Limitations and considerations for Snowpipe Streaming with high-performance architecture¶

General and service-level limitations¶

• The service is available in all AWS regions except for government-specific regions and regions in China.

• The default role is used for authorization. Ensure users are granted access to the default role when they connect to the service.

Table limits¶

• Maximum throughput: A table can achieve an aggregate throughput of 10 GBps uncompressed.

Pipe limits¶

• Channels per pipe: By default, a single pipe can have up to 2,000 active channels. Contact Snowflake Support if you require more channels for your use case.

Channel limits¶

Each channel has the following soft limits. If your application requires higher throughput per channel, contact Snowflake Support to discuss increasing these limits.

• SDK throughput: 12 MBps uncompressed

• REST endpoint throughput: 1 MBps uncompressed

• Request rate: 10 requests per second (RPS).

Ingestion and data-specific limitations¶

• The ON_ERROR option in Snowpipe Streaming with high-performance architecture only supports CONTINUE.

• Sudden spikes in data throughput might experience brief increases in end-to-end latency because the service is elastically scaling to support the new throughput level.

• Partitioned Iceberg tables aren’t supported.

SDK and architectural limitations¶

• Supported architectures (Rust Core): ARM64 Mac, Windows, ARM64-Linux, and x86_64-Linux.

• Linux requirements: If you use the SDK on Linux, your system must have glibc version 2.26 or later.

• Timezone: The SDK automatically uses UTC, and this setting can’t be changed by the user.

• OAuth and Personal Access Tokens aren’t supported.