Redshift Spectrum truncates Iceberg metadata_location, causing performance issues

Issue

Redshift Spectrum appears to be truncating the metadata_location parameter for Iceberg tables to 128 characters. This prevents Spectrum from properly reading Iceberg metadata files, resulting in full sequential scans instead of partition-pruned reads.

Relevant content reviewed

• Why does Redshift Spectrum do full table scans for Iceberg tables which have a nested schema?

Evidence & Impact

Identical query run with Athena vs Spectrum:

SELECT count(*)
FROM <my_table>
WHERE timestamp >= CURRENT_TIMESTAMP - interval '6' hour';

• Athena: 3.2 seconds (scans 3.2MB)
• Spectrum: 10+ minutes (full sequential scan)

All Iceberg tables in our schema show exactly 128-character parameter lengths:

SELECT tablename, 
       len(parameters) as param_length,
       parameters
FROM SVV_EXTERNAL_TABLES 
WHERE schemaname = '<my_schema>'
ORDER BY tablename;

Attempting to extract the metadata_location shows JSON parsing errors due to truncation:

SELECT tablename,
    json_extract_path_text(parameters, 'metadata_location') AS metadata_location
FROM SVV_EXTERNAL_TABLES 
WHERE schemaname = '<my_schema>';

Error:

JSON parsing error Detail: invalid json object {"metadata_location":"s3://<bucket>/.../tables/my_table/metadata/08857-a1f504fe-608c-4bab-92e6-749240a

(Note the incomplete UUID, missing .metadata.json extension & closing ").

Query plan shows Spectrum unable to utilize partitioning:

        ->  S3 HashAggregate  (cost=236199230.50..236199230.51 rows=2000 width=8)    
              ->  S3 Seq Scan my_schema.my_table location:"s3://<bucket>/.../tables/my_table" format:ICEBERG  (cost=0.00..235727774.95 rows=94291110 width=8)    
                    Filter: ((event)::text = 'Page View'::text)

Setup Details

Our Iceberg table is properly partitioned by hour:

CREATE TABLE my_table(
  timestamp timestamp,
  [other fields...]
)
PARTITIONED BY (hour(timestamp))
LOCATION '<location>'
TBLPROPERTIES ('table_type'='ICEBERG')

Iceberg metadata confirms partitioning is as expected:

{
  "format-version" : 2,
  "partition-specs" : [ {
    "spec-id" : 0,
    "fields" : [ {
      "name" : "timestamp_hour",
      "transform" : "hour",
      "source-id" : 1,
      "field-id" : 1000
    } ]
  } ]
}

Expected Behavior

Per AWS documentation:

Amazon Redshift can use the table statistics stored in Apache Iceberg metadata to optimize query plans and reduce file scans during query processing

Questions

1. Is the 128-character limit on the svv_external_tables.parameters column, or this issue more broadly, a known limitation?
2. Are there any workarounds to allow Spectrum to properly access Iceberg metadata files?
3. Is there a timeline for fixing this issue?