CLIENT_RESULT_CHUNK_SIZE: Streaming result chunks honor the configured limit (Pending)¶

The CLIENT_RESULT_CHUNK_SIZE session parameter is meant to cap how much uncompressed query result data each chunk can contain. Previously, Snowflake used your value mainly to adjust an internal compression-related setting instead of lowering the maximum uncompressed chunk size. Chunks could still be large after decompression, and because clients prefetch multiple chunks, memory use could far exceed the limit you configured. That led to out-of-memory failures for some customers, especially in memory-constrained environments, when fetching very large result sets.

Pipelines and applications continue to work with this behavior change, but more chunks can be produced for the same result set, which can mean more network round trips and possible performance impact. Monitor throughput and latency after you enable the bundle.

Before the change:

Large compressed chunks were returned whose uncompressed size could still exceed what CLIENT_RESULT_CHUNK_SIZE was intended to allow. Total in-flight memory could be much higher than your setting, which sometimes caused out-of-memory errors.

After the change:

When the 2026_04 behavior change bundle is enabled in your account, Snowflake enforces a backend chunk size limit so each chunk delivered to the client has no more uncompressed data than CLIENT_RESULT_CHUNK_SIZE specifies. Prefetching stays within the bounds implied by your setting.

If you do not set CLIENT_RESULT_CHUNK_SIZE and rely on the account default (160 MB), you are not affected.

If you set CLIENT_RESULT_CHUNK_SIZE (to any supported value), you get the corrected chunking behavior. Snowflake does not require additional customer readiness steps. Monitor your pipelines and workloads after you enable the bundle; if you need higher throughput and can afford larger per-chunk memory, increase CLIENT_RESULT_CHUNK_SIZE within the allowed range (see the parameter reference).

Ref: 2319