtractor.
URL to connect to for REST requests. This parameter is only used for the Salesforce source.
None
No
Version number of communication protocol. This parameter is only used for the Salesforce source.
None
No
The timeout set for connecting to the source in milliseconds.
500000
No
The value of the port to connect to.
None
Required for SftpExtractor, MySQLExtractor, SqlServerExtractor.
Table name in Hadoop which is different table name in source.
Source table name
No
True if this pull should treat the data as a full dump of table from the source, false otherwise
False
No
List of columns that will be used as the delta field for the data.
None
No
List of columns that will be used as the primary key for the data.
None
No
This limits the number of records read by Gobblin. In Gobblin's extractor the readRecord() method is expected to return records until there are no more to pull, in which case it runs null. This parameter limits the number of times readRecord() is executed. This parameter is useful for pulling a limited sample of the source data for testing purposes.
Unbounded
No
The following table lists the query based extractor configuration properties.
The format of the watermark that is used when extracting data from the source. Possible types are timestamp, date, hour, simple.
timestamp
Yes
Value for the watermark to start pulling data from, also the default watermark if the previous watermark cannot be found in the old task states.
None
Yes
Number of hours to pull in each partition.
1
No
Delta column with hour for hourly extracts (Ex: hour_sk)
None
No
If it is true, skips high watermark calculation in the source and it will use partition higher range as high watermark instead of getting it from source.
False
No
The query that the extractor should execute to pull data.
None
No
True if hourly extract is required.
False
No
"snapshot" for the incremental dimension pulls. "append_daily", "append_hourly" and "append_batch" for the append data append_batch for the data with sequence numbers as watermarks
None
No
The high watermark which this entire job should pull up to. If this is not specified, pull entire data from the table
None
No
max limit of the high watermark for the append data. CURRENT_DATE - X CURRENT_HOUR - X where X>=1
CURRENT_DATE for daily extract CURRENT_HOUR for hourly extract
No
True if this pull should override previous watermark with start.value and end.value. False otherwise.
False
No
Number of seconds that needs to be backup from the previous high watermark. This is to cover late data. Ex: Set to 3600 to cover 1 hour late data.
0
No
Database name
None
No
True if this pull needs to use source specific apis instead of standard protocols. Ex: Use salesforce bulk api instead of rest api
False
No
A boolean, if true then the QueryBasedExtractor will skip the source count calculation.
False
No
This parameter is currently only used in JDBCExtractor. The JDBCExtractor will process this many number of records from the JDBC ResultSet at a time. It will then take these records and return them to the rest of the Gobblin flow so that they can get processed by the rest of the Gobblin components.
1000
No
When a query is specified in the configuration file, it is possible a user accidentally adds in a column name that does not exist on the source side. By default, this parameter is set to false, which means that if a column is specified in the query and it does not exist in the source data set, Gobblin will just skip over that column. If it is set to true, Gobblin will actually take the config specified column and check to see if it exists in the source data set. If it doesn't exist then the job will fail.
False
No
A boolean specifying whether or not compression should be enabled when pulling data from the source. This parameter is only used for MySQL sources. If set to true, the MySQL will send compressed data back to the source.
False
No
The number of rows to pull through JDBC at a time. This is useful when the JDBC ResultSet is too big to fit into memory, so only "x" number of records will be fetched at a time.
1000
No
The following table lists the jdbc based extractor configuration properties.
The fully qualified path of the JDBC driver used to connect to the external source.
None
Yes
A enum specifying whether or not to convert the column names to a specific case before performing a query. Possible values are TOUPPER or TOLOWER.
NOCHANGE
No
The following table lists the file based extractor configuration properties.
The data directory from which to pull data from.
None
Yes
A list of files to pull - this should be set in the Source class and the extractor will pull the specified files.
None
Yes
The FileBasedExtractor will report it's status every time it processes the number of records specified by this parameter. The way it reports status is by logging out how many records it has seen.
10000
No
The URI of the filesystem to connect to.
None
Required for HadoopExtractor.
A boolean, if true then the original file names will be preserved when they are are written to the source.
False
No
The schema of the data that will be pulled by the source.
None
Yes
File location of the private key used for key based authentication. This parameter is only used for the SFTP source.
None
Yes
File location of the known hosts file used for key based authentication.
None
Yes
Properties for Gobblin converters.
Comma-separated list of fully qualified names of the Converter classes. The order is important as the converters will be applied in this order.
None
No
This converter takes in text data separated by a delimiter (converter.csv.to.json.delimiter), and splits the data into a JSON format recognized by JsonIntermediateToAvroConverter.
The regex delimiter between CSV based files, only necessary when using the CsvToJsonConverter - e.g. ",", "/t" or some other regex
None
Yes
This converter takes in JSON data in a specific schema, and converts it to Avro data.
Source format of the date columns for Avro-related converters.
None
No
Source format of the timestamp columns for Avro-related converters.
None
No
Source format of the time columns for Avro-related converters.
None
No
Source format of the time columns for Avro-related converters.
UTF-8
No
A boolean specifying whether or not a epoch time field in the JSON object is in seconds or not.
None
Yes
This converter is will fail for this many number of records before throwing an exception.
0
No
This converter takes in an Avro record, and filters out records by performing an equality operation on the value of the field specified by converter.filter.field and the value specified in converter.filter.value. It returns the record unmodified if the equality operation evaluates to true, false otherwise.
The name of the field in the Avro record, for which the converter will filter records on.
None
Yes
The value that will be used in the equality operation to filter out records.
None
Yes
This converter takes a specific field from an Avro record and returns its value.
The field in the Avro record to retrieve. If it is a nested field, then each level must be separated by a period.
None
Yes
Properties for Gobblin's fork operator.
Fully qualified name of the ForkOperator class.
com.linkedin.uif.fork.IdentityForkOperator
No
Number of fork branches.
1
No
Name of a fork branch with the given index, e.g., 0 and 1.
fork_${branch index}, e.g., fork_0 and fork_1.
No
Comma-separted list of fully qualified names of the TaskLevelPolicy classes that will run at the end of each Task.
None
No
OPTIONAL implies the corresponding class in qualitychecker.task.policies is optional and if it fails the Task will still succeed, FAIL implies that if the corresponding class fails then the Task will fail too.
OPTIONAL
No
Comma-separted list of fully qualified names of the RowLevelPolicy classes that will run on each record.
None
No
OPTIONAL implies the corresponding class in qualitychecker.row.policies is optional and if it fails the Task will still succeed, FAIL implies that if the corresponding class fails then the Task will fail too, ERR_FILE implies that if the record does not pass the test then the record will be written to an error file.
OPTIONAL
No
The quality checker will write the current record to the location specified by this parameter, if the current record fails to pass the quality checkers specified by qualitychecker.row.policies; this file will only be written to if the quality checker policy type is ERR_FILE.
None
No
Writer destination type; currently only writing to HDFS is supported.
HDFS
No
Writer output format; currently only Avro is supported.
AVRO
No
File system URI for writer output.
file:///
No
Staging directory of writer output. All staging data that the writer produces will be placed in this directory, but all the data will be eventually moved to the writer.output.dir.
None
Yes
Output directory of writer output. All output data that the writer produces will be placed in this directory, but all the data will be eventually moved to the final directory by the publisher.
None
Yes
Fully qualified name of the writer builder class.
com.linkedin.uif.writer.AvroDataWriterBuilder
No
The Path where the writer will write it's data. Data in this directory will be copied to it's final output directory by the DataPublisher.
None
Yes
The name of the file the writer writes to.
part
Yes
Writer buffer size in bytes. This parameter is only applicable for the AvroHdfsDataWriter.
4096
No
Writer deflate level. Deflate is a type of compression for Avro data.
9
No
This is used to specify the type of compression used when writing data out. Possible values are NOCOMPRESSION, DEFLATE, SNAPPY.
DEFLATE
No
The fully qualified name of the DataPublisher class to run. The DataPublisher is responsible for publishing task data once all Tasks have been completed.
None
Yes
The final output directory where the data should be published.
None
Yes
A boolean, if true and the the final output directory already exists, then the data will not be committed. If false and the final output directory already exists then it will be overwritten.
None
Yes
The final name of the file that is produced by Gobblin. By default, Gobblin already assigns a unique name to each file it produces. If that default name needs to be overridden then this parameter can be used. Typically, this parameter should be set on a per workunit basis so that file names don't collide.
No
These properties are used throughout multiple Gobblin components.
Default file system URI for all file storage; over-writable by more specific configuration properties.
file:///
No
- Home
- [Getting Started](Getting Started)
- Architecture
- User Guide
- Working with Job Configuration Files
- [Deployment](Gobblin Deployment)
- Gobblin on Yarn
- Compaction
- [State Management and Watermarks] (State-Management-and-Watermarks)
- Working with the ForkOperator
- [Configuration Glossary](Configuration Properties Glossary)
- [Partitioned Writers](Partitioned Writers)
- Monitoring
- Schedulers
- [Job Execution History Store](Job Execution History Store)
- Gobblin Build Options
- Troubleshooting
- [FAQs] (FAQs)
- Case Studies
- Gobblin Metrics
- [Quick Start](Gobblin Metrics)
- [Existing Reporters](Existing Reporters)
- [Metrics for Gobblin ETL](Metrics for Gobblin ETL)
- [Gobblin Metrics Architecture](Gobblin Metrics Architecture)
- [Implementing New Reporters](Implementing New Reporters)
- [Gobblin Metrics Performance](Gobblin Metrics Performance)
- Developer Guide
- [Customization: New Source](Customization for New Source)
- [Customization: Converter/Operator](Customization for Converter and Operator)
- Code Style Guide
- IDE setup
- Monitoring Design
- Project
- [Feature List](Feature List)
- Contributors/Team
- [Talks/Tech Blogs](Talks and Tech Blogs)
- News/Roadmap
- Posts
- Miscellaneous