Skip to content

Latest commit

 

History

History
314 lines (261 loc) · 11.6 KB

README.md

File metadata and controls

314 lines (261 loc) · 11.6 KB

Simple Bigtable

Overview

Cloud Bigtable is a datastore supported by Google for storing huge amounts of data and maintaining very low read latency. The main drawback to using Bigtable is that Google does not currently have an official asynchronous client. Within Spotify we have been using the RPC client which is a pain to use. This library aims to fix that by making the most common interactions with Bigtable clean and easy to use while not preventing you from doing anything you could do with the RPC client.

To import with maven, add this to your pom:

<dependency>
    <groupId>com.spotify</groupId>
    <artifactId>simple-bigtable</artifactId>
    <version>LATEST_RELEASE</version>
</dependency>

Raw RPC Client vs Bigtable Client Comparison

Using The RPC Client

To give an example of using the base RPC client (which gives the BigtableSession object), this is how you would request a single cell from Bigtable.

String projectId;
String zone;
String cluster;
BigtableSession session;

String fullTableName = String.format("projects/%s/zones/%s/clusters/%s/tables/%s",
        projectId,
        zone,
        cluster,
        "table");

// Could also use a filter chain, but you can't actually set all the filters within the same RowFilter object
// without a merge or chain of some sort
final RowFilter.Builder filter = RowFilter.newBuilder().setFamilyNameRegexFilter("column-family");
filter.mergeFrom(RowFilter.newBuilder().setColumnQualifierRegexFilter(ByteString.copyFromUtf8("column-1")).build());
filter.mergeFrom(RowFilter.newBuilder().setCellsPerColumnLimitFilter(1).build()); // By default it is 1

final ReadRowsRequest readRowsRequest = ReadRowsRequest.newBuilder()
        .setTableName(fullTableName)
        .setRowKey(ByteString.copyFromUtf8("row"))
        .setNumRowsLimit(1)
        .setFilter(filter.build())
        .build();

final ListenableFuture<List<Row>> future = session.getDataClient().readRowsAsync(readRowsRequest);
final ListenableFuture<Cell> cell = FuturesExtra.syncTransform(future, rows -> {
  // This doesnt actually check if the row, column family, and qualifier exist
  // IndexOutOfBoundsException might be thrown
  return rows.get(0).getFamilies(0).getColumns(0).getCells(0);
});

Bigtable Client

The goal of this client is to let you query what you want with minimal overhead (there should be no need to create all these filter objects) as well as give you the object you want without needing to constantly convert a list of rows down to a single cell. Note that these examples use a String as a row key. Bigtable keys are really byte arrays. Strings in this api is just a convenience. Under the cover the string "row" is converted to a ByteString. In reality you should use byte arrays as keys as that will be more efficient.

Here is the same query as above using this client wrapper.

String projectId;
String zone;
String cluster;
BigtableSession session;

Bigtable bigtable = new Bigtable(session, projectId, zone, cluster);
final ListenableFuture<Optional<Cell>> cell = bigtable.read("table")
    .row("row")
    .column("family:qualifier") // specify both column family and column qualifier separated by colon
    .latestCell()
    .executeAsync();

Performing Reads

The goal of this client is to make the most tedious and common interactions with Bigtable as painless as possible. Therefore reading data is an extremely large focus. Here are some examples of reading data.

Get full column family within row

final ListenableFuture<Optional<Family>> family = bigtable.read("table")
    .row("row")
    .family("family")
    .executeAsync();

Get multiple columns within a row (Currently all need to be in the same column family but hopefully that gets fixed)

// Get the entire column
final ListenableFuture<List<Column>> family = bigtable.read("table")
    .row("row")
    .family("family")
    .columnQualifiers(Lists.newArrayList("qualifier-1", "qualifier-2"))
    .executeAsync();

// Get the latest cell in each column
final ListenableFuture<List<Column>> family = bigtable.read("table")
    .row("row")
    .family("family")
    .columnQualifiers(Lists.newArrayList("qualifier1", "qualifier2"))
    .latestCell()
    .executeAsync();

Get columns within a single family and within column qualifier range

final ListenableFuture<List<Column>> columns = bigtable.read("table")
    .row("row")
    .family("family")
    .columns()
    .startQualifierInclusive(startBytestring)
    .endQualifierExclusive(endBytestring)
    .executeAsync();

Get cells between certain timestamps within a column

final ListenableFuture<List<Cell>> cells = bigtable.read("table")
    .row("row")
    .column("family:qualifier")
    .cells()
    .startTimestampMicros(someTimestamp)
    .endTimestampMicros(someLatertimestamp)
    .executeAsync();

Get the latest cell of a certain value within a column

final ListenableFuture<Optional<Cell>> cells = bigtable.read("table")
    .row("row")
    .column("family:qualifier")
    .cells()
    .startValueInclusive(myValueByteString)
    .endValueInclusive(myValueByteString)
    .latest()
    .executeAsync();

Get the latest cell of a between 2 timestamps within a column for multiple rows

final ListenableFuture<List<Row>> cells = bigtable.read("table")
    .rows(ImmutableSet.of("row1", "row2"))
    .column("family:qualifier")
    .cells()
    .startTimestampMicros(someTimestamp)
    .endTimestampMicros(someLatertimestamp)
    .latest()
    .executeAsync();

Get the multiple column families and column qualifiers (will match all combinations)

final ListenableFuture<List<Row>> cells = bigtable.read("table")
    .row("row")
    .families(ImmutableSet.of("family1, family2"))
    .columnQualifiers(ImmutableSet.of("qualifier1", "qualifier2")
    .cells()
    .startTimestampMicros(someTimestamp)
    .endTimestampMicros(someLatertimestamp)
    .latest()
    .executeAsync();

Get all rows between different ranges or with certain specific keys (these functions add rows to the row set, instead of filtering)

final ListenableFuture<List<Row>> rows = bigtable.read("table")
    .rows()
    .addRowRangeOpen(myStartKeyOpen, myEndKeyOpen) // add an exclusive range
    .addRowRangeClosed(myStartKeyClosed, myEndKeyClosed) // add an inclusive range
    .addKeys(extraKeys) // add some keys you always want
    .executeAsync();

Note that currently there is no half open, half closed range.

Other Operations

The client supports other Bigtable operations as well, with hopefully the rest of all possible operations coming soon.

Mutations (Writes, Deletions)

Mutations are performed on the row level with many mutations possible within a single call. Mutations include writing new values as well as deleting a column, column family, or an entire row and all data help in each.

Write a new cell within a column

final ListenableFuture<Empty> mutation = bigtable.mutateRow("table", "row")
    .write("family:qualifier", ByteString.copyFromUtf8("value"))
    .executeAync()

Perform multiple writes in different columns setting an explicit timestamp on some

final ListenableFuture<Empty> mutation = bigtable.mutateRow("table", "row")
    .write("family:qualifier", ByteString.copyFromUtf8("value-1"), timestampMicros)
    .write("family", "qualifier", ByteString.copyFromUtf8("value-2"))
    .executeAync()

Delete a column and then write to the same column

final Empty mutation = bigtable.mutateRow("table", "row")
    .deleteColumn("family:qualifier")
    .write("family:qualifier", ByteString.copyFromUtf8("brand-new-value"))
    .execute()

ReadModifyWrite (Atomically Update or Append To A Column)

ReadModifyWrite is useful for either incrementing the latest cell within a column by a long or appending bytes to the value. If the column is empty, is will write a new value. Once again this operation is on the row level with multiple ReadModifyWrites possible in a single request.

Increment a couple counter columns and append a value to another

bigtable.readModifyWrite("table", "row")
    .read("request-numbers:number-1")
    .thenIncrementAmount(1L)
    .read("request-numbers:number-2")
    .thenIncrementAmount(5L)
    .read("family:values")
    .thenAppendValue(ByteString.copyFromUtf8("new-value"))
    .executeAsync();

SampleRowKeys

Sample some row keys from a table.

final List<SampleRowKeysResponse> sampleRowKeys = bigtable.sampleRowKeys("table").execute();

CheckAndMutateRow - NOT YET IMPLEMENTED

Perform a read and a set of mutations depending on whether the read returns data. This is not yet implemented but here are some ideas on how this operation might be implemented in the future.

Have the check specified like a read, then allow mutations to be added.

bigtable.checkAndMutateRow("table", "row")
    .column("family:qualifier")
    .cells()
    .endTimestampMicros(timestamp)
    .ifExists()
    .deleteColumn("family:qualifier")
    .write("family:qualifier", "had-data")
    .ifDoesNotExist()
    .write("family:qualifier", "did-not-have-data")
    .executeAsync();

Pass in Bigtable protobuf objects, kind of against the purpose of the library but keeps things simple.

bigtable.checkAndMutateRow("table", "row")
    .rowFilter(someRowFilter)
    .ifExists(someMutation)
    .ifExists(someOtherMutation)
    .ifDoesNotExist(someOtherMutation)
    .executeAsync();

Pull requests with other ideas are encouraged.

Table and Cluster Admin Operations - NOT YET IMPLEMENTED

It is unclear whether there is a need this wrapper to provide the admin operations, though it would be pretty easy to include.

How to Release

  • A local build and deploy is the easiest way to make a release at this point. For setup follow instructions in: scio instructions modified to work with maven. For credentials to push to sonatype, create an access token with your sonatype login and place the access token in your maven settings.xml as in:
    <server>
      <!-- sonatype repository -->
      <id>ossrh</id>
      <username>access-token-name</username> <!-- access token tied to an account sonatype.org -->
      <password>access-token-password</password>
    </server>
  • mvn clean javadoc:jar source:jar deploy should do the rest.

  • Release deployments (from mvn deploy) will be put into a staging area at sonatype. Read the following to release the staged deployment how to release and maven.

  • With effort we could get automatic deployments via travis. The travis build console is here: travis.org. You'll need access to travis.org (not .com) to access the builds.

Open Problems and Questions

  • One problem is that currently it is not really possible to do queries for nested lists. For example there really is no way to do a ColumnRange within a RowRange or request multiple columns within different column families. Something like BigtableColumnsWithinFamilies could be added where it keeps track of needing different column families but that is confusing. Another option is adding the filtering methods to every Read object which would also be super confusing.

Code of conduct

This project adheres to the Open Code of Conduct. By participating, you are expected to honor this code.