Stream Data from Google Cloud Storage

You can create a data connector between TigerGraph’s internal Kafka server and your GCS service with a specified topic. The connector will stream data from the data source in your GCS buckets to TigerGraph’s internal Kafka cluster. You can then create and run a loading job to load data from Kafka into the graph store using the Kafka loader.

1. Prerequisites

2. Procedure

2.1. Specify connector configurations

The connector configurations provide the following information:

  • Connector class

  • Your GCP service account credentials

  • Information on how to parse the source data

  • Mapping between connector and source file

2.1.1. Specify connector class

The connector class indicates what type of connector the configuration file is used to create. Connector class is specified by the connector.class key. For connecting to GCS, its value is set to com.tigergraph.kafka.connect.filesystem.gcp.GCPSourceConnector.

2.1.2. Provide GCP account credentials

You need to provide following information for the connector to access the files in your GCS bucket. You should be able to find all items on this list in your service account key:

  • file.reader.settings.fs.gs.auth.service.account.email: The email address associated with the service account

  • file.reader.settings.fs.gs.auth.service.account.private.key.id: The private key ID associated with the service account used for GCS access.

  • file.reader.settings.fs.gs.auth.service.account.private.key: The private key associated with the service account used for GCS access.

  • file.reader.settings.client_email: The service account email.

  • file.reader.settings.fs.gs.project.id: Google Cloud Project ID with access to GCS buckets.

  • file.reader.settings.fs.gs.auth.service.account.enable: Whether to create objects for the parent directories of objects with / in their path. For example, creating gs://bucket/foo/ upon deleting or renaming gs://bucket/foo/bar.

  • file.reader.settings.fs.gs.impl: The FileSystem for gs: (GCS) uris. The value for this configuration should be com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem.

  • file.reader.settings.fs.AbstractFileSystem.gs.impl: The AbstractFileSystem for gs: (GCS) URIs. Only necessary for use with Hadoop 2. The value for this configuration should be com.google.cloud.hadoop.fs.gcs.GoogleHadoopFS.

2.1.3. Specify parsing rules

The streaming connector supports the following file types:

  • CSV files

  • JSON files. Each JSON object must be on a separate line.

  • directories

  • tar files

  • zip files

For URIs that point to directories and compressed files, the connector looks for all files inside the directory or the compressed file that match the file.regexp parameter. By default, this includes both CSV and JSON files.

If you set file.recursive to true, the connector looks for files recursively.

The following parsing options are available:

Name Description Default

file.regexp

The regular expression to filter which files to read. The default value matches all files.

.*

file.recursive

Whether to retrieve files recursively if the URI points to a directory.

true

file.reader.type

The type of file reader to use. The only supported value is text.

text

file.reader.delimited.separator

The character that separates columns. This parameter does not affect JSON files.

,

file.reader.delimited.quote

The explicit boundary markers for string tokens, either single or double quotation marks. This parameter does not affect JSON files.

The parser will not treat separator characters found within a pair of quotation marks as a separator.

Accepted values:

  • single: single quotes ' are boundary markers.

  • double: double quotes " are boundary markers.

  • Empty string "": Quotation marks are treated as ordinary characters.

Empty string ""

file.reader.delimited.value.default

The default value for a column when its value is null. This parameter does not affect JSON files.

Empty string ""

file.reader.batch.size

The maximum number of lines to include in a single batch.

1000

file.reader.text.eol

End of line character.

\n

file.reader.text.header

Whether the first line of the files is a header line. If the value is set to true, the first line of the file is ignored during data loading. If you are using JSON files, set this parameter to false.

false

file.reader.text.archive.type

File type for archive files. Setting the value of this configuration to auto will allow the connector to decide file types automatically based on the file extensions. Accepted values:

  • tar

  • zip

  • gzip

  • none

  • auto

auto

file.reader.text.archive.extensions.tar

If a file has this extension, treat it as a tar file

tar

file.reader.text.archive.extensions.zip

If a file has this extension, treat it as a zip file

zip

file.reader.text.archive.extensions.gzip

If a file has this extension, treat it as a gzip file

gz

file.reader.text.archive.extensions.tar.gz

If a file has this extension, treat it as a tar.gz file

tar.gz,tgz

2.1.4. Map source file to connector

The below configurations are required:

Name Description Default

name

Name of the connector.

None. Must be provided by the user.

topic

Name of the topic to create in Kafka.

None. Must be provided by the user.

tasks.max

The maximum number of tasks which can run in parallel.

1

file.uris

The path(s) to the data files on Google Cloud Storage. The URI may point to a CSV file, a zip file, a gzip file, or a directory

None. Must be provided by the user.

2.1.5. Example

Below is an example configuration:

connector.class=com.tigergraph.kafka.connect.filesystem.gcp.GCPSourceConnector
file.reader.settings.fs.gs.auth.service.account.email=gcsconnect@example.iam.gserviceaccount.com
file.reader.settings.fs.gs.auth.service.account.private.key.id=<example>
file.reader.settings.fs.gs.auth.service.account.private.key="<example>"
file.reader.settings.client_email="gcsconnect@example.iam.gserviceaccount.com"
file.reader.settings.fs.gs.project.id="example"
file.reader.settings.fs.gs.auth.service.account.enable=true
file.reader.settings.fs.gs.impl=com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem
file.reader.settings.fs.AbstractFileSystem.gs.impl="com.google.cloud.hadoop.fs.gcs.GoogleHadoopFS"

mode=eof
file.regexp=".*"
file.recursive=true
file.reader.type=text
file.reader.batch.size=10000
file.reader.text.eol="\\n"
file.reader.text.header=true
file.reader.text.archive.type=auto
file.reader.text.archive.extensions.tar=tar
file.reader.text.archive.extensions.zip=zip
file.reader.text.archive.extensions.gzip=gz
file.reader.text.archive.extensions.tar.gz=tar.gz,tgz

[connector_person]
name = fs-person-demo-104
tasks.max=10
topic=person-demo-104
file.uris=gs://example/p.csv

[connector_friend]
name = fs-friend-demo-104
tasks.max=10
topic=friend-demo-104
file.uris=gs://example/f.csv

2.2. Create connector

Once you have prepared the configuration file, run command gadmin connector create and specify the configuration file to create the connector:

gadmin connector create --c <config_file>

The connectors start streaming from the data source immediately after creation if the configurations are valid. You can run gadmin connector status to verify the status of the connectors. If the configurations are valid, the connectors should be in RUNNING status. Data streamed from the source stays in the destination cluster Kafka topics.

If the source data are in Avro or Parquet formats, the connector converts the data into JSON format when it loads the data into the destination Kafka cluster (often TigerGraph’s internal Kafka cluster).

If the destination cluster is TigerGraph’s internal Kafka cluster, the messages are removed as they are loaded in to the graph during the course of the loading job at regular intervals.

Automatic removal of loaded Kafka messages is an alpha feature and may be subject to change.

2.3. Create data source

Now that the connector has started loading data into TigerGraph’s internal Kafka cluster, you can create a data source and point it to the Kafka cluster:

  1. Create a data source configuration file. The only required field is the address of the broker. The broker’s IP and hostname is the private IP of your TigerGraph server and port of TigerGraph’s internal Kafka server:

    • If you don’t know your private IP, you can retrieve it by running gmyip in the bash terminal.

    • To retrieve the port of your Kafka cluster, run gadmin config get Kafka.Port to retrieve the port number. The default port is 30002.

    In the kafka.config field, you can provide additional configurations from librdkafka Global Configurations. For example, you can specify group.id to be tigergraph to specify the group that this consumer belongs to:

    {
    "broker":"10.128.0.240:30002", (1)
    "kafka_config":
        {
            "session.timeout.ms":"20000"
        }
}
    1 Make sure to use the internal ID of the machine instead of localhost for the hostname of the broker. localhost only works for single-server instances.

  2. Run CREATE DATA SOURCE to create the data source and assign the configuration file to the data source you just created:

    CREATE DATA_SOURCE KAFKA <datasource_name> FOR GRAPH <graph_name>
SET <datasource_name> = <path_to_datasource_config>

    For example:

    CREATE DATA_SOURCE KAFKA k1 FOR GRAPH Social
SET k1 = "/home/tigergraph/social/k1.config"

2.4. Create loading job

Create a loading job to load data from the data source:

  1. Create a topic-partition configuration for each topic.

    {
  "topic": <topic_name>, (1)
  "partition_list": [ (2)
    {
      "start_offset": <offset_value>, (3)
      "partition": <partition_number> (4)
    },
    {
      "start_offset": <offset_value>, (3)
      "partition": <partition_number> (4)
    }
    ...
  ]
}
    1 Replace <topic_name> with the name of the topic this configuration applies to. This must be one of the topics you configured in the connector configuration step.
    2 List of partitions you want to stream from. For each partition, you can set a start offset. If this list is empty, or partition_list isn’t included, all partitions are used with the default offset.
    3 Replace <offset_value> with the offset value you want. The default offset for loading is -1, which means you will load data from the most recent message in the topic. If you want to load from the beginning of a topic, the start_offset value should be -2.
    4 Replace <partition_number> with the partition number if you want to configure.

  2. Create a loading job and map data to graph. See Kafka loader guide for how to map data from a Kafka data source to the graph. See Loading JSON data on how to create a loading job for JSON data.

Known bug: to use the -1 value for offset, delete the start_offset key instead of setting it to -1.

For example, suppose we have the following two CSV files and schema:

  • Schema

  • p.csv

  • f.csv

CREATE VERTEX Person (PRIMARY_ID name STRING, name STRING, age INT, gender STRING, state STRING)
CREATE UNDIRECTED EDGE Friendship (FROM Person, TO Person, connect_day DATETIME)
CREATE GRAPH Social (Person, Friendship)
name,gender,age,state
Tom,male,40,ca
Dan,male,34,ny
Jenny,female,25,tx
Kevin,male,28,az
Amily,female,22,ca
Nancy,female,20,ky
Jack,male,26,fl
A,male,29,ca
person1,person2,date
Tom,Dan,2017-06-03
Tom,Jenny,2015-01-01
Dan,Jenny,2016-08-03
Jenny,Amily,2015-06-08
Dan,Nancy,2016-01-03
Nancy,Jack,2017-03-02
Dan,Kevin,2015-12-30
Amily,Dan,1990-1-1

The following topic-partition configurations and loading job will load the two CSV files into the graph:

  • topic_person.json

  • topic_friend.json

  • Loading job

{
  "topic": "person-demo-104",
  "partition_list": [
    {
      "start_offset": -2,
      "partition": 0
    }
  ]
}
{
  "topic": "friend-demo-104",
  "partition_list": [
    {
      "start_offset": -2,
      "partition": 0
    }
  ]
}
CREATE LOADING JOB load_person FOR GRAPH Social {
    DEFINE FILENAME f1 = "$k1:/home/mydata/topic_person.json"; (1)
    DEFINE FILENAME f2 = "$k1:/home/mydata/topic_friend.json";
    LOAD f1 TO VERTEX person VALUES ($0, $0, $2, $1, $3) USING separator=",";
    LOAD f2 TO EDGE friendship VALUES ($0, $1, $2)  USING separator=",";
}
1 $k1 refers to a data source defined in the graph. See Create data source

2.5. Stream data source updates

To load updates to the data source after you’ve created the connectors, you need to delete the connector and recreate another connector.

In the future, the streaming data connector will support automatically scanning for updates and stream data to TigerGraph.