Configurable max rows per streaming request

[FreCap]

Due to BQ streaming put limitations, the max request size is 10MB.

Hence, considering that in average 1 record takes at least 20 bytes, if we have big batches (e.g. 500000) we might need to run against BigQuery multiple requests that would return a Request Too Large before finding the right size.

This config allows starting from a lower value altogether and reduce the amount of failed requests. Only works with simple TableWriter (no GCS)

Otherwise this can lead to

BigQueryException
Request payload size exceeds the limit: 10485760 bytes.

BigQueryException
Unexpected end of file from server

BigQueryException
Remote host terminated the handshake

BigQueryException
Error writing request body to server

[b-goyal]

Thanks for the PR, @FreCap and apologies for the delay in reviewing. I have left a few comments, please take a look when you get chance.

[b-goyal]

Can we rename this to - maxRowsPerRequest

[b-goyal]

Let's have the default behaviour same. We can use '-1' to say this is disabled and have that as the default

[b-goyal]

The maximum number of rows to be sent in one batch in the request payload to bigquery.
This can reduce number of failed calls due to Request Too Large if the payload exceeds BigQuery specified quota limits. (https://cloud.google.com/bigquery/quotas#write-api-limits)
Setting it to a low value can result in degraded performance of the connector

[b-goyal]

Lets add a validator as well with minimum and maximum values allowed.
-1 -> default
1 -> min
50,000 -> max (https://cloud.google.com/bigquery/quotas#write-api-limits)

[FreCap]

Hi! Unfortunately I moved on another project, but feel free to pick up the MR. Thank you, Francesco

…

On Mon, Aug 28, 2023 at 08:45 Bhagyashree ***@***.***> wrote: ***@***.**** requested changes on this pull request. Thanks for the PR, @FreCap <https://github.com/FreCap> and apologies for the delay in reviewing. I have left a few comments, please take a look when you get chance. ------------------------------ In kcbq-connector/src/main/java/com/wepay/kafka/connect/bigquery/config/BigQuerySinkConfig.java <#237 (comment)> : > @@ -93,6 +93,18 @@ public class BigQuerySinkConfig extends AbstractConfig { "The interval, in seconds, in which to attempt to run GCS to BQ load jobs. Only relevant " + "if enableBatchLoad is configured."; + public static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_CONFIG = "bqStreamingMaxRowsPerRequest"; Can we rename this to - maxRowsPerRequest ------------------------------ In kcbq-connector/src/main/java/com/wepay/kafka/connect/bigquery/config/BigQuerySinkConfig.java <#237 (comment)> : > @@ -93,6 +93,18 @@ public class BigQuerySinkConfig extends AbstractConfig { "The interval, in seconds, in which to attempt to run GCS to BQ load jobs. Only relevant " + "if enableBatchLoad is configured."; + public static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_CONFIG = "bqStreamingMaxRowsPerRequest"; + private static final ConfigDef.Type BQ_STREAMING_MAX_ROWS_PER_REQUEST_TYPE = ConfigDef.Type.INT; + private static final Integer BQ_STREAMING_MAX_ROWS_PER_REQUEST_DEFAULT = 50000; Let's have the default behaviour same. We can use '-1' to say this is disabled and have that as the default ------------------------------ In kcbq-connector/src/main/java/com/wepay/kafka/connect/bigquery/config/BigQuerySinkConfig.java <#237 (comment)> : > @@ -93,6 +93,18 @@ public class BigQuerySinkConfig extends AbstractConfig { "The interval, in seconds, in which to attempt to run GCS to BQ load jobs. Only relevant " + "if enableBatchLoad is configured."; + public static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_CONFIG = "bqStreamingMaxRowsPerRequest"; + private static final ConfigDef.Type BQ_STREAMING_MAX_ROWS_PER_REQUEST_TYPE = ConfigDef.Type.INT; + private static final Integer BQ_STREAMING_MAX_ROWS_PER_REQUEST_DEFAULT = 50000; + private static final ConfigDef.Importance BQ_STREAMING_MAX_ROWS_PER_REQUEST_IMPORTANCE = ConfigDef.Importance.LOW; + private static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_DOC = The maximum number of rows to be sent in one batch in the request payload to bigquery. This can reduce number of failed calls due to Request Too Large if the payload exceeds BigQuery specified quota limits. ( https://cloud.google.com/bigquery/quotas#write-api-limits) Setting it to a low value can result in degraded performance of the connector ------------------------------ In kcbq-connector/src/main/java/com/wepay/kafka/connect/bigquery/config/BigQuerySinkConfig.java <#237 (comment)> : > @@ -93,6 +93,18 @@ public class BigQuerySinkConfig extends AbstractConfig { "The interval, in seconds, in which to attempt to run GCS to BQ load jobs. Only relevant " + "if enableBatchLoad is configured."; + public static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_CONFIG = "bqStreamingMaxRowsPerRequest"; + private static final ConfigDef.Type BQ_STREAMING_MAX_ROWS_PER_REQUEST_TYPE = ConfigDef.Type.INT; + private static final Integer BQ_STREAMING_MAX_ROWS_PER_REQUEST_DEFAULT = 50000; + private static final ConfigDef.Importance BQ_STREAMING_MAX_ROWS_PER_REQUEST_IMPORTANCE = ConfigDef.Importance.LOW; + private static final String BQ_STREAMING_MAX_ROWS_PER_REQUEST_DOC = + "Due to BQ streaming put limitations, the max request size is 10MB. " + + "Hence, considering that in average 1 record takes at least 20 bytes, " + + "if we have big batches (e.g. 500000) we might need to run against BigQuery multiple requests " + + "that would return a `Request Too Large` before finding the right size. " + + "This config allows starting from a lower value altogether and reduce the amount of failed requests. " + + "Only works with simple TableWriter (no GCS)"; + Lets add a validator as well with minimum and maximum values allowed. -1 -> default 1 -> min 50,000 -> max (https://cloud.google.com/bigquery/quotas#write-api-limits) — Reply to this email directly, view it on GitHub <#237 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAM4TGJ3Y2A2JRXQEVPMLRTXXSHF5ANCNFSM6AAAAAAQM4XRKA> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>