Introduction to Cloud Bigtable

原创

Freddy_Shen 2022-02-04 13:46:48 ©著作权

文章标签 java git sed 文章分类 代码人生

©著作权归作者所有：来自51CTO博客作者Freddy_Shen的原创作品，请联系作者获取转载授权，否则将追究法律责任

Cloud Bigtable

使用 NoSQL 宽列数据库服务存储 TB 或 PB 级数据。

Cloud Bigtable 是 Google 的全代管式 NoSQL 大数据数据库服务。它也是为 Google 搜索、Analytics（分析）、地图和 Gmail 等众多核心 Google 服务提供支撑的数据库。

1. Introduction

In this codelab, you’ll get introduced to using Cloud Bigtable with the Java HBase client.

You’ll learn how to

Avoid common mistakes with schema design
Import data in a sequence file
Query your data

When you’re done, you’ll have several maps that show New York City bus data. For example, you’ll create this heatmap of bus trips in Manhattan:

Introduction to Cloud Bigtable_git

2. About the dataset

You’ll be looking at a dataset about New York City buses. There are more than 300 bus routes and 5,800 vehicles following those routes. Our dataset is a log that includes destination name, vehicle id, latitude, longitude, expected arrival time, and scheduled arrival time. The dataset is made up of snapshots taken around every 10 minutes for June 2017.

3. Schema design

To get the best performance from Cloud Bigtable, you have to be thoughtful when you design your schema. Data in Cloud Bigtable is automatically sorted lexicographically, so if you design your schema well, querying for related data is very efficient. Cloud Bigtable allows for queries using point lookups by row key or row-range scans that return a contiguous set of rows. However, if your schema isn’t well thought out, you might find yourself piecing together multiple row lookups, or worse, doing full table scans, which are extremely slow operations.

为了从 Cloud Bigtable 获得最佳性能，您在设计模式时必须体贴周到。Cloud Bigtable中的数据是自动按词典排序的，因此，如果您设计良好的模式，查询相关数据非常有效。Cloud Bigtable允许使用行键或行范围扫描的点查找查询，这些扫描返回连续的行集。然而，如果您的模式没有经过深思熟虑，您可能会发现自己拼凑出多个行查找，或者更糟的是，进行完整的表格扫描，这是非常缓慢的操作。

It is common for tables to have billions of rows, so doing a full table scan could take several minutes and take away resources for other queries.
表有数十亿行是常见的，因此进行完整的表扫描可能需要几分钟，并带走其他查询的资源。

Plan out the queries

Our data has a variety of information, but for this codelab, you will use the location and destination of the bus.

With that information, you could perform these queries:

Get the location of a single bus over a given hour.
Get a day’s worth of data for a bus line or specific bus.
Find all the buses in a rectangle on a map.
Get the current locations of all the buses (if you were ingesting this data in real time).

This set of queries can’t all be done together optimally. For example, if you are sorting by time, you can’t do a scan based on a location without doing a full table scan. You need to prioritize based on the queries you most commonly run.

For this codelab, you’ll focus on optimizing and executing the following set of queries:

Get the locations of a specific vehicle over an hour.
Get the locations of an entire bus line over an hour.
Get the locations of all buses in Manhattan in an hour.
Get the most recent locations of all buses in Manhattan in an hour.
Get the locations of an entire bus line over the month.
Get the locations of an entire bus line with a certain destination over an hour.

Design the row key

For this codelab, you will be working with a static dataset, but you will design a schema for scalability. You’ll design a schema that allows you to stream more bus data into the table and still perform well.

Here is the proposed schema for the row key:

[Bus company/Bus line/Timestamp rounded down to the hour/Vehicle ID]. Each row has an hour of data, and each cell holds multiple time-stamped versions of the data.

For this codelab, you will use one column family to keep things simple. Here is an example view of what the data looks like. The data is sorted by row key.

Row key	cf:VehicleLocation.Latitude	cf:VehicleLocation.Longitude	…
MTA/M86-SBS/1496275200000/NYCT_5824	40.781212 @20:52:54.0040.776163 @20:43:19.0040.778714 @20:33:46.00	-73.961942 @20:52:54.00-73.946949 @20:43:19.00-73.953731 @20:33:46.00	…
MTA/M86-SBS/1496275200000/NYCT_5840	40.780664 @20:13:51.0040.788416 @20:03:40.00	-73.958357 @20:13:51.00 -73.976748 @20:03:40.00	…
MTA/M86-SBS/1496275200000/NYCT_5867	40.780281 @20:51:45.0040.779961 @20:43:15.0040.788416 @20:33:44.00	-73.946890 @20:51:45.00-73.959465 @20:43:15.00-73.976748 @20:33:44.00	…
…	…	…	…

Common mistake: You might think that making time the first value in the row key would be ideal, because you probably care about more recent data, and would want to run queries mainly around certain times. Doing this causes hotspots in the data, however, so you compromise by putting time third. This makes some of your queries more difficult, but you need to do this in order to get the full performance Cloud Bigtable has to offer. Also, you probably don’t need to get all buses for a certain time at once. Check out this talk by Twitter for information about how they optimized their schema.

4. Create instance, table, and family

Next, you’ll create a Cloud Bigtable table.

First, create a new project. Use the built-in Cloud Shell, which you can open by clicking the “Activate Cloud Shell” button in the upper-righthand corner.

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-5JeP9ToE-1617197786363)(https://codelabs.developers.google.com/codelabs/cloud-bigtable-intro-java/img/a74d156ca7862b28.png)]

Set the following environment variables to make copying and pasting the codelab commands easier:

INSTANCE_ID="bus-instance"
CLUSTER_ID="bus-cluster"
TABLE_ID="bus-data"
CLUSTER_NUM_NODES=3
CLUSTER_ZONE="us-central1-c"

Cloud Shell comes with the tools that you’ll use in this codelab, the gcloud command-line tool, the cbt command-line interface, and Maven, already installed.

Enable the Cloud Bigtable APIs by running this command.

gcloud services enable bigtable.googleapis.com bigtableadmin.googleapis.com

Create an instance by running the following command:

gcloud bigtable instances create $INSTANCE_ID \
    --cluster=$CLUSTER_ID \
    --cluster-zone=$CLUSTER_ZONE \
    --cluster-num-nodes=$CLUSTER_NUM_NODES \
    --display-name=$INSTANCE_ID

Make sure you delete the instance when you are done with the codelab to prevent recurring charges.

After you create the instance, populate the cbt configuration file and then create a table and column family by running the following commands:

echo project = $GOOGLE_CLOUD_PROJECT > ~/.cbtrc
echo instance = $INSTANCE_ID >> ~/.cbtrc

cbt createtable $TABLE_ID
cbt createfamily $TABLE_ID cf

5. Import data

Import a set of sequence files for this codelab from gs://cloud-bigtable-public-datasets/bus-data with the following steps:

Enable the Cloud Dataflow API by running this command.

gcloud services enable dataflow.googleapis.com

Run the following commands to import the table.

NUM_WORKERS=$(expr 3 \* $CLUSTER_NUM_NODES)
gcloud beta dataflow jobs run import-bus-data-$(date +%s) \
--gcs-location gs://dataflow-templates/latest/GCS_SequenceFile_to_Cloud_Bigtable \
--num-workers=$NUM_WORKERS --max-workers=$NUM_WORKERS \
--parameters bigtableProject=$GOOGLE_CLOUD_PROJECT,bigtableInstanceId=$INSTANCE_ID,bigtableTableId=$TABLE_ID,sourcePattern=gs://cloud-bigtable-public-datasets/bus-data/*

Monitor the import

You can monitor the job in the Cloud Dataflow UI. Also, you can view the load on your Cloud Bigtable instance with its monitoring UI. It should take 5 minutes for the entire import.

6. Get the code

git clone https://github.com/googlecodelabs/cbt-intro-java.git
cd cbt-intro-java

Change to Java 11 by running the following commands:

sudo update-java-alternatives -s java-1.11.0-openjdk-amd64 && export JAVA_HOME=/usr/lib/jvm/java-11-openjdk-amd64/

7. Perform a lookup

The first query you’ll perform is a simple row lookup. You’ll get the data for a bus on the M86-SBS line on June 1, 2017 from 12:00 am to 1:00 am. A vehicle with id NYCT_5824 is on the bus line then.

With that information, and knowing the schema design (Bus company/Bus line/Timestamp rounded down to the hour/Vehicle ID,) you can deduce that the row key is:

MTA/M86-SBS/1496275200000/NYCT_5824

BusQueries.java

private static final byte[] COLUMN_FAMILY_NAME = Bytes.toBytes("cf");
private static final byte[] LAT_COLUMN_NAME = Bytes.toBytes("VehicleLocation.Latitude");
private static final byte[] LONG_COLUMN_NAME = Bytes.toBytes("VehicleLocation.Longitude");

String rowKey = "MTA/M86-SBS/1496275200000/NYCT_5824";
Result getResult =
    table.get(
        new Get(Bytes.toBytes(rowKey))
            .addColumn(COLUMN_FAMILY_NAME, LAT_COLUMN_NAME)
            .addColumn(COLUMN_FAMILY_NAME, LONG_COLUMN_NAME));

The result should contain the most recent location of the bus within that hour. But you want to see all the locations, so set the maximum number of versions on the get request.

BusQueries.java

Result getResult =
    table.get(
        new Get(Bytes.toBytes(rowKey))
            .setMaxVersions(Integer.MAX_VALUE)
            .addColumn(COLUMN_FAMILY_NAME, LAT_COLUMN_NAME)
            .addColumn(COLUMN_FAMILY_NAME, LONG_COLUMN_NAME));

In the Cloud Shell, run the following command to get a list of latitudes and longitudes for that bus over the hour:

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=lookupVehicleInGivenHour

You can copy and paste the latitudes and longitudes into MapMaker App to visualize the results. After a few layers, it will tell you to create a free account. You can create an account or just delete the existing layers you have. This codelab includes a visualization for each step, if you just want to follow along. Here is the result for this first query:

Introduction to Cloud Bigtable_git_02

8. Perform a scan

Now, let’s view all the data for the bus line for that hour. The scan code looks pretty similar to the get code. You give the scanner a starting position and then indicate you only want rows for the M86-SBS bus line within the hour denoted by the timestamp 1496275200000.

BusQueries.java

Scan scan;
scan = new Scan();
scan.setMaxVersions(Integer.MAX_VALUE)
    .addColumn(COLUMN_FAMILY_NAME, LAT_COLUMN_NAME)
    .addColumn(COLUMN_FAMILY_NAME, LONG_COLUMN_NAME)
    .withStartRow(Bytes.toBytes("MTA/M86-SBS/1496275200000"))
    .setRowPrefixFilter(Bytes.toBytes("MTA/M86-SBS/1496275200000"));
ResultScanner scanner = table.getScanner(scan);

Run the following command to get the results.

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=scanBusLineInGivenHour

Introduction to Cloud Bigtable_git_03

The Map Maker app can display multiple lists at once, so you can see which of the buses are the vehicle from the first query you ran.

Introduction to Cloud Bigtable_java_04

An interesting modification to this query is to view the entire month of data for the M86-SBS bus line, and this is very easy to do. Remove the timestamp from the start row and prefix filter to get the result.

BusQueries.java

scan.withStartRow(Bytes.toBytes("MTA/M86-SBS/"))
    .setRowPrefixFilter(Bytes.toBytes("MTA/M86-SBS/"));

// Optionally, reduce the results to receive one version per column
// since there are so many data points.
scan.setMaxVersions(1);

Run the following command to get the results. (There will be a long list of results.)

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=scanEntireBusLine

If you copy the results into MapMaker, you can view a heatmap of the bus route. The orange blobs indicate the stops, and the bright red blobs are the start and end of the route.

Introduction to Cloud Bigtable_git_05

9. Introduce filters

Next, you will filter on buses heading east and buses heading west, and create a separate heatmap for each.

Note: This filter will only check the latest version, so you will set the max versions to one to ensure the results match the filter. Read more about configuring your filter.

BusQueries.java

Scan scan;
ResultScanner scanner;
scan = new Scan();
SingleColumnValueFilter valueFilter =
    new SingleColumnValueFilter(
        COLUMN_FAMILY_NAME,
        Bytes.toBytes("DestinationName"),
        CompareOp.EQUAL,
        Bytes.toBytes("Select Bus Service Yorkville East End AV"));

scan.setMaxVersions(1)
    .addColumn(COLUMN_FAMILY_NAME, LAT_COLUMN_NAME)
    .addColumn(COLUMN_FAMILY_NAME, LONG_COLUMN_NAME);
scan.withStartRow(Bytes.toBytes("MTA/M86-SBS/"))
    .setRowPrefixFilter(Bytes.toBytes("MTA/M86-SBS/"));
scan.setFilter(valueFilter);
scanner = table.getScanner(scan);

Run the following command to get the results for buses going east.

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=filterBusesGoingEast

To get the buses going west, change the string in the valueFilter:

BusQueries.java

SingleColumnValueFilter valueFilter =
    new SingleColumnValueFilter(
        COLUMN_FAMILY_NAME,
        Bytes.toBytes("DestinationName"),
        CompareOp.EQUAL,
        Bytes.toBytes("Select Bus Service Westside West End AV"));

Run the following command to get the results for buses going west.

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=filterBusesGoingWest

Buses heading east

Introduction to Cloud Bigtable_git_06

Buses heading west

Introduction to Cloud Bigtable_sed_07

By comparing the two heatmaps, you can see the differences in the routes as well as notice differences in the pacing. One interpretation of the data is that on the route heading west, the buses are getting stopped more, especially when entering Central Park. And on the buses heading east, you don’t really see many choke points.

10. Perform a multi-range scan

For the final query, you’ll address the case when you care about many bus lines in an area:

BusQueries.java

private static final String[] MANHATTAN_BUS_LINES = {"M1","M2","M3",...

Scan scan;
ResultScanner scanner;
List<RowRange> ranges = new ArrayList<>();
for (String busLine : MANHATTAN_BUS_LINES) {
  ranges.add(
      new RowRange(
          Bytes.toBytes("MTA/" + busLine + "/1496275200000"), true,
          Bytes.toBytes("MTA/" + busLine + "/1496275200001"), false));
}
Filter filter = new MultiRowRangeFilter(ranges);
scan = new Scan();
scan.setFilter(filter);
scan.setMaxVersions(Integer.MAX_VALUE)
    .addColumn(COLUMN_FAMILY_NAME, LAT_COLUMN_NAME)
    .addColumn(COLUMN_FAMILY_NAME, LONG_COLUMN_NAME);
scan.withStartRow(Bytes.toBytes("MTA/M")).setRowPrefixFilter(Bytes.toBytes("MTA/M"));
scanner = table.getScanner(scan);

Run the following command to get the results.

mvn package exec:java -Dbigtable.projectID=$GOOGLE_CLOUD_PROJECT \
-Dbigtable.instanceID=$INSTANCE_ID -Dbigtable.table=$TABLE_ID \
-Dquery=scanManhattanBusesInGivenHour

Introduction to Cloud Bigtable_java_08

11. Finish up

Clean up to avoid charges

To avoid incurring charges to your Google Cloud Platform account for the resources used in this codelab you should delete your instance.

gcloud bigtable instances delete $INSTANCE_ID

What we’ve covered

Schema design
Setting up an instance, table, and family
Importing sequence files with dataflow
Querying with a lookup, a scan, a scan with a filter, and a multi-range scan

Next steps

Learn more about Cloud Bigtable in the documentation.
Try out other Google Cloud Platform features for yourself. Have a look at our tutorials.
Learn how to monitor time-series data with the OpenTSDB integration