Streaming Data into BigQuery

Instead of using a job to load data into BigQuery, you can choose to stream your data into BigQuery one record at a time by using the tabledata().insertAll() method. This approach enables querying data without the delay of running a load job. This document discusses several important trade-offs to consider before choosing an approach, including streaming quotas, data availability, and data consistency.

Before you begin

  1. Ensure that you have write access to the dataset that contains your destination table. The table must exist before you begin writing data to it unless you are using template tables. For more information on template tables, see Creating tables automatically using template tables.

  2. Check the quota policy for streaming data.

Checking for data availability

Streamed data is available for real-time analysis within a few seconds of the first streaming insertion into a table.

Data can take up to 90 minutes to become available for copy and export operations. Also, when streaming to a partitioned table, data in the streaming buffer has a NULL value for the _PARTITIONTIME pseudo column. To see whether data is available for copy and export, check the tables.get response for a section named streamingBuffer. If that section is absent, your data should be available for copy or export, and should have a non-null value for the _PARTITIONTIME pseudo column.

Ensuring data consistency

To help ensure data consistency, you can supply insertId for each inserted row. BigQuery remembers this ID for at least one minute. If you try to stream the same set of rows within that time period and the insertId property is set, BigQuery uses the insertId property to de-duplicate your data on a best effort basis.

You might have to retry an insert because there's no way to determine the state of a streaming insert under certain error conditions, such as network errors between your system and BigQuery or internal errors within BigQuery. If you retry an insert, use the same insertId for the same set of rows so that BigQuery can attempt to de-duplicate your data. For more information, see troubleshooting streaming inserts.

In the rare instance of a Google datacenter losing connectivity unexpectedly, automatic deduplication may not be possible.

If you have stronger requirements for your data, Google Cloud Datastore is an alternative service that supports transactions.

Streaming data across data locations

You can stream data to datasets in both the US and EU. Data can flow through machines outside the dataset's location while BigQuery processes the insertAll request. If you are streaming data from a location outside of the dataset's location, you might experience increased latency and error rates.

Streaming into partitioned tables

When streaming into partitioned tables, you can stream directly to the table. In this case, rows will have a NULL value for the [_PARTITIONTIME] pseudo column. Rows are assigned a value for [_PARTITIONTIME] based on the UTC timestamp at which they leave the streaming buffer. You can also stream directly to a partition of a partitioned table. For example, you can stream to the partition corresponding to 2017-03-01 for table mydataset.table using the partition decorator:


When streaming using a partition decorator, you can stream to partitions within the last 30 days in the past and 5 days in the future relative to the current date, based on current UTC time. To write to partitions for dates outside these allowed bounds, you can use load or query jobs, as described in Restating data in a partition.

Creating tables automatically using template tables

A common usage pattern for streaming data into BigQuery is to split a logical table into many smaller tables, either for creating smaller sets of data (e.g., by date or by user ID) or for scalability (e.g., streaming more than the current limit of 100,000 rows per second). To split a table into many smaller tables without adding complex client-side code, use the BigQuery template tables feature to let BigQuery create the tables for you.

To use a template table via the BigQuery API, add a templateSuffix parameter to your insertAll request. For the bq command line tool, add the template_suffix flag to your insert command. If BigQuery detects a templateSuffix parameter or the template_suffix flag, it treats the targeted table as a base template, and creates a new table that shares the same schema as the targeted table and has a name that includes the specified suffix:

<targeted_table_name> + <templateSuffix>

By using a template table, you avoid the overhead of creating each table individually and specifying the schema for each table. You need only create a single template, and supply different suffixes so that BigQuery can create the new tables for you. BigQuery places the tables in the same project and dataset. Templates also make it easier to update the schema because you need only update the template table.

Tables created via template tables are usually available within a few seconds. On rare occasions they may take longer to become available.

Changing the template table schema

If you change a template table schema, all subsequently generated tables will use the updated schema. Previously generated tables will not be affected, unless the existing table still has a streaming buffer.

For existing tables that still have a streaming buffer, if you modify the template table schema in a backward compatible way, the schema of those actively streamed generated tables will also be updated. However, if you modify the template table schema in a non-backward compatible way, any buffered data that uses the old schema will be lost. Additionally, you will not be able to stream new data to existing generated tables that use the old, but now incompatible, schema.

After you change a template table schema, wait until the changes have propagated before you try to insert new data or query generated tables. Requests to insert new fields should succeed within a few minutes. Attempts to query the new fields might require a longer wait of up to 90 minutes.

If you want to change a generated table's schema, do not change the schema until streaming via the template table has ceased and the generated table's streaming statistics section is absent from the tables.get() response, which indicates that no data is buffered on the table.

Template table details

Template suffix value
The templateSuffix (or --template_suffix) value must contain only letters (a-z, A-Z), numbers (0-9), or underscores (_). The maximum combined length of the table name and the table suffix is 1024 characters.
The same quotas apply to all tables, whether they are based on templates or created manually.
Time to live
The generated table inherits its expiration time from the dataset. As with normal streaming data, generated tables cannot be copied or exported immediately.
Deduplication only happens between uniform references to a destination table. For example, if you simultaneously stream to a generated table using both template tables and a regular insertAll command, no deduplication occurs between rows inserted by template tables and a regular insertAll command.
The template table and the generated tables should not be views.

Example use cases

High volume event logging

If you have an app that collects a large amount of data in real-time, streaming inserts can be a good choice. Generally, these types of apps have the following criteria:

  • Not transactional. High volume, continuously appended rows. The app can tolerate a rare possibility that duplication might occur or that data might be temporarily unavailable.
  • Aggregate analysis. Queries generally are performed for trend analysis, as opposed to single or narrow record selection.

One example of high volume event logging is event tracking. Suppose you have a mobile app that tracks events. Your app, or mobile servers, could independently record user interactions or system errors and stream them into BigQuery. You could analyze this data to determine overall trends, such as areas of high interaction or problems, and monitor error conditions in real-time.

Real-time dashboards and queries

In certain situations, streaming data into BigQuery enables real-time analysis over transactional data. Since streaming data comes with a possibility of duplicated data, ensure that you have a primary, transactional data store outside of BigQuery.

You can take a few precautions to ensure that you'll be able to perform analysis over transactional data, and also have an up-to-the-second view of your data:

  1. Create two tables with an identical schema. The first table is for the reconciled data, and the second table is for the real-time, unreconciled data.
  2. On the client side, maintain a transactional data store for records.
  3. Fire-and-forget insertAll() requests for these records. The insertAll() request should specify the real-time, unreconciled table as the destination table.
  4. At some interval, append the reconciled data from the transactional data store and truncate the unreconciled data table.
  5. For real-time dashboards and queries, you can select data from both tables. The unreconciled data table might include duplicates or dropped records.

Manually removing duplicates

You can use the following manual process to ensure that no duplicate rows exist after you are done streaming.

  1. Add the insertId as a column in your table schema and include the insertId value in the data for each row.
  2. After streaming has stopped, perform the following query to check for duplicates:

      MAX(count) FROM(
        count(*) as count
      GROUP BY

    If the result is greater than 1, duplicates exist.
  3. To remove duplicates, perform the following query. You should specify a destination table, allow large results, and disable result flattening.

      * EXCEPT(row_number)
    FROM (
              OVER (PARTITION BY [ID_COLUMN]) row_number
      row_number = 1

Notes about the duplicate removal query:

  • The safer strategy for the duplicate removal query is to target a new table. Alternatively, you can target the source table with write disposition WRITE_TRUNCATE.
  • The duplicate removal query adds a row_number column with the value 1 to the end of the table schema. The query uses a SELECT * EXCEPT statement from standard SQL to exclude the row_number column from the destination table. The #standardSQL prefix enables standard SQL for this query. Alternatively, you can can select by specific column names to omit this column.
  • For querying live data with duplicates removed, you can also create a view over your table using the duplicate removal query. Be aware that query costs against the view will be calculated based on the columns selected in your view, which can result in large bytes scanned sizes.

Troubleshooting streaming inserts

For information about how to troubleshoot errors during streaming inserts, see Troubleshooting streaming inserts on the Troubleshooting Errors page.

Back to top

Streaming insert examples


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

public void UploadJson(string datasetId, string tableId, BigQueryClient client)
    // Note that there's a single line per JSON object. This is not a JSON array.
    IEnumerable<string> jsonRows = new string[]
        "{ 'title': 'exampleJsonFromStream', 'unique_words': 1}",
        "{ 'title': 'moreExampleJsonFromStream', 'unique_words': 1}",
        //add more rows here...
    }.Select(row => row.Replace('\'', '"')); // Simple way of representing C# in JSON to avoid escaping " everywhere.

    // Normally we'd be uploading from a file or similar. Any readable stream can be used.
    var stream = new MemoryStream(Encoding.UTF8.GetBytes(string.Join("\n", jsonRows)));

    // This example uploads data to an existing table. If the upload will create a new table
    // or if the schema in the JSON isn't identical to the schema in the table,
    // create a schema to pass into the call instead of passing in a null value.
    BigQueryJob job = client.UploadJson(datasetId, tableId, null, stream);
    // Use the job to find out when the data has finished being inserted into the table,
    // report errors etc.

    // Wait for the job to complete.


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

u := client.Dataset(datasetID).Table(tableID).Uploader()
items := []*Item{
	// Item implements the ValueSaver interface.
	{Name: "n1", Count: 7},
	{Name: "n2", Count: 2},
	{Name: "n3", Count: 1},
if err := u.Put(ctx, items); err != nil {
	return err


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

TableId tableId = TableId.of(datasetName, tableName);
// Values of the row to insert
Map<String, Object> rowContent = new HashMap<>();
rowContent.put("booleanField", true);
// Bytes are passed in base64
rowContent.put("bytesField", "Cg0NDg0="); // 0xA, 0xD, 0xD, 0xE, 0xD in base64
// Records are passed as a map
Map<String, Object> recordsContent = new HashMap<>();
recordsContent.put("stringField", "Hello, World!");
rowContent.put("recordField", recordsContent);
InsertAllResponse response = bigquery.insertAll(InsertAllRequest.newBuilder(tableId)
    .addRow("rowId", rowContent)
    // More rows can be added in the same RPC by invoking .addRow() on the builder
if (response.hasErrors()) {
  // If any of the insertions failed, this lets you inspect the errors
  for (Entry<Long, List<BigQueryError>> entry : response.getInsertErrors().entrySet()) {
    // inspect row error


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

// Imports the Google Cloud client library
const BigQuery = require('@google-cloud/bigquery');

// The project ID to use, e.g. "your-project-id"
// const projectId = "your-project-id";

// The ID of the dataset of the table into which data should be inserted, e.g. "my_dataset"
// const datasetId = "my_dataset";

// The ID of the table into which data should be inserted, e.g. "my_table"
// const tableId = "my_table";

// Instantiates a client
const bigquery = BigQuery({
  projectId: projectId

// Inserts data into a table
  .then((insertErrors) => {
    rows.forEach((row) => console.log(row));

    if (insertErrors && insertErrors.length > 0) {
      console.log('Insert errors:');
      insertErrors.forEach((err) => console.error(err));
  .catch((err) => {
    console.error('ERROR:', err);


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

use Google\Cloud\BigQuery\BigQueryClient;

 * Stream a row of data into your BigQuery table
 * Example:
 * ```
 * $data = [
 *     "field1" => "value1",
 *     "field2" => "value2",
 * ];
 * stream_row($projectId, $datasetId, $tableId, $data);
 * ```.
 * @param string $projectId The Google project ID.
 * @param string $datasetId The BigQuery dataset ID.
 * @param string $tableId   The BigQuery table ID.
 * @param string $data      An associative array representing a row of data.
 * @param string $insertId  An optional unique ID to guarantee data consistency.
function stream_row($projectId, $datasetId, $tableId, $data, $insertId = null)
    // instantiate the bigquery table service
    $bigQuery = new BigQueryClient([
        'projectId' => $projectId,
    $dataset = $bigQuery->dataset($datasetId);
    $table = $dataset->table($tableId);

    $insertResponse = $table->insertRows([
        ['insertId' => $insertId, 'data' => $data],
        // additional rows can go here
    if ($insertResponse->isSuccessful()) {
        print('Data streamed into BigQuery successfully' . PHP_EOL);
    } else {
        foreach ($insertResponse->failedRows() as $row) {
            foreach ($row['errors'] as $error) {
                printf('%s: %s' . PHP_EOL, $error['reason'], $error['message']);


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

def stream_data(dataset_name, table_name, json_data):
    bigquery_client = bigquery.Client()
    dataset = bigquery_client.dataset(dataset_name)
    table = dataset.table(table_name)
    data = json.loads(json_data)

    # Reload the table to get the schema.

    rows = [data]
    errors = table.insert_data(rows)

    if not errors:
        print('Loaded 1 row into {}:{}'.format(dataset_name, table_name))


For more on installing and creating a BigQuery client, refer to BigQuery Client Libraries.

# project_id = "Your Google Cloud project ID"
# dataset_id = "ID of the dataset containing table"
# table_id   = "ID of the table to import data into"
# row_data   = [{ column1: value, column2: value }, ...]

require "google/cloud/bigquery"

bigquery = project: project_id
dataset  = bigquery.dataset dataset_id
table    = dataset.table table_id

response = table.insert row_data

if response.success?
  puts "Inserted rows successfully"
  puts "Failed to insert #{response.error_rows.count} rows"

Back to top

Send feedback about...

BigQuery Documentation